This page intentionally left blank
cognitive load theory Cognitive load theory (CLT) is one of the most influential theories in instructional design, a highly effective guide for the design of multimedia and other learning materials. This edited volume brings together the most prolific researchers from around the world who study various aspects of cognitive load to discuss its current theoretical as well as practical issues. The book is divided into three parts: The first part describes the theoretical foundations and assumptions of CLT, the second discusses the empirical findings about the application of CLT to the design of learning environments, and the third part concludes the book with discussions and suggestions for new directions for future research. It aims to become the standard handbook in CLT for researchers and graduate students in psychology, education, and educational technology. Jan L. Plass is Associate Professor of Educational Communication and Technology in the Steinhardt School of Culture, Education, and Human Development at New York University (NYU), where he co-directs the Games for Learning Institute. He is also the founding director of the Consortium for Research and Evaluation of Advanced Technologies in Education (CREATE). His research is at the intersection of cognitive science, learning sciences, and design, and seeks to enhance the educational effectiveness of visual environments. Dr. Plass’s current focus is on the cognitive and emotional aspects of information design and the interaction design of simulations and educational games for science education and second language acquisition. He has received funding for his research from the U.S. Department of Education’s Institute of Education Sciences, the National Science Foundation, the National Institutes of Health, and, most recently, Microsoft Research and the Motorola Foundation. Roxana Moreno is Educational Psychology Professor at the University of New Mexico. Her research interests are in applying cognitive–affective theories of learning to derive principles of instructional design for a diversity of learners. Her investigations involve undergraduate students as well as K–12 students who are culturally and linguistically diverse. Dr. Moreno’s most recent projects include an engineering education grant aimed at applying empirically based technology tools to foster problem solving and cognitive flexibility in pre-college students and the “Bridging the Gap Between Theory and Practice in Teacher Education: Guided Interactive Virtual Environments for CaseBased Learning” grant, for which she received the prestigious Presidential Early Career Award in Science and Engineering. Other awards and honors include the American Psychological Association Richard E. Snow Award, being a Fulbright Senior Specialist in the areas of education and instructional media design, and an appointment as a veteran social scientist for the Department of Education. Roland Br¨unken is Full Professor in Education and Dean for Student Affairs of the Faculty of Empirical Human Sciences at Saarland University, Germany. He is also Speaker of the special interest group Educational Psychology of the German Psychological Association (DGPs). His main research interests are concerned with using new technology for education, direct measurement of cognitive load by behavioral measures, and applying cognitive psychology to the instructional design of multimedia learning environments.
Cognitive Load Theory Edited by
Jan L. Plass New York University
Roxana Moreno University of New Mexico
Roland Br¨unken Saarland University, Germany
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo, Delhi, Dubai, Tokyo Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521860239 © Cambridge University Press 2010 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2010 ISBN-13
978-0-511-72963-8
eBook (NetLibrary)
ISBN-13
978-0-521-86023-9
Hardback
ISBN-13
978-0-521-67758-5
Paperback
Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.
contents
Contributors
page vii
Introduction
1
part one. theory 1 Cognitive Load Theory: Historical Development and Relation to Other Theories Roxana Moreno and Babette Park
9
2 Cognitive Load Theory: Recent Theoretical Advances John Sweller
29
3 Schema Acquisition and Sources of Cognitive Load Slava Kalyuga
48
4 Individual Differences and Cognitive Load Theory Jan L. Plass, Slava Kalyuga, and Detlev Leutner
65
part two. empirical evidence 5 Learning from Worked-Out Examples and Problem Solving Alexander Renkl and Robert K. Atkinson 6 Instructional Control of Cognitive Load in the Design of Complex Learning Environments Liesbeth Kester, Fred Paas, and Jeroen J. G. van Merri¨enboer 7 Techniques That Reduce Extraneous Cognitive Load and Manage Intrinsic Cognitive Load during Multimedia Learning Richard E. Mayer and Roxana Moreno v
91
109
131
vi
Contents
8 Techniques That Increase Generative Processing in Multimedia Learning: Open Questions for Cognitive Load Research Roxana Moreno and Richard E. Mayer
153
part three. discussion 9 Measuring Cognitive Load Roland Br¨unken, Tina Seufert, and Fred Paas 10 From Neo-Behaviorism to Neuroscience: Perspectives on the Origins and Future Contributions of Cognitive Load Research Richard E. Clark and Vincent P. Clark 11 Cognitive Load in Learning with Multiple Representations Holger Horz and Wolfgang Schnotz 12 Current Issues and Open Questions in Cognitive Load Research Roland Br¨unken, Jan L. Plass, and Roxana Moreno Index
181
203 229
253
273
contributors
robert k. atkinson Arizona State University
roxana moreno University of New Mexico
¨ roland brunken Saarland University
fred paas Erasmus University Rotterdam
richard e. clark University of Southern California
babette park Saarland University
vincent p. clark University of New Mexico
jan l. plass New York University
holger horz University of Applied Sciences, Northwestern Switzerland
alexander renkl University of Freiburg
slava kalyuga University of New South Wales
wolfgang schnotz University of Koblenz-Landau
liesbeth kester Open University of the Netherlands
tina seufert Ulm University
detlev leutner University of Duisburg-Essen
john sweller University of New South Wales
richard e. mayer University of California, Santa Barbara
jeroen j. g. van merri¨enboer University of Maastricht
vii
cognitive load theory
introduction
What is Cognitive Load Theory (CLT)? The objective of CLT is to predict learning outcomes by taking into consideration the capabilities and limitations of the human cognitive architecture. The theory can be applied to a broad range of learning environments because it links the design characteristics of learning materials to principles of human information processing. CLT is guided by the idea that the design of effective learning scenarios has to be based on our knowledge about how the human mind works. Starting from this premise, different processes of knowledge acquisition and understanding are described in terms of their demands on the human cognitive system, which is seen as an active, limited-capacity information processing system. Taking into account the demands on cognitive resources induced by the complexity of the information to be learned, the way in which the instruction is presented to the learner, and the learner’s prior experience and knowledge, CLT aims to predict what makes learning successful and how learning can be effectively supported by teaching and instruction. Because of its applicability for a broad range of instructional materials, including Web-based and multimedia instruction, CLT is a frequently discussed concept in educational psychology and applied learning sciences. A growing body of empirical research has become available in recent years that describes the relationships among human cognitive architecture, the design of educational materials, and successful learning. Moreover, the research conducted in past years has led to a more detailed description of the theoretical components of CLT, including processes of schema acquisition, capacity limitations, and different causes for load, namely, intrinsic load (generated by the difficulty of the materials), extraneous load (generated by the design of the instruction and materials), and germane load (the amount of invested mental effort). 1
2
Introduction
Considering the theoretical and empirical developments that have been made in this area, as well as the importance of the implications of CLT for the design of learning environments, especially for those using Webbased or multimedia formats for the delivery of instruction, there is a need to present the current knowledge about CLT in a handbook for research, education, and application. This edited volume brings together the most prolific researchers from around the world who study various aspects of cognitive load to discuss current theoretical as well as practical issues of CLT. The book is divided into three parts: The first part describes the theoretical foundation and assumptions of CLT, the second part examines the empirical findings about the application of CLT to the design of learning environments, and the third part concludes the book with a discussion and directions for future research. The chapters in the first part of this book discuss the theoretical underpinnings of CLT. In Chapter 1, Moreno and Park place CLT into the broader context of the learning sciences by providing a historical review of the assumptions underlying CLT and by relating the theory to other relevant theories in psychology and education. In Chapter 2, Sweller presents five assumptions underlying CLT using an analogy between evolution by natural selection and human cognitive architecture. Specifically, the chapter describes Sweller’s most recent information store, borrowing, randomness as genesis, narrow limits of change, and environment organizing and linking CLT assumptions (Sweller, 2004). In addition, Chapter 2 describes the three categories of cognitive load and the additive load hypothesis, according to which intrinsic, extraneous, and germane cognitive load add to produce a total cognitive load level during learning. In Chapter 3, Kalyuga describes more fully the process of schema acquisition according to CLT and presents three instructional principles in its support: the direct initial instruction principle, the expertise principle, and the small step-size of knowledge change principle. In Chapter 4, Plass, Kalyuga, and Leutner expand on the first three chapters by offering a typology of individual differences that may have an effect on learners’ working memory capacity. To this end, they distinguish between differences in information gathering, information processing, and regulation of processing and explain how such differences may affect cognitive load during learning. Taken together, the first four chapters of this book synthesize the history of CLT, describe the main principles underlying the current CLT, highlight the relation of CLT to individual learner differences, and relate CLT to other theoretical models. As Sweller argues in Chapter 2, not only is the type of load imposed by the difficulty of the material (intrinsic load) and the instructional design
Introduction
3
(germane or extraneous loads) critical to CLT, but the learner’s prior knowledge is as well. Information or instructional activities that are crucial to novices may interfere with further learning by more expert learners, giving rise to the expertise reversal effect (Kalyuga, Chapter 3, this volume). Instructional methods that promote schema acquisition in novices (leading to increased germane cognitive load) may contribute to extraneous cognitive load for more expert learners. Moreover, as Plass, Kalyuga, and Leutner argue in Chapter 4, cognitive load is most likely to arise when spatial ability is low or when students do not have good metacognitive skills. In sum, the relationship between the three types of load and learners’ characteristics is far from simple. The second part of this book synthesizes the findings of recent empirical studies conducted by the leading researchers in the cognitive load field and translates the insights gained from this work into guidelines for the design of learning environments. In Chapter 5, Renkl and Atkinson summarize research in which CLT is used to design learning environments that promote problem solving with worked-out examples. In Chapter 6, Kester, Paas, and van Merri¨enboer summarize research in which CLT is used to design learning environments that promote complex cognitive processes such as air traffic control systems. Finally, Chapters 7 and 8 summarize the research program of Mayer and Moreno, who have developed a set of empirically based principles to reduce intrinsic and extraneous cognitive load (Chapter 7) and increase generative processing (Chapter 8) in multimedia learning. CLT began as an instructional theory that, based on assumptions regarding the characteristics of the human cognitive architecture, was used to generate a series of cognitive load effects in randomised, controlled experiments. Some examples are the modality effect, according to which multiple sources of information that are unintelligible in isolation result in less learning when they are presented in single-modality as opposed to dual-modality format (Low & Sweller, 2005; Mayer & Moreno, Chapter 7, this volume); the redundancy effect, according to which the presence of information that does not contribute to schema acquisition or automation interferes with learning (Mayer & Moreno, Chapter 7, this volume; Sweller, 2005); and the worked example effect, according to which studying worked examples promotes problem solving compared with solving the equivalent problems (Renkl, 2005; Renkl & Atkinson, Chapter 5, this volume). The fact that each one of these cognitive load effects was replicated across a variety of learning environments and domains led cognitive load researchers to derive corresponding evidence-based instructional principles. CLT can, therefore, provide instructional designers with guidelines for the design
4
Introduction
of multimedia learning environments that include verbal representations of information (e.g., text, narrated words) and pictorial representations of information (e.g., animation, simulation, video, photos), as well as for the design of Web-based learning environments. The research reviewed in the second part of this volume focuses on empirical work that applied CLT to multimedia and online learning environments. According to CLT’s additivity hypothesis, learning is compromised when the sum of intrinsic, extraneous, and germane loads exceeds available working memory capacity and any cognitive load effect is caused by various interactions among these sources of cognitive load. For example, many cognitive load effects occur because a reduction in extraneous cognitive load permits an increase in germane cognitive load, which in turn enhances learning. This is presumably the underlying cause of the redundancy effect reviewed by Mayer and Moreno in Chapter 7. However, these effects only occur when intrinsic cognitive load is high. If intrinsic cognitive load is low, sufficient working memory resources are likely to be available to overcome a poor instructional design that imposes an unnecessary extraneous cognitive load. Kester, Paas, and van Merri¨enboer (Chapter 6, this volume) explore this hypothesis by examining cognitive load effects in complex instructional environments. Ideally, good instructional design should reduce extraneous cognitive load and use the liberated cognitive resources to increase germane cognitive load and learning. Renkl and Atkinson (Chapter 5, this volume) explore this hypothesis by examining cognitive load effects in worked-out example instruction (aimed at reducing extraneous cognitive load) that includes different cognitive activities to engage students in deeper learning (aimed at increasing germane cognitive load). As Moreno and Park describe in Chapter 1, the empirical findings produced by cognitive load researchers over the past twenty years motivated a series of revisions of CLT since its inception. The third part of the book includes chapters that discuss the current state of CLT as well as open questions for future developments. In Chapter 9, Br¨unken, Seufert, and Paas discuss the general problem of measuring cognitive load, summarize the different types of measures that are commonly used, and discuss current issues in cognitive load measurement, such as the problem of global versus differential measurement of the three types of cognitive load and the relationship between cognitive load and learners’ prior knowledge. A critical evaluation of CLT from the perspective of the broader field of educational psychology and cognitive psychology is provided by Clark and Clark in Chapter 10. In Chapter 11, Horz and Schnotz compare CLT with other theoretical models used in the field of instructional
Introduction
5
design, namely Mayer’s (2005) cognitive theory of multimedia learning and Schnotz’s (2005) integrated model of text and picture comprehension. The last chapter of this volume presents some current open questions in cognitive load research (Br¨unken, Plass, & Moreno, Chapter 12). This edited volume could not have been completed without the help of numerous collaborators. We thank the contributing authors for their patience in completing this book, which changed its form more than once based on the chapters we received. We would also like to thank Simina Calin, our editor at Cambridge University Press, who has patiently guided us through the process of completing this volume, as well as her assistant Jeanie Lee. Our work has been generously supported by a number of funding agencies, including the National Science Foundation,1 the Institute of Education Sciences,2 the National Institutes of Health,3 and the German Research Foundation (DFG),4 as well as by Microsoft Research5 and the Motorola Foundation.6 Any opinions, findings, conclusions, or recommendations expressed in this book are those of the authors and do not necessarily reflect the views of the funding agencies. We especially thank our partners and loved ones, who made this work possible through their enduring emotional support. Jan L. Plass, Roxana Moreno, and Roland Br¨unken New York, Albuquerque, and Saarbr¨ucken August 2009 references Low, R., & Sweller, J. (2005). The modality principle in multimedia learning. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 147–158). New York: Cambridge University Press. Mayer, R. E. (2005). Cognitive theory of multimedia learning. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 31–48). New York: Cambridge University Press. 1
Support from the National Science Foundation: Grant Nos. 0238385 and 0648568 to Roxana Moreno, and Grant No. HSD-0332898 to Jan L. Plass. 2 Support from the Institute of Education Sciences, U.S. Department of Education: Grant Nos. R305K050140, R305A090203 and R305B080007 to Jan L. Plass. 3 Support from the National Institutes of Health: Grant No. 1R01LM009538–01A1 from the National Library of Medicine to Jan L. Plass. 4 Support from the Deutsche Forschungsgemeinschaft (German Research Foundation): Grant No. BR2082/6–1 to Roland Br¨unken. 5 Grant to Jan L. Plass and Ken Perlin. 6 Innovation Generation Grant to Jan L. Plass.
6
Introduction
Renkl, A. (2005). The worked-out examples principle in multimedia learning. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 229–245). New York: Cambridge University Press. Schnotz, W. (2005). An integrated model of text and picture comprehension. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 49–69). New York: Cambridge University Press. Sweller, J. (2004). Instructional design consequences of an analogy between evolution by natural selection and human cognitive architecture. Instructional Science, 32, 9–31. Sweller, J. (2005). The redundancy principle in multimedia learning. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 159–167). New York: Cambridge University Press.
part one THEORY
1 Cognitive Load Theory: Historical Development and Relation to Other Theories roxana moreno and babette park
The goal of this introductory chapter is to provide a historical review of the assumptions underlying Cognitive Load Theory (CLT) and to place the theory into the broader context of the learning sciences. The chapter focuses on the theoretical developments that guided the research on cognitive load and learning for the past twenty years and is organized in the following way. First, we examine the nature of the cognitive load construct and compare it to similar psychological constructs. Second, we present a historical review of the development of CLT’s assumptions in the following four stages: (a) extraneous cognitive load in problem solving, (b) intrinsic cognitive load and the first additivity hypothesis, (c) germane cognitive load and the second additivity hypothesis, and (d) the evolutionary interpretation of CLT. Finally, we conclude the chapter by examining the constructs and assumptions of CLT in relation to other theories in psychology and education.
the cognitive load construct CLT is a psychological theory because it attempts to explain psychological or behavioral phenomena resulting from instruction. Psychological theories are concerned with the possible relationships among psychological constructs or between a psychological construct and an observable phenomenon of practical consequence. A psychological construct is an attribute or skill that happens in the human brain. In CLT, the main constructs of interest are cognitive load, hence the name of the theory, and learning. CLT was developed to explain the effects of instructional design on these two constructs. The idea of cognitive load, however, was not new at the time the theory was developed. A similar psychological construct called “mental load” was 9
10
Roxana Moreno and Babette Park
already defined in the human factors psychology domain by Moray (1979) as the difference between task demands and the person’s ability to master these demands. The mental load construct is essential to the human factors science, which is concerned with understanding how human-specific physical, cognitive, and social properties may interact with technological systems, the human natural environment, and human organizations. The relation of mental load or workload and performance has been investigated in many fields, including cognitive ergonomics, usability, human computer/human machine interaction, and user experience engineering (Hancock & Desmond, 2001; Hancock & Meshkati, 1988; Huey & Wickens, 1993; Wickens & Hollands, 2000). Likewise, the construct of task difficulty was used to refer to the mental load experienced during performance with the practical goal of developing measures of job difficulty for several professional specialties (Madden, 1962; Mead, 1970). The influence of this work in the development of CLT is clear. For instance, the development of the first subjective cognitive load scale (Paas & van Merri¨enboer, 1994) was inspired by a previously developed scale to assess perceived item difficulty in cognitive tests (Bratfisch, Borg, & Dornic, 1972). After conducting a careful review of the human factors literature, MacDonald (2003) concluded that mental workload is more than just the amount of work that has to be done to accomplish a task. Other psychological factors, such as demand expectations, the actual effort expended during performance, and the perceived adequacy of performance, need to be taken into consideration when predicting mental load. For example, even if the amount of work that needs to be done to accomplish a task is high, different workload levels will result from individual differences in the willingness to spend effort on such a task. This willingness will depend on, among other factors, the learner’s self-schemas and how relevant the task is perceived to be in terms of helping the learner achieve meaningful, personal goals (Eccles, Wigfield, & Schiefele, 1998; Thrash & Elliott, 2001). The cognitive load construct is similar to the workload construct in that it takes into consideration the demands that a certain task imposes on an individual. However, it does not take into consideration the psychological effects that individuals’ beliefs, expectations, and goals have on their load perceptions. This has been argued to be one of the limitations of CLT (Bannert, 2002; Moreno, 2006). Early psychological theories have recognized the multidimensional nature of the mental load construct by defining it as the psychological experience that results from the interaction of subjective individual characteristics and objective task characteristics (Campbell, 1988; Wood, 1986). In the words of Kantowitz (1987), mental load is “a
Cognitive Load Theory
11
subjective experience caused by . . . motivation, ability, expectations, training, timing, stress, fatigue, and circumstances in addition to the number, type and difficulty of tasks performed, effort expended, and success in meeting requirements” (p. 97). CLT has mostly focused on how the objective characteristics of the task affect cognitive load and, in turn, learning. The only individual characteristic that is explicitly included in its theoretical framework is students’ prior knowledge (Kalyuga, Chandler, & Sweller, 1998). Other individual characteristics that are highly predictive of learning, such as cognitive abilities and styles, self-regulation, motivation, and affect, are not considered within the CLT framework (Moreno, 2005). Nevertheless, several studies have examined additional individual differences that are relevant to cognitive load and learning (see Chapter 4, this volume).
the stages of clt development Stage I: Extraneous Cognitive Load in Problem Solving Traditional CLT focused on the relation between the type of cognitive processes elicited by different problem-solving methods and schema acquisition. Although not fully developed as a theory, the first articles using the term cognitive load date to the late 1980s (Sweller, 1988, 1989). In this work, the founder of CLT, John Sweller, focused on the cognitive demands of the means–ends analysis method used in conventional problem-solving practice, a method in which learners independently solve a large number of problems to develop expertise. Using a production system approach, Sweller argued that means–ends analysis imposes a higher cognitive load on students’ limited cognitive processing capacity than using a non-specific goal strategy to solve problems. The theoretical conclusion was that the cognitive effort spent in means–ends analysis leads to problem solution (the goal of the immediate task) but does not leave sufficient cognitive resources for schema acquisition (the goal of instruction). Therefore, the first hypothesis raised by CLT established a relationship between the instructional methods used to promote problem solving and the cognitive load induced by such methods. More specifically, “cognitive processing load is an important factor reducing learning during means-ends analysis” (Sweller, 1988, p. 263). Later empirical studies cite the 1988 and 1989 articles as the main reference to CLT and further elaborate on its initial ideas. For instance, Sweller, Chandler, Tierney, and Cooper (1990) state that CLT “is concerned with how cognitive resources are distributed during learning and problem solving.
12
Roxana Moreno and Babette Park
Many learning and problem-solving activities impose a heavy, extraneous cognitive load that interferes with the primary goal of the task at hand” (p. 176). This article and several others show that the cognitive load construct in this first stage of the theory was mainly concerned with the unnecessary cognitive demands imposed by instructional design (van Merri¨enboer & Ayres, 2005). Because this source of load can be eliminated by appropriately redesigning the instructional materials, it was called extraneous cognitive load. In addition to stating the goal of CLT, this stage is characterized by having set the following main assumptions (Cooper & Sweller, 1987; Sweller & Cooper, 1985; Sweller et al., 1990): (a) Schema acquisition is the building block of skilled performance. (b) Schema acquisition requires attention directed to problem states and their associated solution moves. (c) Learning is enhanced when learners attend to schema acquisition. (d) Other cognitive activities must remain limited to avoid imposing a heavy cognitive load that interferes with learning. As seen from the previous list, the early stage of CLT was concerned with the question of how to design instruction to promote problem solving in well-defined domains and was inspired by the research on expert problemsolving and schema theory (Bartlett, 1932; Chase & Simon, 1973; Chi, Glaser, & Rees, 1982; De Groot, 1966; Simon & Simon, 1978). According to schema theory, people represent knowledge as networks of connected facts and concepts that provide a structure for making sense of new information (Anderson & Bower, 1983; Rumelhart & Ortony, 1976). Experts in a domain have well-structured schemata that are automatically activated during problem solving, which allows them to categorize problems according to their structural characteristics (Egan & Schwartz, 1979; Simon & Gilmartin, 1973). This is why CLT states that instruction should avoid using methods that are unrelated to the development of problem-solving schemas such as means– ends analysis. Although studying worked-out problems is a less demanding method to develop problem-solving skills than the means–ends analysis method (see Chapter 5, this volume), it is also susceptible to inducing extraneous cognitive load when the worked examples are poorly designed. Therefore, during the initial stage of CLT development, researchers also began to examine the effects that manipulations of the design of worked examples had on students’ learning, such as the five cognitive-load effects listed in Table 1.1 (Sweller, van Merri¨enboer, & Paas, 1998).
13
Cognitive Load Theory
table 1.1. Traditional cognitive-load effects focusing on the reduction of extraneous cognitive load Effect and references
Description
Goal-free effect (Owen & Sweller, 1985; Sweller, Mawer, & Ward, 1983; Tarmizi & Sweller, 1988)
Goal-free problems reduce extraneous cognitive load compared with means–ends analysis by focusing students’ attention on problem states and available operators. Replacing means–ends analysis with the study of worked examples reduces extraneous cognitive load by focusing students’ attention on problem states and solution steps. Replacing multiple sources of mutually referring information with a single, integrated source of information reduces extraneous cognitive load by avoiding the need to mentally integrate the information sources. Completing partially completed problems rather than solving entire problems reduces extraneous cognitive load by reducing the size of the problem space, which helps focus attention on problem states and solution steps. Replacing multiple sources of information that can be understood in isolation with one source of information reduces extraneous cognitive load by eliminating the processing of redundant information.
Worked-example effect (Cooper & Sweller, 1987; Sweller & Cooper, 1985)
Split-attention effect (Chandler & Sweller, 1991, 1992; Sweller & Chandler, 1994; Sweller et al., 1990)
Completion effect (Paas, 1992; van Merri¨enboer & De Croock, 1992)
Redundancy effect (Chandler & Sweller, 1991; Sweller & Chandler, 1994)
For instance, presenting non-integrated mutually referring pieces of information (e.g., graphics, symbols) in worked-out problems was predicted to produce extraneous cognitive load by forcing the learner to mentally integrate the information, a process that is unrelated to the development of problem-solving schemas. Therefore, CLT predicts that presenting integrated rather than non-integrated problem information sources should promote learning by eliminating extraneous load (Chandler & Sweller, 1991). Several studies have shown, indeed, that integrated mutually referring sources of information promote better learning (Chandler & Sweller, 1991, 1996; Kalyuga, Chandler, & Sweller, 1999; Tarmizi & Sweller, 1988).
14
Roxana Moreno and Babette Park
An interesting development that started in the early 1990s was the use of a seven-point self-reported rating of participants’ perceived difficulty to test the theory’s assumptions (Paas & van Merri¨enboer, 1993). It is important to note that the assumptions articulated at this early stage were heavily influenced by the theoretical and empirical advances of cognitive psychology, which eventually launched the development of cognitive approaches to learning (Ausubel, 1960; Bruner, Goodnow, & Austin, 1956; Chomsky, 1957). For instance, the theory assumes that the human cognitive architecture is characterized by a very limited short-term memory (Kahneman, 1973; Miller, 1956) and a very large long-term memory, which are two of the basic assumptions in early information-processing models (Atkinson & Shiffrin, 1968). In addition, the theory assumes that the function of learning is to store automated schemas in long-term memory so that working memory load can be reduced during problem solving (Sweller, 1994). This assumption is based on the distinction between automatic and controlled processing proposed by Schneider and Shiffrin (1977) and Shiffrin and Schneider (1977) more than a quarter-century ago. Considerable research had identified two qualitatively distinct ways to process information. Automatic processing is fast, effortless, not limited by working memory capacity, and developed with extensive practice; in contrast, controlled processing is relatively slow, mentally demanding, and highly dependent on working memory capacity (Fisk & Schneider, 1983; Logan, 1979, 1980; Posner & Snyder, 1975). In sum, schema development reduces the constraints of a limited capacity working memory in two ways. First, a highly complex schema can be manipulated as one element rather than as multiple interacting elements when brought into working memory. Second, well-developed schemata are processed automatically, minimizing the demands of cognitive resources to tackle the task at hand. Consequently, the thrust of CLT during its first development stage was to assist instructional designers to structure information appropriately so that extraneous cognitive load was reduced and novice learners could spend their limited cognitive resources in schema development. Two issues related to these assumptions, however, were still underspecified during the first stage of CLT. First, the theory did not explain which cognitive processes lead to schema acquisition. Although the literature on the role of extensive practice was cited as the basis for developing automated schemas in problem solving (Kotovsky, Hayes, & Simon, 1985; Schneider & Schiffrin, 1977; Shiffrin & Schneider, 1977), only examples of potential schema acquisition activities were provided by the early CLT developments.
Cognitive Load Theory
15
Free Capacity
Total Working Memory Capacity
Schema Acquisition & Automation Activities
Other Mental Activities
Extraneous Load = Total Cognitive Load
figure 1.1. A visual representation of the assumptions underlying the first stage of cognitive load theory development.
A precise definition of the mental activities that promote schema acquisition is not only necessary to make predictions about learning from different instructional designs, it is crucial to accurately predict cognitive load and learning (the main goal of CLT) because the definition of extraneous processing is, by exclusion, “any activity not directed to schema acquisition and automation” (Sweller & Chandler, 1994, p. 192). Second, there was a question as to whether engaging in schema acquisition and automation activities would have any effects on the learner’s cognitive load. If so, how might this type of load be distinguished from the extraneous load that was posited to produce negative learning effects? As will be seen in the next sections, this is still a topic of intense debate and one that led to the current triarchic theory of cognitive load (see Chapter 7, this volume). Figure 1.1 summarizes the assumptions of CLT during its first development stage. Stage II: Intrinsic Cognitive Load and the First Additivity Hypothesis The second stage of CLT is characterized by the introduction of an additional source of cognitive load, namely, intrinsic cognitive load. CLT moved from focusing solely on the extraneous cognitive load that may originate by the way in which instructional materials and methods are designed to including the load that “is imposed by the basic characteristics of information” (Sweller, 1994, p. 6). More specifically, some materials are difficult to learn or some problems are difficult to solve because they require processing several elements that simultaneously interact with each other. Intrinsic load can, therefore, be estimated “by counting the number of elements that
16
Roxana Moreno and Babette Park
must be considered simultaneously in order to learn a particular procedure” (Sweller & Chandler, 1994, p. 190). The very idea of intrinsic load stemmed from the research of Halford, Maybery, and Bain (1986) and Maybery, Bain, and Halford (1986), as cited by Sweller and colleagues (Sweller, 1993; Sweller & Chandler, 1994). In these investigations, the researchers found that difficulty in the processing of transitive inference in children’s reasoning (e.g., a is taller than b; b is taller than c; which is the largest?) was heavily influenced by the need to consider all the elements of the premises simultaneously. According to CLT, intrinsic cognitive load depends on two factors: the number of elements that must be simultaneously processed in working memory on any learning task and the prior knowledge of the learner. The load resulting from element interactivity varies among and within different subject areas. For instance, solving algebra problems involves dealing with higher element interactivity than learning the vocabulary of a second language, and creating grammatically correct sentences in a second language involves higher element interactivity than learning the vocabulary itself (Sweller, 1993). In addition, prior knowledge has an effect on intrinsic load in that a large number of interacting elements for a novice may be a single element for an expert who has integrated the interacting elements in one schema. In addition to introducing a second source of cognitive load, the first version of what we have called the “additivity hypothesis” was developed during this stage (see Chapter 12, this volume): When people are faced with new material, the cognitive load imposed by that material will consist of the intrinsic cognitive load due to element interactivity and extraneous cognitive load determined by the instructional design used. If that total cognitive load is excessive, learning and problem solving will be inhibited. (Sweller, 1993, p. 7)
The original additivity hypothesis motivated the development of the following two additional CLT assumptions. The first one is that extraneous load is the only source of load that can be reduced by good instructional design. In contrast, instructors have no control over intrinsic load (Sweller, 1994), a claim that continues to be held by many cognitive load researchers (Paas, Renkl, & Sweller, 2003). The practical implication of this new assumption is that the only way to manage high intrinsic load is to help students develop cognitive schemata that incorporate the interacting elements. The second new assumption embedded in the original additivity hypothesis is that the extent to which extraneous load should be reduced depends
17
Cognitive Load Theory
Free Capacity
Total Working Memory Capacity
Extraneous Load (Reducible by instructional design) Intrinsic Load (Irreducible by instructional design)
Total Cognitive Load
figure 1.2. A visual representation of the assumptions underlying the second stage of cognitive load theory development.
on the existing level of intrinsic load: if the level of intrinsic load is low, then a high extraneous load may not impede learning because students are able to handle low interactivity material; if intrinsic load is high, adding a high extraneous load will result in a total load that might exceed cognitive resources (Sweller, 1994; Sweller & Chandler, 1994). This assumption refocused CLT as a theory that is mainly concerned with the learning of complex tasks, where students are typically overwhelmed by the amount of elements and interactions that need to be processed simultaneously (Paas, Renkl, & Sweller, 2004). Despite the fact that the theory has offered methods to measure element interactivity (Sweller & Chandler, 1994), what is not clear is the criterion for determining a priori when a task is sufficiently complex to be likely to produce cognitive-load effects on learning. Figure 1.2 summarizes the assumptions of CLT at its second stage of development. Stage III: Germane Cognitive Load and the Second Additivity Hypothesis More recently, CLT has undergone two major revisions. The first one was the introduction of the third source of cognitive load, germane load. The distinctive characteristic of germane cognitive load is that, unlike the other two, it has a positive relationship with learning because it is the result of devoting cognitive resources to schema acquisition and automation rather than to other mental activities. The idea of germane load originated from the need to specify the cognitive-load effects of the schema acquisition and automation activities that were proposed to be beneficial to learning according to the original CLT. In addition, many scholars questioned the theory’s focus on the need to reduce extraneous load and the idea
18
Roxana Moreno and Babette Park
Free Capacity Germane Load (Increasable by instructional design) Total Working Memory Capacity
Extraneous Load (Reducible by instructional design)
Total Cognitive Load
Intrinsic Load (Irreducible by instructional design??)
figure 1.3. A visual representation of the assumptions underlying the most recent cognitive load theory development.
that the purpose of instruction is to keep mental effort at a minimum during the learning process. According to the revised theory, “as long as the load is manageable, it is not the level of load that matters but its source” (Paas et al., 2004, p. 3). Freeing available cognitive capacity by reducing extraneous load will not necessarily result in increased learning unless the freed resources are directed to activities that are relevant for schema acquisition. The second development during this stage was the revision of the additivity hypothesis, which now integrates the three load sources as follows: Intrinsic, extraneous, and germane cognitive loads are additive in that, together, the total load cannot exceed the working memory resources available if learning is to occur. The relations between the three forms of cognitive load are asymmetric. Intrinsic cognitive load provides a base load that is irreducible other than by constructing additional schemas and automating previously acquired schemas. (Paas et al., 2003, p. 2)
Figure 1.3 summarizes the assumptions of CLT in its current form. The revised theory has inspired a large amount of empirical work that has been useful in pointing out the theoretical and methodological issues that still need investigation. We discuss a few of the major open questions next. First, a careful review of the most recent stage of CLT development reveals that the theory is still unclear about the irreducible nature of intrinsic
Cognitive Load Theory
19
cognitive load. Several theoretical articles hold the initial assumption that intrinsic load cannot be altered by the instructional designer (Paas et al., 2003; Paas, Tuovinen, Tabbers, & van Gerven, 2003). Others suggest that intrinsic load may be altered by reducing element interactivity, although “by artificially reducing intrinsic cognitive load, understanding is also reduced” (van Merri¨enboer & Sweller, 2005, p. 157), suggesting that a reduction in the intrinsic difficulty of the materials to be learned hurts learning. This conclusion is puzzling in that, according to the additivity hypothesis, reductions in this load type should free cognitive resources that can be used in other, productive mental activities. The disagreement on this assumption is represented by the question marks next to the intrinsic cognitive load construct in Figure 1.3 and has motivated several studies that attempted to reduce intrinsic load using instructional methods such as pretraining, sequencing, and chunking (Gerjets, Scheiter, & Catrambone, 2004, 2006; Lee, Plass, & Homer, 2006; Pollock, Chandler, & Sweller, 2002; van Merri¨enboer, Kirschner, & Kester, 2003). However, an issue that emerges when reviewing this research is that it is difficult to reconcile the findings because they do not consistently support either prediction. Moreover, even in the cases in which a positive learning effect is found (for all or a subgroup of students), it is possible to offer an alternative explanation using the same theory. More specifically, the original instructional design of the materials (without the intervention to reduce intrinsic load) may have been poorly designed. The success of the intervention can then be explained as the result of having reduced the extraneous source of load imposed by the original instructional design. Unfortunately, the lack of good measures of intrinsic, extraneous, and germane load has prevented this research from advancing CLT because it is not possible to understand the causes of the diverse findings (see Chapter 9, this volume). In the same vein, the introduction of the third load type in CLT has inspired many studies in which manipulations of germane sources of load are examined (Renkl, Atkinson, & Grobe, 2004; Gerjets et al., 2004; Berthold & Renkl, 2009). For instance, some studies have investigated the learning effects of combining methods aimed at reducing extraneous load and increasing germane load, thus redirecting students’ cognitive resources from irrelevant to relevant schema acquisition activities (Seufert & Br¨unken, 2006; Seufert, J¨anen, & Br¨unken, 2007). In these studies, the investigators tested the CLT hypothesis that methods that reduce extraneous load will free cognitive resources, which then can be used to engage in schema acquisition activities.
20
Roxana Moreno and Babette Park
Stage IV: The Evolutionary Interpretation of CLT For the past twenty years, cognitive load research has focused on investigating the effects of instructional design on learning and cognitive load using the theoretical developments summarized in the previous sections. Very recently, however, Sweller (2003, 2004) has suggested that the principles underlying evolution by natural selection are the same as those underlying the human cognitive architecture, therefore offering an evolutionary perspective about CLT. More specifically, five biological principles, namely, the information store, borrowing, randomness as genesis, narrow limits of change, and environmental organizing and linking principles, are proposed as an essential connection between the human cognitive system and human biology. We do not describe this new interpretation of CLT here because John Sweller provides a detailed description of the five principles and their instructional implications in the next chapter of this book.
where does clt fit in the big picture? In this section, we examine the constructs and assumptions of CLT in relation to other theories in psychology and education. The relation of CLT to the theoretical models of multimedia learning offered by Mayer (2001) and Schnotz (2001) are given special consideration in a separate chapter (see Chapter 11, this volume); therefore, they are not discussed in this chapter. An important first point that needs to be made is that CLT was never claimed to be a learning theory. Instead, it was articulated to explain the relation between the human cognitive architecture, instructional design, and learning. Nevertheless, CLT should be parsimonious with contemporary thought in the learning sciences. In this regard, it should be noted that when CLT was first advanced, it was heavily inspired by the contributions that cognitive psychology had to offer at that point in time. Although several developments have refined the theory throughout its lifespan, none of them have challenged the basic assumptions drawn from the computational models of the mind that inspired the theory (Simon & Gilmartin, 1973). Cognitively based approaches to learning, however, are only one of the four main frameworks under which learning theories fall. The other three are behaviorism, which focuses only on the objectively observable aspects of learning (Guthrie, 1959); sociocognitive theories of learning, which focus on the type of learning that occurs even when there is no direct interaction with the environment (Bandura, 2000); and constructivist learning theories,
Cognitive Load Theory
21
which view learning as a process in which the learner actively constructs or builds new ideas or concepts. How does CLT relate to these approaches? Similar to other cognitively based theories, CLT looks beyond behavior by concentrating on “unobservable” phenomena, specifically, the cognitive load experienced by individuals in different instructional conditions. Unlike sociocognitive theories of learning, CLT has limited its scope to enactive learning scenarios, in which students learn by experiencing the consequences of their own actions. Moreover, a sociocognitive area that continues to receive increasing attention from educational psychologists and educators alike is self-regulation (Boekaerts, Pintrich, & Zeidner, 2000; Winne, 2005; Zimmerman, 2002). Self-regulated learners are able to set more specific learning goals, use more learning strategies, better monitor their learning, and more systematically evaluate their progress toward learning goals than their counterparts (Boekaerts, 2006). Although these learner characteristics are very likely to have an effect on how successfully students deal with high cognitive load situations, CLT does not currently provide insight into how the internal and external management of cognitive load may interact (Bannert, 2000). Finally, the relation between CLT and constructivism merits special consideration. Constructivist theories have been very influential in guiding educational practices and curriculum, and have become the basis for the standards of teaching developed by national education groups, such as the National Council of Teachers of Mathematics (2000) and the American Association for the Advancement of Science (1993). Constructivist learning theories focus not only on how people construct knowledge within themselves (i.e., individual constructivism) but also on how they co-construct knowledge with others (i.e., social constructivism). Furthermore, social constructivist perspectives extend sociocognitive theories by considering a wider range of social influences, such as those stemming from individuals’ culture, history, and direct interaction with others. CLT does not include assumptions about the relationship between cognitive load, instruction, and co-constructing knowledge with others. The psychological theory of distributed cognition, however, suggests that cognitive processes may be distributed across the members of a social group (Hutchins, 1995), and Vygotsky’s (1978) social constructivism asserts that the interactions with “more knowledgeable others” support the development of individuals’ schemas and thinking processes. Therefore, differences in effort and learning are likely to arise when students work alone compared with learning with more capable others or in groups (Moreno, 2009). Nevertheless, CLT has provided explanations for the ineffective use of some of the most advocated constructivist strategies of the past three or
22
Roxana Moreno and Babette Park
four decades, such as discovery learning, inquiry, and problem-based learning. Some forms of discovery ask students to try to find a solution to a problem or an explanation for a phenomenon with minimum guidance (Kato, Honda, & Kamii, 2006). According to CLT, when learners are novices in a domain, the cognitive load associated with unguided discovery is too high to promote learning because novices lack well-developed schemas to guide their knowledge construction process (Kirshner, Sweller, & Clark, 2006; Tuovinen & Sweller, 1999). This idea has received empirical support. Poorer learning outcomes related to unguided discovery appear rather general (Mayer, 2004; Moreno & Mayer, 2007; Taconis, Ferguson-Hessler, & Broekkamp, 2001). However, the research shows that when problem-based learning and inquiry methods are designed to support high-order thinking, they can be highly effective. As explained by Hmelo-Silver, Duncan, and Chinn (2007), an important difference between the discovery method and the other two constructivist methods is that the latter typically include a large range of scaffolds to guide the process of constructing knowledge individually or in groups (see Chapter 6, this volume). Interestingly, cognitive load theorists disagree among each other when called to evaluate the effectiveness of problem-based learning and inquiry methods using the CLT framework. Some argue that inquiry and problem-based learning are instructional approaches that allow for flexible adaptation of guidance; therefore, they are compatible with the way in which the human cognitive structures are organized and can be effectively used to promote learning (Schmidt, Loyens, van Gog, & Paas, 2007). Others cite the classic literature on the worked-example effect and argue that presenting the solution of a problem to novice students should always lead to better learning than requiring them to search for a solution, which will necessarily lead to a heavy working memory load (Sweller, Kirschner, & Clark, 2007). Because constructivist methods are aimed at more actively engaging learners in the learning process, it seems that they are likely to be good candidates for promoting germane load and learning (see Chapter 8, this volume). Similar to other open questions identified in this chapter, the question of whether and how instruction should be designed to support students’ knowledge construction is likely to remain open until carefully controlled experimental studies, including appropriate controls and measures of the three cognitive load types, are conducted. The third part of this volume is dedicated to discussing in detail some potential venues that basic research (Chapter 12, this volume) and neuroscience (Chapter 10, this volume) have to offer to advance CLT in future years.
Cognitive Load Theory
23
references American Association for the Advancement of Science. (1993). Benchmarks for science literacy. Washington, DC: Author. Anderson, J. R., & Bower, G. (1983). Human associative memory. Washington, DC: Winston. Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motivation (Vol. 2, pp. 89–195). New York: Academic Press. Ausubel, D. P. (1960). The use of advance organizers in the learning and retention of meaningful verbal materials. Journal of Educational Psychology, 51(5), 267–272. Bandura, A. (2000). Social-cognitive theory. In A. E. Kazdin (Ed.), Encyclopedia of psychology (Vol. 7, pp. 329–332). Washington, DC: American Psychological Association. Bannert, M. (2000). Effects of training wheels and self-learning materials in software training. Journal of Computer-Assisted Learning, 16(4), 336–346. Bannert, M. (2002). Managing cognitive load, recent trends in cognitive load theory. Learning and Instruction, 12, 139–146. Bartlett, F. (1932). Remembering: A study in experimental and social psychology. New York: Cambridge University Press. Berthold, K., & Renkl, A. (2009). Instructional aids to support a conceptual understanding of multiple representations. Journal of Educational Psychology, 101(1), 70–87. Boekaerts, M. (2006). Self-regulation and effort investment. In E. Sigel & K. A. Renninger, (Vol. Eds.), Handbook of child psychology (6th ed., Vol. 4) Child psychology in practice (pp. 345–377). Hoboken, NJ: Wiley. Boekaerts, M., Pintrich, P., & Zeidner, M. (Eds.). (2000). Handbook of self-regulation. San Diego, CA: Academic Press. Bratfisch, O., Borg, G., & Dornic, S. (1972). Perceived item-difficulty in three tests of intellectual performance capacity (Tech. Rep. No. 29). Stockholm: Institute of Applied Psychology. Bruner, J. S., Goodnow, J. J., & Austin, G. A. (1956). A study of thinking. New York: Wiley. Campbell, D. T. (1988). Task complexity: A review and analysis. Academy of Management Review, 13, 40–52. Chandler, P., & Sweller, J. (1991). Cognitive load theory and the format of instruction. Cognition and Instruction, 8, 293–332. Chandler, P., & Sweller, J. (1992). The split-attention effect as a factor in the design of instruction. British Journal of Educational Psychology, 62, 233–246. Chandler, P., & Sweller, J. (1996). Cognitive load while learning to use a computer program. Applied Cognitive Psychology, 10, 151–170. Chase, W. G., & Simon, H. A. (1973). The mind’s eye in chess. In W. G. Chase (Ed.), Visual information processing (pp. 215–281). New York: Academic Press. Chi, M. T. H., Glaser, R., & Rees, E. (1982). Expertise in problem solving. In R. Sternberg (Ed.), Advances in the psychology of human intelligence (pp. 7–76). Hillsdale, NJ: Erlbaum. Chomsky, N. (1957). Syntactic structures. Oxford, England: Mouton.
24
Roxana Moreno and Babette Park
Cooper, G., & Sweller, J. (1987). The effects of schema acquisition and rule automation on mathematical problem-solving transfer. Journal of Educational Psychology, 79, 347–362. De Groot, A. (1966). Perception and memory versus thought: Some old ideas and recent findings. In B. Kleinmuntz (Ed.), Problem solving: Research, method, and theory (pp. 19–50). New York: Wiley. Eccles, J., Wigfield, A., & Schiefele, U. (1998). Motivation to succeed. In W. Damon & N. Eisenberger (Eds.), Handbook of child psychology (5th ed., Vol. 3, pp. 1017–1095). New York: Wiley. Egan, D. E., & Schwartz, B. J. (1979). Chunking in recall of symbolic drawings. Memory & Cognition, 7, 149–158. Fisk, A. D., & Schneider, W. (1983). Category and work search: Generalizing search principles to complex processing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9, 117–195. Gerjets, P., Scheiter, K., & Catrambone, R. (2004). Designing instructional examples to reduce intrinsic cognitive load: Molar versus modular presentation of solution procedures. Instructional Science, 32, 33–58. Gerjets, P., Scheiter, K., & Catrambone, R. (2006). Can learning from molar and modular worked-out examples be enhanced by providing instructional explanations and prompting self-explanations? Learning and Instruction, 16, 104–121. Guthrie, E. R. (1959). Association by contiguity. In S. Koch (Ed.), Psychology: A study of science (Vol. 2, pp. 158–195). New York: McGraw-Hill. Halford, G., Maybery, M., & Bain, J. (1986). Capacity limitations in children’s reasoning: A dual task approach. Child Development, 57, 616–627. Hancock, P. A., & Desmond, P. A. (2001). Stress, workload and fatigue. London: Lawrence Erlbaum Associates. Hancock, P. A., & Meshkati, N. (1988). Human mental workload. Amsterdam: NorthHolland. Hmelo-Silver, C. E., Duncan, R. G., & Chinn, C. A. (2007). Scaffolding and achievement in problem-based and inquiry learning: A response to Kirschner, Sweller, and Clark (2006). Educational Psychologist, 42, 99–107. Huey, B. M., & Wickens, C. D. (1993). Workload transition: Implications for individual and team performance. Washington, DC: National Academy Press. Hutchins, E. (1995). Cognition in the wild. Cambridge, MA: MIT Press. Kahneman, D. (1973). Attention and effort. Englewood Cliffs, NJ: Prentice-Hall. Kalyuga, S., Chandler, P., & Sweller, J. (1998). Levels of expertise and instructional design. Human Factors, 40, 1–17. Kalyuga, S., Chandler, P., & Sweller, J. (1999). Managing split-attention and redundancy in multimedia instruction. Applied Cognitive Psychology, 13, 351–357. Kantowitz, B. H. (1987). Mental workload. In P. A. Hancock (Ed.), Human factors psychology (pp. 81–121). Amsterdam: North-Holland. Kato, Y., Honda, M., & Kamii, C. (2006). Kindergartners play lining up the 5s: A card game to encourage logical-mathematical thinking. Young Children, 61(4), 82–88. Kirshner, P., Sweller, J., & Clark, R. (2006). Why minimal guidance during instruction does not work: An analysis of the failure of constructivist, discovery, problembased, experiential, and inquiry-based teaching. Educational Psychologist, 41, 75– 86.
Cognitive Load Theory
25
Kotovsky, K., Hayes, J. R., & Simon, H. A. (1985). Why are some problems hard? Evidence from Tower of Hanoi. Cognitive Psychology, 17, 248–294. Lee, H., Plass, J. L., & Homer, B. D. (2006). Optimizing cognitive load for learning from computer science simulations. Journal of Educational Psychology, 98, 902– 913. Logan, G. D. (1979). On the use of concurrent memory load to measure attention and automaticity. Journal of Experimental Psychology: Human Perception and Performance, 5, 189–207. Logan, G. D. (1980). Attention and automaticity in Stroop and priming tasks: Theory and data. Cognitive Psychology, 12, 523–553. MacDonald, W. (2003). The impact of job demands and workload on stress and fatigue. Australian Psychologist, 36, 102–117. Madden, J. M. (1962). What makes work difficult? Personnel Journal, 41, 341–344. Maybery, M., Bain, J., & Halford, G. (1986). Information processing demands of transitive inference. Journal of Experimental Psychology: Learning Memory and Cognition, 12, 600–613. Mayer, R. E. (2001). Multimedia learning. New York: Cambridge University Press. Mayer, R. E. (2004). Should there be a three strikes rule against pure discovery learning? The case for guided methods of instruction. American Psychologist, 59(1), 14–19. Mead, D. F. (1970). Development of an equation for evaluating job difficulty. Brooks, TX: Air Force Human Resources Laboratory. Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97. Moray, N. (1979). Mental workload: Its theory and measurement. New York: Plenum. Moreno, R. (2005). Instructional technology: Promise and pitfalls. In L. PytlikZillig, M. Bodvarsson, & R. Bruning (Eds.), Technology-based education: Bringing researchers and practitioners together (pp. 1–19). Greenwich, CT: Information Age Publishing. Moreno, R. (2006). When worked examples don’t work: Is cognitive load theory at an impasse? Learning and Instruction, 16, 170–181. Moreno, R. (2009). Constructing knowledge with an agent-based instructional program: A comparison of cooperative and individual meaning making. Learning and Instruction, 19, 433–444. Moreno, R., & Mayer, R. E. (2007). Interactive multimodal learning environments. Educational Psychology Review, 19, 309–326. National Council of Teachers of Mathematics. (2000). Principles and standards for school mathematics. Reston, VA: Author. Owen, E., & Sweller, J. (1985). What do students learn while solving mathematics problems? Journal of Educational Psychology, 77, 272–284. Paas, F. (1992). Training strategies for attaining transfer of problem-solving skill in statistics: A cognitive load approach. Journal of Educational Psychology, 84, 429–434. Paas, F., Renkl, A., & Sweller, J. (2003). Cognitive load theory and instructional design: Recent developments. Educational Psychologist, 38, 1–4. Paas, F., Renkl, A., & Sweller, J. (2004). Cognitive load theory: Instructional implications of the interaction between information structures and cognitive architecture [Guest editorial statement]. Instructional Science, 32, 1–8.
26
Roxana Moreno and Babette Park
Paas, F., Tuovinen, J., Tabbers, H., & van Gerven, P. W. M. (2003). Cognitive load measurement as a means to advance cognitive load theory. Educational Psychologist, 38, 63–71. Paas, F., & van Merri¨enboer, J. J. G. (1993). The efficiency of instructional conditions: An approach to combine mental-effort and performance measures. Human Factors, 35, 737–743. Paas, F., & van Merri¨enboer, J. J. G. (1994). Variability of worked examples and transfer of geometrical problem-solving skills: A cognitive-load approach. Journal of Educational Psychology, 86, 122–133. Pollock, E., Chandler, P., & Sweller, J. (2002). Assimilating complex information. Learning and Instruction, 12, 61–86. Posner, M. I., & Snyder, C. R. R. (1975). Attention and cognitive control. In R. L. Solso (Ed.), Information processing and cognition: The Loyola symposium (pp. 212– 225). Hillsdale, NJ: Erlbaum. Renkl, A., Atkinson, R., & Grobe, C. S. (2004). How fading worked solution steps works – A cognitive load perspective. Instructional Science, 32, 59–82. Rumelhart, D. E., & Ortony, A. (1976). The representation of knowledge in memory. In R. C. Anderson, R. J. Spiro, & W. E. Montague (Eds.), Schooling and the acquisition of knowledge (pp. 99–136). Hillsdale, NJ: Erlbaum. Schmidt, H. G., Loyens, S. M. M., van Gog, T., & Paas, F. (2007). Problem based learning is compatible with human cognitive architecture: Commentary on Kirschner, Sweller, and Clark (2006). Educational Psychologist, 42, 91–97. Schneider, W., & Schiffrin, R. M. (1977). Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 84, 1–66. Schnotz, W. (2001). Sign systems, technologies, and the acquisition of knowledge. In J. F. Rouet, J. Levonen, & A. Biardeau (Eds.), Multimedia learning? Cognitive and instructional issues (pp. 9–29). Amsterdam: Elsevier. Seufert, T., & Br¨unken, R. (2006). Cognitive load and the format of instructional aids for coherence formation. Applied Cognitive Psychology, 20, 321– 331. Seufert, T., J¨anen, I., & Br¨unken R. (2007). The impact of intrinsic cognitive load on the effectiveness of graphical help for coherence formation. Computers in Human Behavior, 23, 1055–1071. Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing. II. Perceptual learning, automatic attending and a general theory. Psychological Review, 84(2), 127–190. Simon, D., & Simon, H. (1978). Individual differences in solving physics problems. In R. S. Siegler (Ed.), Children’s thinking: What develops? (pp. 325–348). Hillsdale, NJ: Erlbaum. Simon, H. A., & Gilmartin, K. J. (1973). A simulation of memory for chess positions. Cognitive Psychology, 5, 29–46. Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12, 257–285. Sweller, J. (1989). Cognitive technology: Some procedures for facilitating learning and problem solving in mathematics and science. Journal of Educational Psychology, 81, 457–466.
Cognitive Load Theory
27
Sweller, J. (1993). Some cognitive processes and their consequences for the organisation and presentation of information. Australian Journal of Educational Psychology, 45, 1–8. Sweller, J. (1994). Cognitive load theory, learning difficulty, and instructional design. Learning and Instruction, 4, 295–312. Sweller, J. (2003). Evolution of human cognitive architecture. In B. Ross (Ed.), The psychology of learning and motivation (Vol. 43, pp. 215–266). San Diego, CA: Academic Press. Sweller, J. (2004). Instructional design consequences of an analogy between evolution by natural selection and human cognitive architecture. Instructional Science, 32, 9–31. Sweller, J., & Chandler, P. (1994). Why some material is difficult to learn. Cognition and Instruction, 12(3), 185–233. Sweller, J., Chandler, P., Tierney, P., & Cooper, M. (1990). Cognitive load as a factor in the structuring of technical material. Journal of Experimental Psychology: General, 119, 176–192. Sweller, J., & Cooper, G. (1985). The use of worked examples as a substitute for problem solving in learning algebra. Cognition and Instruction, 2(1), 59–89. Sweller, J., Kirschner, P., & Clark, R. (2007). Why minimally guided teaching techniques do not work: A reply to commentaries. Educational Psychologist, 42, 115–121. Sweller, J., Mawer, R., & Ward, M. (1983). Development of expertise in mathematical problem solving. Journal of Experimental Psychology: General, 112(4), 639–661. Sweller, J., van Merri¨enboer, J., & Paas, F. (1998). Cognitive architecture and instructional design. Educational Psychology Review, 10, 251–296. Taconis, R., Ferguson-Hessler, M. G. M., & Broekkamp, H. (2001). Teaching science problem solving: An overview of experimental work. Journal of Research in Science Teaching, 38, 442–468. Tarmizi, R. A., & Sweller, J. (1988). Guidance during mathematical problem solving. Journal of Educational Psychology, 80, 424–436. Thrash, T., & Elliott, A. (2001). Delimiting and integrating achievement motive and goal constructs. In A. Efklides, J. Kuhl, & R. Sorrentino (Eds.), Trends and prospects in motivation research (pp. 3–21). Boston: Kluwer. Tuovinen, J., & Sweller, J. (1999). A comparison of cognitive load associated with discovery learning and worked examples. Journal of Educational Psychology, 91, 334–341. van Merri¨enboer, J. J. G., & Ayres, P. (2005). Research on cognitive load theory and its design implications for e-learning, educational technology, research and development, Educational Technology Research and Development, 53(3), 5–13. van Merri¨enboer, J. J. G., & de Croock, M. (1992). Strategies for computer-based programming instruction: Program completion vs. program generation. Journal of Educational Computing Research, 8(3), 365–394. van Merri¨enboer, J. J. G., Kirschner, P. A., & Kester, L. (2003). Taking the load off a learner’s mind: Instructional design for complex learning. Educational Psychology, 38, 5–13. van Merri¨enboer, J., & Sweller, J. (2005). Cognitive load theory and complex learning: Recent developments and future directions. Educational Psychology Review, 17, 147–177.
28
Roxana Moreno and Babette Park
Vygotsky, L. S. (1978). Mind in society: The development of higher mental process. Cambridge, MA: Harvard University Press. Wickens, C. D., & Hollands, J. G. (2000). Engineering psychology and human performance. Upper Saddle River, NJ: Prentice-Hall. Winne, P. H. (2005). Key issues in modeling and applying research on self-regulated learning. Applied Psychology: An International Review, 54, 232–238. Wood, R. E. (1986). Task complexity: Definition of the construct. Organizational Behavior and Human Decision Processes, 37, 60–82. Zimmerman, B. J. (2002). Achieving self-regulation: The trial and triumph of adolescence. In F. Pajares & T. Urdan (Eds.), Academic motivation of adolescents (pp. 1–27). Greenwich, CT: Information Age Publishing.
2 Cognitive Load Theory: Recent Theoretical Advances john sweller
Cognitive Load Theory (CLT) began as an instructional theory based on our knowledge of human cognitive architecture. It proved successful in generating a series of cognitive load effects derived from the results of randomised, controlled experiments (Clark, Nguyen, & Sweller, 2006). This chapter summarises the theory, including its general instructional implications. Many of the theory’s specific instructional implications, which provide its prime function and purpose, are discussed in other chapters in this volume and therefore will not be discussed in detail in this chapter (see Table 2.1 for a summary).
human cognition The processes of human cognition constitute a natural information-processing system that mimics the system that gave rise to human cognitive architecture: evolution by natural selection. Both human cognition and biological evolution create novel information, store it for subsequent use, and are capable of disseminating that information indefinitely over space and time. By considering human cognition within an evolutionary framework, our understanding of the structures and functions of our cognitive architecture are being transformed. In turn, that cognitive architecture has profound instructional consequences. CLT is an amalgam of human cognitive architecture and the instructional consequences that flow from that architecture. From an evolutionary perspective, there are two categories of human knowledge: biologically primary and biologically secondary knowledge (Geary, 2007, 2008). Biologically primary knowledge is knowledge we have evolved to acquire over many generations. Examples are general problemsolving techniques, recognising faces, engaging in social relations, and 29
30
John Sweller table 2.1. Cognitive load effects
Cognitive load effect Worked-Example
Completion
Split-Attention
Modality
Redundancy
Expertise reversal
Guidance fading
Goal-Free
Element interactivity
Isolated/interacting elements
Variable examples
Imagination
Description Studying worked examples results in better performance on subsequent tests of problem solving than solving the equivalent problems (Renkl, 2005). Requiring learners to complete partially solved problems can be just as effective as worked examples (Paas & van Merri¨enboer, 1994). Multiple sources of information that are unintelligible in isolation result in less learning when they are presented in split-attention as opposed to integrated format (Ayres & Sweller, 2005). Multiple sources of information that are unintelligible in isolation result in less learning when they are presented in single-modality as opposed to dual-modality format (Low & Sweller, 2005). The presence of sources of information that do not contribute to schema acquisition or automation interfere with learning (Sweller, 2005). With increasing expertise, instructional procedures that are effective with novices can lose their effectiveness, whereas ineffective techniques can become effective (Kalyuga, 2005). With increasing expertise, learners should be presented worked examples followed by completion problems and then full problems rather than worked examples alone (Renkl, 2005). Problems presented in goal-free form enhance learning compared with conventional problems (Paas, Camp, & Rikers, 2001). Cognitive load effects are only obtainable using high rather than low element interactivity material (Sweller, 1994). Learning is enhanced if very high element interactivity material is first presented as isolated elements followed by interacting elements versions rather than as interacting elements form initially (Pollock, Chandler & Sweller, 2002). Examples with variable surface features enhance learning compared with examples with similar features (Paas & van Merri¨enboer, 1994). Imagining procedures or concepts enhance learning compared with studying materials (Leahy & Sweller, 2004).
Primary cognitive load source Extraneous
Extraneous
Extraneous
Extraneous
Extraneous
Extraneous
Extraneous
Extraneous
Intrinsic
Intrinsic
Germane
Germane
Cognitive Load Theory: Recent Theoretical Advances
31
listening to and speaking our native language. Primary knowledge is modular in that we have independent, cognitive modules that allow us to acquire the relevant knowledge unconsciously, effortlessly, and without external motivation simply by membership in a human society. Learning to speak our native language provides a clear example. We are not normally explicitly taught how to organise our lips, tongue, voice, and breath when learning to speak. We have evolved to learn these immensely complex procedures just by listening to others speaking. In contrast, biologically secondary knowledge is culturally dependent. We have evolved to acquire such knowledge in a general sense rather than having evolved to acquire particular knowledge modules such as speaking. Biologically secondary knowledge is acquired consciously and usually requires mental effort. In modern times, we invented educational institutions to impart biologically secondary knowledge precisely because, unlike biologically primary knowledge, it tends not be learned simply by immersion in a functioning society. Virtually everything taught in educational institutions consists of biologically secondary knowledge. For example, unlike listening and speaking, few people are likely to learn to read and write without being explicitly taught to read and write. Simple immersion in a reading and writing society is unlikely to be sufficient. CLT and, indeed, instructional design in general, applies to biologically secondary knowledge (Sweller, 2007, 2008). It does not apply to biologically primary knowledge. Thus, CLT is relevant to those aspects of knowledge dealt with in educational institutions (secondary knowledge) rather than the possibly far larger body of primary knowledge that we have specifically evolved to acquire. When dealing with biologically secondary knowledge, human cognition can be characterised by five basic principles that govern its functions and processes. These principles apply equally to the processes that govern biological evolution (Sweller, 2003, 2004; Sweller & Sweller, 2006) and as such constitute a natural information processing system. They will be discussed in more detail subsequently but can be summarised as follows. The information store principle states that human cognition includes a large store of information that governs the bulk of its activity. Long-term memory provides this function. The borrowing and reorganising principle states that almost all of the information held in long-term memory has been borrowed from other long-term memory stores. Information obtained by imitation, listening, or reading exemplifies this process. The randomness as genesis principle indicates that random generation followed by tests of effectiveness provide the initial source for the generation
32
John Sweller
of all information held in long-term memory. When faced with a problem for which solution knowledge is not available or only partly available, the random generation of moves followed by tests of the effectiveness of those moves is an example. The narrow limits of change principle indicates that all effective changes to long-term memory occur slowly and incrementally. The capacity limitations of working memory when dealing with novel information exemplifies this principle. The environment organising and linking principle states that unlimited amounts of organised information from long-term memory can be used by working memory to determine interactions with the external world. These principles constitute a natural information processing system and derive from evolutionary theory. Although the derivation will not be discussed in this chapter (see Sweller, 2003, 2004; Sweller & Sweller, 2006), the link with biological evolution establishes an essential connection between the human cognitive system and human biology. A detailed description of the five principles follows. Long-Term Memory and the Information Store Principle Functioning in a complex environment requires a complex store of information to govern activity in that environment. The primary driver of activity of human cognition is its large store of information held in long-term memory. The realisation that long-term memory is not simply a repository of isolated, near-random facts but, rather, the central structure of human cognition, developed slowly. Its origins can probably be traced to the early work on expertise in the game of chess. When De Groot (1965) followed by Chase and Simon (1973) found that the only difference between chess masters and less able players was in memory of chess-board configurations taken from real games, it established the central importance of long-term memory to cognition. Chess has long been seen, appropriately, as a game that required the most sophisticated of human cognitive processes, a game of problem solving and thought. The discovery that the major difference between people who differed in ability was in terms of what they held in long-term memory changed our view of cognition. (It should be noted that what is held in long-term memory includes problem-solving strategies.) Long-term memory was not just used by humans to reminisce about the past but, rather, was a central component of problem solving and thought. If long-term memory is essential to problem solving, we might expect results similar to those obtained using the game of chess to also be obtained in educationally more relevant areas. Increasing levels of expertise should be associated with increasing ability to reproduce relevant problem states
Cognitive Load Theory: Recent Theoretical Advances
33
and, indeed, the same expert–novice differences obtained in chess have been established in a variety of educationally relevant areas (e.g. Egan & Schwartz, 1979; Jeffries, Turner, Polson, & Atwood, 1981). The amount of information required by the human cognitive system is huge. Although we have no metric for measuring the amount of information held in long-term memory, it might be noted that Simon and Gilmartin (1973) estimated that chess grand masters have learned to recognise many tens of thousands of the board configurations that are required for their level of competence. It is reasonable to assume that similar numbers of knowledge elements are required for skilled performance in areas more relevant to everyday life, including areas covered in educational contexts. If so, long-term memory holds massive amounts of information to permit adequate levels of performance in the various areas in which an individual is competent. Although a large amount of information is held in long-term memory, where does the information come from? The next two sections discuss this issue. Schema Theory and the Borrowing and Reorganising Principle Based on the information store principle, human cognition includes a large store of information that governs most activity. What is the immediate source of that information? The borrowing and reorganising principle explains how most of the information found in any individual’s longterm memory is acquired. Almost all information in long-term memory is obtained by imitating other people’s actions or hearing or reading what others have said. In effect, our knowledge base is borrowed almost entirely from the long-term memory of other people. Nevertheless, the information borrowed is almost invariably altered and constructed. We do not remember exactly what we have heard or seen but, rather, construct a representation based on knowledge already held in long-term memory. Schema theory reflects that constructive process. A schema permits multiple elements of information to be treated as a single element according to the manner in which it will be used. Thus, a problem-solving schema permits us to classify problems according to their solution mode. A chess master has schemas that allow the classifying of chess-board configurations according to the moves required. The modern origins of schema theory can be found in Piaget (1928) and Bartlett (1932), although the theory was largely ignored for several decades during the Behaviourist era. The relevance of schemas to problem solving
34
John Sweller
was emphasised by Larkin, McDermott, Simon, and Simon (1980) and Chi, Glaser, and Rees (1982), who provided theory and data indicating that the possession of domain-specific schemas differentiated novices from experts in a particular area. The extent to which one is skilful in an area depends on the number and sophistication of one’s schemas stored in long-term memory. Schema construction, by indicating the form in which information is stored in long-term memory, provides us with a learning mechanism; therefore, learning includes the construction of schemas. Learning also includes the automation of schemas. Automation has occurred when knowledge is processed unconsciously (Schneider & Shiffrin, 1977; Shiffrin & Schneider, 1977) rather than consciously in working memory. Problem solving using automated knowledge is much easier and more rapid than when basic processes must be consciously considered (Kotovsky, Hayes, & Simon, 1985). The automation of lower level schemas is frequently essential for the construction of higher level schemas. For example, without the automatic processing of the letters of the alphabet, because of the automation of schemas associated with recognising those letters, it would be difficult to combine those letters into words and sentences to permit reading. A well-known study of Bartlett (1932), in providing a graphic example of the process of schema acquisition, also indicates how the borrowing and reorganising principle functions. One person read several paragraphs of a Native American legend and then wrote down from memory as much as possible of the passage. That remembered passage was then given to another person who again wrote down as much as possible from memory, with the process repeated with ten people. There were many alterations to the passage as it passed from person to person, and those alterations provided a window into memory. The alterations were not random, with Bartlett identifying two major categories. First, there was a levelling or flattening of aspects of the passage that were novel to the participants, resulting in a reduced emphasis or disappearance of these aspects entirely. Second, there was a sharpening of those aspects that accorded with knowledge in long-term memory, resulting in those aspects being emphasised. Thus, participants did not remember the passage as it was presented but rather, remembered a construction that consisted of a combination of the passage and previous information held in long-term memory. What we hold in memory consists of schematised constructions – schemas. While being constructed, the information held by schemas is essentially borrowed, via the borrowing and reorganising principle, from schemas held by others.
Cognitive Load Theory: Recent Theoretical Advances
35
The various cognitive load effects shown in Table 2.1 provide strong evidence for the borrowing and reorganising principle. Each of the CLT effects listed in the table is concerned with techniques for presenting information to learners rather than having them generate information. The narrow limits of change principle, discussed later in the chapter, indicates why generating information is ineffective; therefore, since its inception, CLT has been concerned with techniques for presenting information to learners rather than having learners attempt to generate information. All cognitive load effects are intended to indicate how to provide auditory and visual information in a manner that best facilitates learning. In other words, CLT is concerned with how information held in the long-term memory of instructors can be borrowed for use by learners via schema acquisition. Based on the borrowing and reorganising principle, learned information can be maintained by transmitting it among people indefinitely. But the transmission is rarely exact. It normally includes a constructive element that, over time, can result in substantial changes to the store. There is an inevitable random component to this process, and whether any changes are retained or jettisoned depends on their adaptive value, with beneficial changes retained and non-beneficial changes jettisoned. Adaptive value also is critical to the next principle discussed. Problem Solving and the Randomness as Genesis Principle Despite its constructive elements, the borrowing and reorganising principle is basically a device for combining and communicating information. It does not generate new information, which begs the question, how is the information that is transmitted via the borrowing and reorganising principle created in the first instance? A likely answer seems to be random generation followed by tests of effectiveness. Consider a person dealing with a novel set of circumstances, for example, solving a problem. Most of the activity of that person will be based on knowledge held in long-term memory acquired through the borrowing and reorganising principle. Nevertheless, on many occasions, a problem solver will be faced with two or more possible moves and have no knowledge indicating which move should be made. Under these circumstances, random generation of novel problem solving moves followed by tests of effectiveness are needed. Random generation could be expected to lead to many deadends that will only be discovered after the event and, of course, when faced with a difficult novel problem, most problem solvers will, indeed, reach
36
John Sweller
many dead-ends. A difficult problem may result in many more dead-ends than appropriate moves. The relation between knowledge-based move generation and randomly generated moves can be described in terms of a central executive (Sweller, 2003). A central executive must know what the consequences of a move are prior to it occurring and then arrange for it to occur. An executive system such as this is possible for familiar moves generated by knowledge in longterm memory but impossible for novel moves. The moves made to solve a problem can originate from long-term memory but only to the extent that information is available in long-term memory indicating potential solution moves. Knowledge held in long-term memory can indicate what type of problem we are faced with and what types of moves are appropriate for that problem. That knowledge is acquired from previous experience – it is learned. Such knowledge provides the characteristics expected of a central executive and, in that sense, it is a learned central executive. If information concerning potential solution moves is not available in long-term memory, the problem solver can select a move randomly and test the effectiveness of that move. Failing knowledge held in long-term memory, there may be no logical alternative. The potential effectiveness of a move cannot be determined prior to its selection, so random selection is needed. Prior knowledge of either the correct move or knowledge of the potential usefulness of a range of moves that permits a hierarchy of moves to be established eliminates the need for random selection. Failing that knowledge held in long-term memory, random selection can be used as a substitute. It should be noted that for the reasons outlined here, computational models of human problem solving require random generation of moves as a last resort when knowledge is unavailable (e.g., see Sweller, 1988). Of course, if knowledge is available to us, we are highly likely to use it. On this argument, human cognition does include an executive but it is a learned executive held in the information store. That information acts as an executive indicating what should be done, when it should be done, and how it should be done. To the extent that this executive is unavailable because of a lack of relevant information held by an information store, random generation followed by tests of effectiveness are necessary procedures to alter the store. Because the procedures are required and because there is no substitute for them unless another store can be found from which the relevant information can be communicated, random generation followed by effectiveness testing provides the genesis of all information held by longterm memory. Indeed, it can be argued that it is the source of all human creativity (Sweller, 2009).
Cognitive Load Theory: Recent Theoretical Advances
37
Novice Working Memory and the Narrow Limits of Change Principle A major consequence that flows from the randomness as genesis principle is that all random alterations to the information store must be incremental and slow. A large, rapid, random change to long-term memory is unlikely to be adaptive in the sense that it is likely to destroy critical aspects of the store. Small, incremental changes will leave most of the store intact and functioning and are therefore unlikely to be fatal. The larger the change, the larger the probability that previous structures that have been established as effective have been compromised. Furthermore, small changes can be reasonably tested for effectiveness. Assume a small working memory that must deal with three elements that must be combined in some manner. Assume further that the various combinations are tested using the logic of permutations. There are 4! = 4 × 3 × 2 × 1 = 24 possible permutations. It is within the bounds of possibility to test 24 permutations. In contrast, assume a somewhat larger working memory that can handle ten elements. This working memory must test 10! = 10 × 9 × 8 × 7 × 6 × 5 × 4 × 3 × 2 × 1 = 3,628,800 permutations, which in most situations is unlikely to be possible. Thus, working memory is the structure that ensures that alterations to long-term memory are limited, with working memory unable to handle large amounts of novel information. We are unable to hold more than about seven items of novel information in working memory (Miller, 1956) and can probably process no more than about four items (Cowen, 2001). The narrow limits of change principle provides a central plank of CLT. Competence derives from a large store of information held in long-term memory and largely borrowed from the long-term memories of others. The narrow limits of change principle suggests that that information must be carefully structured to ensure working memory is not overloaded and that schemas are effectively constructed and transferred to long-term memory. The CLT effects listed in Table 2.1 are intended to meet this objective. The narrow limits of change principle indicates why CLT, with its emphasis on a limited working memory, argued that encouraging learners to generate information that could readily be presented to them was likely to be an ineffective instructional technique. A working memory that can process no more than two to four elements of information is unlikely to be capable of rapidly generating the knowledge required. That knowledge is available via the borrowing and reorganising principle. Each of the effects listed in Table 2.1 is based on the assumption that instruction should emphasise the borrowing and reorganising of knowledge from knowledgeable instructors
38
John Sweller
and should be structured in a manner that reduces unnecessary cognitive load. According to CLT, the more that is borrowed and the less that learners need to generate themselves, the more effective the instruction is likely to be (Kirschner, Sweller, & Clark, 2006). It must be emphasised that the limitations imposed by the narrow limits of change principle only apply to novices who, by definition, are dealing with information with which they are unfamiliar. As indicated next, information that has been organised by schemas in long-term memory does not suffer these limitations. Expert Working Memory and the Environment Organising and Linking Principle In contrast to working memory limitations when dealing with novel information, there are no known limits to the amount of information that working memory can process if it has been organised and tested for effectiveness, that is, if it comes from long-term memory. The environment organising and linking principle explains how we are able to transfer massive amounts of organised information from long-term to working memory to effect the complex actions required of the human cognition. To account for the huge amount of organised information from long-term memory that can be handled by working memory compared with the small amount of novel information that can be handled, Ericsson and Kintsch (1995) suggested a new construct, long-term working memory. Unlike short-term working memory, long-term working memory allows the rapid processing of large amounts of information providing that information has previously been organised in long-term memory. The environment organising and linking principle provides the ultimate justification for human cognition: the ability to function in a complex external environment. Experts can transfer large amounts of organised, schematic information from long-term to working memory to perform appropriately in their environment. CLT assumes that novice–expert differences (see next chapter) primarily result from differences in schematic information held in long-term memory that can be transferred as a single entity to working memory to generate actions appropriate to an environment. Furthermore, the environment organising and linking principle requires the preceding principles. According to the environment organising and linking principle, we need a large information store and learning mechanisms to build that store. The learning mechanisms are the borrowing and reorganising and randomness as genesis principles. In turn, the narrow limits of change principle permits the learning principles to function
Cognitive Load Theory: Recent Theoretical Advances
39
without destroying the information store. Once information is stored in the information store, the environment organising and linking principle allows that information to be used to guide appropriate action.
instructional implications These five principles provide a base for human cognitive architecture. Together, they result in a self-perpetuating, integrated, information processing system. As indicated earlier, it is a natural information processing system that is also used during biological evolution (Sweller, 2003, 2004; Sweller & Sweller, 2006), thus effecting a necessary connection between cognition and the biological world. The elimination of any one of the five principles will eliminate the functionality of the system. Its ability to generate action and accommodate to changing circumstances to continue to generate action requires all five principles. The centrality of the five principles to human cognitive architecture mirrors their centrality to CLT and to instructional design. Each principle has instructional design consequences. From the information store principle, we know that the major function of instruction is to alter the information held in long-term memory. Based on the information store principle, learning can be defined as a change in long-term memory. According to this definition, we have no grounds for assuming the effectiveness of proposed instructional techniques that cannot specify the changes in long-term memory that follow from use of the techniques (Kirschner et al., 2006). In contrast, the accumulation of knowledge in long-term memory is central to CLT. The borrowing and reorganising principle suggests that the bulk of human knowledge is learned from others rather than discovered by problem solving or a similar process. We acquire some of the knowledge held in the long-term memory of other people by copying what they do, listening to what they say, or reading what they have written. We must engage in the difficult task of constructing knowledge in our own long-term memories by these processes of copying, listening, or reading (Kirschner et al., 2006; Mayer, 2004). Accordingly, CLT emphasizes ways of communicating knowledge in educational contexts through observing, listening, and reading, including reading diagrams. Discovery occurs over generations, with its guiding assumption being the randomness as genesis principle. The random components of true discovery ensure that the process may be difficult or impossible to teach. In contrast, teaching can assist in the accumulation of knowledge held in longterm memory. What instructional designs are likely to assist in the goal
40
John Sweller
of changing the information held in long-term memory? Those designs that take into account the narrow limits of change principle. In human cognition, that assumption is expressed through the limitations of working memory when dealing with novel information. In contrast, there are not only no limits to the amount of organised information that can be held in long-term memory, there are no known limits to the amount of organised information from long-term memory that can be used by working memory. This alteration in the characteristics of working memory from a limited capacity, limited duration structure when dealing with novel information to an unlimited capacity, unlimited duration structure when dealing with familiar, organised information from long-term memory is expressed through the environment organising and linking principle and is central to CLT. The primary purpose of CLT has been to indicate how to present novel information structured according to the narrow limits of change principle to reduce unnecessary working memory load and facilitate change in long-term memory. In turn, changes in long-term memory permit complex actions through the environment organising and linking principle. There are three categories of cognitive load that affect working memory and these will be discussed next.
categories of cognitive load Whereas the function of instruction is to increase schematic knowledge in long-term memory, novel information must first be processed by working memory, and when dealing with novel information, working memory is limited in capacity (e.g., Miller, 1956) and duration (e.g. Peterson and Peterson, 1959). All instructional material imposes a working memory or cognitive load, and that cognitive load can be divided into two independent categories – intrinsic and extraneous – with a third category, germane cognitive load, dependent on intrinsic cognitive load. Intrinsic and extraneous cognitive load are additive, and together, they determine the total cognitive load. If that cognitive load exceeds working memory capacity, information processing, including learning, will be compromised. In other words, if total working memory load is excessive, the probability of useful changes to long-term memory is reduced. Each of the three categories of cognitive load will be discussed separately before a discussion of how they interact. Intrinsic Cognitive Load Some material is intrinsically difficult to understand and learn regardless of how it is taught. The critical factor is element interactivity, which refers to
Cognitive Load Theory: Recent Theoretical Advances
41
the number of elements that must be simultaneously processed in working memory to understand and learn material under instruction. For example, assume a student is learning the symbols for chemical elements. Each element can be learned independently of every other element. The task may be difficult because there are many elements that must be learned, but it does not impose a heavy working memory load. Because the working memory load is light, the issue of “understanding” does not arise. We may have failed to learn or forgotten the symbol for a particular element, but we are not likely to use the term “understanding” in this context. In contrast, assume that the learner is learning how to balance a chemical equation or deal with an algebraic equation. The number of relevant elements may be far less than the number dealt with when learning chemical element symbols, but element interactivity is high. When learning how to solve the problem (a + b)/c = d, solve for a, one cannot just attend to one of the elements in the equation or one of the solution steps while ignoring the others if the solution is to be understood. All of the learning elements interact and unless all are considered simultaneously in working memory, the problem and its solution will not be understood. Element interactivity is high, so working memory load and intrinsic cognitive load are high. In some senses, element interactivity is fixed because it is an intrinsic property of all material that must be learned and cannot be altered. Nevertheless, this statement needs to be modified by two points. First, regardless of element interactivity, it is always possible to learn material one element at a time. In the case of high element interactivity material, that means the interacting elements are treated as though they do not interact. Learning can proceed in this manner but understanding cannot. Until all of the elements are processed in working memory, understanding will not occur. For very complex material, the learning of interacting elements as isolated elements may be unavoidable, leading to the isolated/interacting elements effect (Pollock et al., 2002; van Merri¨enboer & Sweller, 2005). By presenting learners with high element interactivity material as isolated elements and only requiring them to learn the relevant interactions later, learning is enhanced compared with requiring learners to learn the interacting elements immediately when instruction commences. Learning itself provides the second way in which the effects of high element interactivity can be reduced. Element interactivity cannot be determined merely by analysing the nature of the material that needs to be learned. Depending on the schemas that have been acquired, material that is complex for one individual may be very simple for another. If a set of interacting elements have become incorporated into a schema, only that schema needs to be processed in working memory, not the interacting
42
John Sweller
elements. Accordingly, working memory load is low. For readers of this chapter, the word “accordingly” is treated as a single element because we all have schemas for this written word. For someone who is just beginning to learn to read written English, the individual letters may need to be processed simultaneously, and that is a high element interactivity task that may exceed the capacity of working memory. Once the word is learned, individual words can be treated as a single element rather than as multiple elements. In this way, a high intrinsic cognitive load due to element interactivity is altered by learning. As can be seen, this alteration in the intrinsic cognitive load due to learning is a consequence of the alteration in the characteristics of working memory depending on whether information is organised or random. According to the environment organising and linking principle, there are no limits to the amount of organised information that can be used for action. This organised information is held in long-term memory, and large amounts of such information can be transferred to working memory. In contrast, according to the randomness as genesis and the narrow limits of change principles, there are severe limits to the extent to which the store can be changed to permit new actions. Accordingly, only a limited amount of novel information can be organised in a manner that alters long-term memory. Extraneous Cognitive Load This category of cognitive load also depends on element interactivity but unlike intrinsic cognitive load, the interacting elements are fully under instructional control, and CLT was devised primarily to provide principles for the reduction of extraneous cognitive load. Whether an instructional procedure imposes an extraneous cognitive load can be assessed by determining whether it is in accord with the cognitive principles outlined earlier. If an instructional procedure does not facilitate change to the information store (long-term memory), if the procedure attempts to alter that store by use of the randomness as genesis principle instead of the borrowing and reorganising principle or functions on the assumption that randomness as genesis can or should be taught, or if the instructional procedure ignores the narrow limits of change principle by either ignoring the limitations of working memory or proceeding on the assumption that working memory has no limitations, that instructional procedure is likely to be ineffective because it unnecessarily introduces interacting elements that should be eliminated. As an example, consider a person attempting to learn via a discovery learning
Cognitive Load Theory: Recent Theoretical Advances
43
technique. Rather than being told a scientific rule, the person is given minimal information and required to work out the rule from that information. The act of discovering a rule is highly likely to make heavy demands on working memory, thus violating the narrow limits of change principle. It depends minimally on the communication of knowledge, thus violating the borrowing principle. To the extent that knowledge is unavailable, discovery relies heavily on random generation followed by tests of effectiveness, which is an extremely slow, ineffective way of accumulating information because discovery introduces a large range of interacting elements unrelated to learning. As a consequence, there is minimal emphasis on building knowledge in the information store – long-term memory – which should be the primary goal of instruction. All discovery and problem-solving based teaching techniques follow this pattern, and as a consequence, all violate every one of the five cognitive principles outlined earlier. Based on the previously described theory, we might expect there to be no systematically organised body of research demonstrating the effectiveness of discovery-based teaching techniques and, indeed, after almost a halfcentury of protagonism, the lack of evidence for these techniques is glaring (Kirschner et al., 2006). Evidence needs to consist of randomised, controlled studies altering one variable at a time. In contrast, there is extensive evidence for the advantages of providing learners with information rather than having them discover it themselves. The worked-example effect, according to which learners provided with worked examples learn more than learners provided with the equivalent problems to solve, flows directly from the cognitive architecture described earlier. The effect indicates in the clearest possible terms the advantages of providing learners with information rather than having them discover it for themselves. There are many other CLT effects. All depend on one or more of the five cognitive principles outlined previously. Many are discussed in the chapters of this volume. All are summarised in Table 2.1. More detailed summaries may also be found in Sweller (2003, 2004). Germane Cognitive Load Reducing extraneous cognitive load would have little function if the working memory resources so freed were not used for productive learning. Instruction should be designed to ensure that the bulk of working memory resources is germane to the goal of schema acquisition and automation. In other words, working memory resources should be devoted to dealing with intrinsic cognitive load rather than extraneous cognitive load because
44
John Sweller
schema acquisition is directed to the interacting elements associated with intrinsic cognitive load. Instructional designs that increase the use of working memory resources devoted to intrinsic cognitive load have the effect of increasing germane cognitive load, which should be increased to the limits of working memory capacity. Beyond that point, increases in germane cognitive load become counterproductive and can be categorised as extraneous cognitive load. Much of the work on extraneous cognitive load assumed that as that category of cognitive load was reduced, germane cognitive load would automatically increase because learners would devote a similar effort to learning regardless of the effectiveness of the instruction. For that reason, there are relatively few germane cognitive load effects. There are some, nevertheless, and the example variability and imagination effects are summarised in Table 2.1. Interactions among Sources of Cognitive Load As indicated earlier, intrinsic and extraneous cognitive load are additive, and if they exceed available working memory capacity, learning (indeed, all information processing) will be compromised and is likely to cease. All of the cognitive load effects listed in Table 2.1 are caused by various interactions among these sources of cognitive load. Most of the effects occur because a reduction in extraneous cognitive load permits an increase in working memory resources devoted to intrinsic cognitive load, increasing germane cognitive load and enhancing learning. These effects only occur if intrinsic cognitive load is high. If intrinsic cognitive load is low, alterations in extraneous cognitive load may not matter because sufficient working memory resources are likely to be available to overcome a poor instructional design that imposes a heavy extraneous cognitive load, giving rise to the element interactivity effect (Table 2.1). Not only is the type of material critical to CLT, so is learner knowledge. Information or learner activities that are important to novices may interfere with further learning by more expert learners, giving rise to the expertise reversal effect (Table 2.1). In other words, as expertise increases, procedures that were important for novices, and therefore part of germane cognitive load, contribute to extraneous cognitive load for more expert learners. Although complex, considerable information is now available concerning these various interactions between sources of cognitive load that give rise to the various cognitive load effects. Nevertheless, interactions between
Cognitive Load Theory: Recent Theoretical Advances
45
various categories of cognitive load constitute a major, current research area, and much still remains to be done.
conclusions There is likely to be widespread agreement that instructional design requires knowledge of cognition. If we do not understand the mechanisms of learning and problem solving, our chances of designing effective instruction are likely to be minimal. The success of CLT as an instructional theory is heavily dependent on its view of human cognition. The theory does more than merely pay lip service to the organization and function of the structures that constitute human cognitive architecture. Cognitive architecture is central to the theory. Unless we have a conception of the bases of human intelligence and thought, effective instructional procedures are likely to elude us. The suggested principles that constitute human cognition provide one such possible base. That cognitive base, in turn, can inform us of the types of instructional procedures that are likely to be effective. The success of CLT in generating the instructional effects shown in Table 2.1 provides some evidence for the validity of the underlying assumptions of the theory. references Ayres, P., & Sweller, J. (2005). The split-attention principle. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 135–146). New York: Cambridge University Press. Bartlett, F. (1932). Remembering: A study in experimental and social psychology. London: Cambridge University Press. Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4, 55–81. Chi, M., Glaser, R., & Rees, E. (1982). Expertise in problem solving. In R. Sternberg (Ed.), Advances in the psychology of human intelligence (pp. 7–75). Hillsdale, NJ: Erlbaum. Clark, R., Nguyen, F., & Sweller, J. (2006). Efficiency in learning: Evidence-based guidelines to manage cognitive load. San Francisco: Pfeiffer. Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24, 87–114. De Groot, A. (1965). Thought and choice in chess. The Hague, Netherlands: Mouton. (Original work published 1946). Egan, D. E., & Schwartz B. J. (1979). Chunking in recall of symbolic drawings. Memory and Cognition, 7, 149–158. Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102, 211–245.
46
John Sweller
Geary, D. (2007). Educating the evolved mind: Conceptual foundations for an evolutionary educational psychology. In J. S. Carlson & J. R. Levin (Eds.), Psychological perspectives on contemporary educational issues (pp. 1–99). Greenwich, CT: Information Age Publishing. Geary, D. (2008). An evolutionarily informed education science. Educational Psychologist, 43, 179–195. Jeffries, R., Turner, A., Polson, P., & Atwood, M. (1981). Processes involved in designing software. In J. R. Anderson (Ed.), Cognitive skills and their acquisition (pp. 255–283). Hillsdale, NJ: Erlbaum. Kalyuga, S. (2005). Prior knowledge principle in multimedia learning. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (pp. 325–337). New York: Cambridge University Press. Kirschner, P., Sweller, J., & Clark, R. (2006). Why minimal guidance during instruction does not work: An analysis of the failure of constructivist, discovery, problembased, experiential and inquiry-based teaching. Educational Psychologist, 41, 75– 86. Kotovsky, K., Hayes, J. R., & Simon, H. A. (1985). Why are some problems hard? Evidence from Tower of Hanoi. Cognitive Psychology, 17, 248–294. Larkin, J., McDermott, J., Simon, D., & Simon, H. (1980). Models of competence in solving physics problems. Cognitive Science, 4, 317–348. Leahy, W., & Sweller, J. (2004). Cognitive load and the imagination effect. Applied Cognitive Psychology, 18, 857–875. Low, R., & Sweller, J. (2005). The modality principle. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (pp. 147–158). New York: Cambridge University Press. Mayer, R. (2004). Should there be a three-strikes rule against pure discovery learning? The case for guided methods of instruction. American Psychologist, 59, 14–19. Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97. Paas, F., Camp, G., & Rikers, R. (2001). Instructional compensation for age-related cognitive declines: Effects of goal specificity in maze learning. Journal of Educational Psychology, 93, 181–186. Paas, F., & van Merri¨enboer, J. (1994). Variability of worked examples and transfer of geometrical problem solving skills: A cognitive-load approach. Journal of Educational Psychology, 86, 122–133. Peterson, L., & Peterson, M. (1959). Short-term retention of individual verbal items. Journal of Experimental Psychology, 58, 193–198. Piaget, J. (1928). Judgement and reasoning in the child. New York: Harcourt. Pollock, E., Chandler, P., & Sweller, J. (2002). Assimilating complex information. Learning and Instruction, 12, 61–86. Renkl, A. (2005). The worked out example principle in multimedia learning. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (pp. 229–245). New York: Cambridge University Press. Schneider, W., & Shiffrin, R. (1977). Controlled and automatic human information processing: I. Detection, search and attention. Psychological Review, 84, 1–66.
Cognitive Load Theory: Recent Theoretical Advances
47
Shiffrin, R., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending, and a general theory. Psychological Review, 84, 127–190. Simon, H., & Gilmartin, K. (1973). A simulation of memory for chess positions. Cognitive Psychology, 5, 29–46. Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12, 257–285. Sweller, J. (1994). Cognitive load theory, learning difficulty and instructional design. Learning and Instruction, 4, 295–312. Sweller, J. (2003). Evolution of human cognitive architecture. In B. Ross (Ed.), The psychology of learning and motivation (Vol. 43, pp. 215–266). San Diego, CA: Academic Press. Sweller, J. (2004). Instructional design consequences of an analogy between evolution by natural selection and human cognitive architecture. Instructional Science, 32, 9–31. Sweller, J. (2005). The redundancy principle. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (pp. 159–167). New York: Cambridge University Press. Sweller, J. (2007). Evolutionary biology and educational psychology. In J. S. Carlson & J. R. Levin (Eds.), Psychological perspectives on contemporary educational issues (pp. 165–175). Greenwich, CT: Information Age Publishing. Sweller, J. (2008). Instructional implications of David Geary’s evolutionary educational psychology. Educational Psychologist, 43, 214–216. Sweller, J. (2009). Cognitive bases of human creativity. Educational Psychology Review, 21, 11–19. Sweller, J., & Sweller, S. (2006). Natural information processing systems. Evolutionary Psychology, 4, 434–458. van Merri¨enboer, J., & Sweller, J. (2005). Cognitive load theory and complex learning: Recent developments and future directions. Educational Psychology Review, 17, 147–177.
3 Schema Acquisition and Sources of Cognitive Load slava kalyuga
introduction The previous chapter outlined the general features of human cognitive architecture relevant to learning. According to this architecture, our schematic knowledge base in long-term memory represents the major critical factor influencing the way we learn new information. In the absence of a relevant knowledge base for a specific situation or task, we apply random search processes to select appropriate actions. Any modifications to our knowledge base for dealing with novel situations occur only under certain restrictive conditions on the amount of such change. Based on these general characteristics of learning within a cognitive load framework, it is possible to formulate general instructional principles that support processes of schema acquisition and enable understanding and learning. This chapter suggests a number of such Cognitive Load Theory (CLT)–generated principles for efficient instruction aimed at acquisition of an organized knowledge base: a direct initial instruction principle, an expertise principle, and a small stepsize of change principle. To substantiate these principles, it is necessary first to describe in more detail the concept of schematic knowledge structures and analyze sources of cognitive load that are irrelevant to learning processes.
learning as schema acquisition Schemas represent knowledge as stable patterns of relationships between elements describing some classes of structures that are abstracted from specific instances and used to categorize such instances. Multiple schemas can be linked together and organized into hierarchical structures. Such organized knowledge structures are a major mechanism for extracting meaning from information, acquiring and storing knowledge in long-term memory, 48
Schema Acquisition and Sources of Cognitive Load
49
circumventing limitations of working memory, increasing the strength of memory, guiding retrieval and recall of information, and providing connections to prior knowledge. A generalized example of a schema is the concept of a chunk of information that has traditionally served as a unit of measurement for memory capacity in studies of short-term memory (Miller, 1956) and expert–novice differences. For example, de Groot’s (1966) classical finding that chess masters could recall many more pieces from briefly exposed real chess positions than novices was explained by masters having larger chunks. Chase and Simon (1973) found that experts placed chess pieces on the board in groups that represented meaningful configurations. Similarly, experts in electronics could reconstruct large circuit diagrams from memory recalling them in chunks of meaningfully related components (Egan & Schwartz, 1979). Schematic knowledge structures can be empirically evaluated by using grouping and categorizing tasks (e.g., by asking learners to cluster problems based on their similarity or to categorize problems after hearing only part of the text); using problems with ambiguous material in the statements (e.g., by replacing some content words with nonsense words); or using the “text editing” technique (classifying problems in terms of whether the text of each problem provides sufficient, missing, or irrelevant information for solution; Low & Over, 1992). Cognitive science methods that are traditionally used in laboratory studies for diagnosing individual knowledge structures are based on interviews, think-aloud procedures, and different forms of retrospective reports. Some recent studies in rapid online methods of cognitive diagnosis of organized knowledge structures are based on the assumption that if schemas in long-term memory alter the characteristics of working memory, then tracing the immediate content of working memory during task performance may provide a measure of levels of acquisition of corresponding schematic knowledge structures. The idea has been practically implemented as the first-step method: learners are presented with a task for a limited time and asked to indicate rapidly their first step towards solution of the task. For learners with different levels of expertise, their first steps could involve different cognitive activities: an expert may immediately provide the final answer; whereas a novice may start attempting some random solution search. Different first-step responses would indicate different levels of acquisition of corresponding knowledge structures (Kalyuga & Sweller, 2004, 2005; Kalyuga, 2006). Studies of expert–novice differences have demonstrated that experts’ performance is determined not by superior problem-solving strategies or
50
Slava Kalyuga
better working memories but rather, a better knowledge base that includes a large interconnected set of domain-specific schematic knowledge structures, well-developed cognitive skills (automated knowledge), and metacognitive self-regulatory skills that allow experts to control their performance, assess their work, predict its results, and, generally, use the available knowledge base (Glaser, 1990). When knowledge becomes automated, conscious cognitive processing requires very limited cognitive resources, and freed capacity can be concentrated on higher, more creative levels of cognition that enable the transfer of learning (Cooper & Sweller, 1987; Kotovsky, Hayes, & Simon, 1985; Schneider & Shiffrin, 1977; Shiffrin & Schneider, 1977). Organized, schematic knowledge structures in long-term memory allow the chunking of many elements of information into a single, higher-level element (Chi, Glaser, & Rees, 1982; Larkin, McDermott, Simon, & Simon, 1980). By treating many elements of information as a single element, longterm memory schematic knowledge structures may reduce informationprocessing demands on working memory. For example, according to the concept of long-term working memory (Ericsson & Kintsch, 1995), longterm memory knowledge structures activated by elements in working memory may form a long-term working memory structure that is capable of holding huge amounts of information for extended periods of time in an interference proof way. Novice learners possess only very limited lower-level knowledge associated with surface aspects of a domain, whereas experts are capable of activating higher-level schematic structures that contain information critical to problem solutions (Chi & Glaser, 1985). These schematic knowledge structures in long-term memory effectively provide necessary executive guidance during high-level cognitive processing. Without such guidance and in the absence of external instructions, learners have to resort to random search or weak problem-solving methods aimed at a gradual reduction of differences between current and goal problem states. Such methods are cognitively inefficient, time consuming, and may impose a heavy working memory load interfering with the construction of new schemas (Sweller, 1988, 2003). As learners become more experienced in the domain, their problem perception and levels of specificity of their schemas change. These schemas become more general and indexed by the underlying principles (Chi, Feltovich, & Glaser, 1981). Using available schematic knowledge bases in long-term memory, expert learners categorize different problem states and select the most appropriate solution moves, perform with greater accuracy, and lower cognitive loads. Organization of their knowledge into large groups of chunks decreases the demands on working memory and allows
Schema Acquisition and Sources of Cognitive Load
51
expert learners to rapidly activate appropriate procedures as soon as they retrieve a problem schema. New information is always encoded in terms of existing domain-specific schemas. As learners acquire more knowledge and experience in the domain, their schemas evolve and become more refined. New schemas may also be created by modeling them on best-suited existing schemas from different task domains, followed by further gradual refinement (learning by analogy). In any case, learning always involves integrating new information with existing knowledge structures. However, learners may often activate different schemas from those intended by the instruction. Students’ existing schemas in particular domains are often quite different from those of experts or teachers. For example, some initial intuitive schemas acquired from everyday experiences (sometimes called “alternative frameworks,” “misconceptions,” “preconceptions,” or “phenomenological primitives”) might make much of new conceptual scientific information incomprehensible for novice learners (diSessa, 1993; Howard, 1987; Slotta, Chi, & Juram, 1995). Such preexisting schemas often resist change, and everything that cannot be understood within the available schematic frameworks is ignored or learned by rote. Cognitive conflicts between mismatching instruction-based cognitive models and learners’ simplified internal knowledge structures may increase processing demands on limited working memory. However, for more advanced learners who have acquired well-organized schemas in the domain, simplified and detailed instruction-based conceptual models may also conflict with the students’ more sophisticated knowledge structures and inhibit learning (Mayer, 1989). Effective instructional methods should always be tuned up to the students’ existing schematic knowledge base to minimize processing loads on working memory. Thus, from a cognitive load perspective, the major goal of learning is the acquisition and automation of schematic knowledge structures in long-term memory. Working memory limitations are of minimal concern to learners whose knowledge in a domain is well organized in long-term memory. Another way to reduce working memory-processing limitations is to practice the skills provided by schemas until they can operate under automatic rather than controlled processing, which is also achieved by acquiring higher levels of expertise in a specific task domain. Therefore, effective instructional procedures and techniques should be aimed at acquiring an organized schematic knowledge base and reducing any diversion of cognitive resources to tasks and activities that are not directly associated with this goal. Because a schematic knowledge base in long-term memory
52
Slava Kalyuga
represents the foundation of our intellectual abilities and skilled performance, the acquisition of the organized knowledge base should be regarded as a key general instructional objective that positions the design of instruction in a way that optimizes cognitive load during learning.
sources of cognitive load in instruction Cognitive load during learning is determined by the demands on working memory by cognitive activities designed to achieve specific learning goals. A major source of cognitive load is provided by activities that are necessary for learning essential connections between elements of information and building new knowledge structures in working memory. This type of load is referred to as intrinsic cognitive load (see Chapter 2, this volume) and is associated with the internal intellectual complexity of the instructional material. The magnitude of the intrinsic cognitive load experienced by a learner is determined by the degree of interactivity between individual learning elements relative to the level of learner expertise in the domain. A learning element is a highest-level chunk of information for a particular learner and specific task domain. The content of various chunks is determined by the schemas learners hold in their long-term memory base. With the development of expertise, the size of a person’s chunks increases: many interacting elements for a novice become encapsulated into a single element for an expert. When learning elements need to be processed simultaneously to comprehend the instruction (even if the number of elements is relatively small), the material is high in element interactivity and can impose a high intrinsic cognitive load. For example, learning the operation of an electronic circuit is much more difficult than learning symbolic meanings of individual elements of this circuit. All elements in the circuit might be well known to the learner, assuming that he or she has pre-acquired lower-level schemas for those components (otherwise, the number of elements will be expanded considerably). Combined in the circuit, they become interconnected and need to be considered simultaneously as a whole to understand the operation of the circuit. Once the interactions of the components of the circuit have been learned, corresponding lower-order schemas become the elements of a higher-order schema that can further act as a single element. If the learner encounters this configuration of electronic components in a new circuit, cognitive processing would require a minimal cognitive effort. When a learner actually attends to the learning elements and attempts to establish connections between them in working memory, he or she
Schema Acquisition and Sources of Cognitive Load
53
experiences intrinsic cognitive load. Without active processing of essential connections between learning elements, the element interactivity (relative to the assumed level of learner expertise) would remain a characteristic of instructional material that indicates only a potential (not actual) level of intrinsic cognitive load for a specific person. Because this load is essential for comprehending the material and constructing new higher-level schematic structures, it is vital to provide all the necessary resources to accommodate the intrinsic cognitive load without exceeding the limits of working memory capacity. The concept of germane load as cognitive load that contributes to schema construction was introduced into CLT to account for learning-relevant demands on working memory (Paas & van Merri¨enboer, 1994; Sweller, van Merri¨enboer, & Paas, 1998). Although, according to this general definition, intrinsic cognitive load should be regarded as germane load (the distinction between them is not clearly delineated yet), germane cognitive load has been traditionally associated with various additional cognitive activities intentionally designed to foster schema acquisition and automation. For example, cognitive load imposed by explicit self-explanations during learning from worked examples (Chi, Bassok, Lewis, Reimann, & Glaser, 1989) or by activities of imagining procedures described in instruction (Cooper, Tindall-Ford, Chandler, & Sweller, 2001) represent typically cited examples of germane load. According to this view, the sources of germane cognitive load are auxiliary cognitive activities designed to enhance learning outcomes or increase levels of learner motivation. Such activities would obviously increase total cognitive load; however, they directly contribute to learning. Thus, cognitive load that directly contributes to schema acquisition includes both intrinsic and germane load. In contrast to this relevant (“good,” productive, or constructive) load, extraneous (“bad,” unproductive, or non-constructive) load is associated with a diversion of cognitive resources to activities irrelevant to learning that are caused by poor instructional design. Extraneous cognitive load is imposed by the design of instruction that can take various forms (written instructions, practical demonstrations, etc.), use various modes (e.g., verbal and/or pictorial) and modalities (e.g., visual and/or auditory), require different activities from learners (solving problems, studying worked examples, exploring task domains, etc.), and involve different sequences and step-sizes of learning tasks (e.g., different arrangements of part-tasks and whole tasks). Extraneous load is associated with cognitive activities that a learner is involved in because of the way the learning tasks are organized and presented rather than because the load
54
Slava Kalyuga
is essential for achieving instructional goals (Sweller, Chandler, Tierney, & Cooper, 1990; Sweller & Chandler, 1994). For example, when some interrelated elements of instruction (textual, graphical, audio, etc.) are separated over distance or time, their integration might require intense search processes and recall of some elements until other elements are attended and processed. Such processes need additional resources and might significantly increase demands on working memory. If a diagrammatic representation of an electronic circuit is accompanied by separate textual instructional explanations, understanding these instructions requires integration of the text and diagram. This involves holding segments of text in working memory until corresponding components of the circuit’s diagram are located, attended, and processed, or keeping active some images of the diagram until corresponding fragments of the text are found, read, and processed. This search-and-match process is likely to increase extraneous cognitive load. Similarly, problem solving using search processes usually involves a large number of interacting statements in working memory (e.g., interconnected sub-goals and steps to solution). Such problem solving might require significant cognitive resources that become unavailable for learning. These cognitive demands are extraneous to the learning goal and should be considered as an extraneous cognitive load. In general, extraneous cognitive load could be imposed by one or more of the following sources: 1. An insufficient learner knowledge base that is not compensated by provided instructional guidance, thus forcing learners to search for solution steps using random procedures (instead of directly learning solution procedures from instruction). 2. An overlap of an available knowledge base with provided instructional guidance, both aimed at the same cognitive activities, thus requiring learners to establish co-references between representations of the same information (instead of using cognitive resources on constructing new representations). 3. An excessive step-size of change of knowledge base required by the instructional sequence of learning tasks that introduces too many new elements of information into working memory to be incorporated into long-term memory structures. 4. Separated (in space and/or time) related instructional representations that require learners to perform extensive search-and-match processes.
Schema Acquisition and Sources of Cognitive Load
55
According to CLT, for instruction to be efficient, intrinsic and extraneous cognitive loads together should not exceed limited working memory capacity. When learning specific instructional materials does not require high levels of intrinsic cognitive load (e.g., because it is low in element interactivity relative to the current level of learner expertise), the extraneous cognitive load imposed by poor instructional design may be of little concern because total cognitive load may not exceed working memory capacity. In contrast, when instructional material is characterized by a high degree of element interactivity relative to the learner level of expertise, it might require a heavy intrinsic cognitive load to comprehend the instruction. In such a situation, an additional extraneous cognitive load caused by an inappropriate design can leave insufficient cognitive resources for learning because total cognitive load may exceed a learner’s working memory capacity. The available cognitive resources may be inadequate for sustaining the required level of intrinsic load and any additional (germane) cognitive activities designed to enhance meaningful learning. In this situation, elimination or reduction of extraneous cognitive load by improving instructional design may be critical for learning. Therefore, a cognitively effective instructional design should provide the necessary resources for sustaining intrinsic cognitive load and reduce extraneous load as much as possible. In some situations, because of an inadequate selection of learning goals, poor sequencing or excessive step-sizes of learning tasks, or other instructional design omissions, required intrinsic load may exceed limits of working memory capacity for a given level of learner expertise. This excessive intrinsic load would cause the design-generated disruption of learning processes and effectively become a form of extraneous load. Similar transformations could also take place with any auxiliary germane cognitive activities that require cognitive resources that exceed available working memory limits. For example, when novice learners are required to explicitly respond to self-explanation prompts when studying complex material with a high degree of element interactivity, this form of learning is unlikely to be productive. The following section provides general guidelines for minimizing extraneous cognitive load by focusing on the instructional goal of building an organized knowledge base in learners’ long-term memory. These guidelines suggest providing direct access to required knowledge structures, eliminating unwarranted random search processes and avoiding diversion of cognitive resources on other irrelevant cognitive activities, and securing manageable incremental changes in the knowledge base within the narrow limits of working memory capacity.
56
Slava Kalyuga
schema acquisition and cognitive load The Direct Initial Instruction Principle As learners become more knowledgeable in a domain, their cognitive activities change. Whereas for novice learners, construction of new schemas represents the dominant cognitive activity, experts tend to use available long-term memory knowledge structures to handle situations and tasks within their domain of expertise. These long-term memory knowledge structures coordinate and control cognitive activities, thus performing an executive function in complex cognitive tasks. The theory of long-term working memory (Ericsson & Kintsch, 1995), which establishes connections between components of working memory and associated experts’ schemas, effectively describes a mechanism of executive functioning of schemas held in long-term memory. Therefore, the content of long-term working memory during complex cognitive activities is a good indicator of the levels of learner expertise in a domain. The executive role of long-term memory knowledge structures consists of determining what information learners attend to and what cognitive activities they engage in. On each specific stage of cognition, such decisions could be based on either available knowledge structures or random factors. Other alternatives inevitably lead to postulating a fixed central executive, which always leads to an infinite regress problem where a higher-level executive is required to control a lower-level one (Sweller, 2003). Schematic knowledge structures in long-term memory eliminate the problem of an infinite regress. They are a major source of knowledge able to carry out an executive function during high-level cognitive processes. Such a knowledgebased central executive is constructed for every specific task by retrieving a set of appropriate schematic knowledge structures from long-term memory and combining them to manage incoming information and cognitive activities for the task. When dealing with a novel situation or task, there might be no longterm memory knowledge available to guide learners’ cognitive activities. In this case, to make sense of this situation, learners may need to process unfamiliar information in working memory element by element by using unorganized, random-search approaches that are cognitively inefficient. For example, substantial empirical evidence has indicated that extensive practice in conventional problem solving is not an effective way of acquiring problem solving schemas (Sweller & Cooper, 1985). Search strategies
Schema Acquisition and Sources of Cognitive Load
57
used during problem solving focus attention on specific features of the problem situation to reduce the difference between current and goal problem states. Maintaining sub-goals and considering alternative solution pathways might result in working memory overload; yet, such activities are unrelated to schema acquisition and the learning of important aspects of the problem structure. Goal-free problems and worked examples are more effective means of acquiring schemas than conventional problem solving (Ayres & Sweller, 1990; Sweller, 1988). Direct instructional explanations and guidance in fully worked-out examples essentially provide a substitute for the schema-based executive at the initial stages of learning by showing the learner exactly how to handle a situation or task. In contrast, problem-solving or discovery learning techniques provide the least effective executive function for novice learners. An inadequate knowledge base or, alternatively, insufficient instructional guidance to serve as an executive function in a given situation, are major sources of extraneous cognitive load. This load is imposed primarily by the random search processes that novice learners need to engage in rather than accessing the required knowledge directly. The direct initial instruction principle ensures that an appropriate level of executive function is provided to novice learners to eliminate or reduce effects of this source of extraneous cognitive load. The Expertise Principle The relative weight of long-term memory schemas and instructional explanations in providing executive functions for a task depends on the levels of learner expertise. For novices, instruction is the only available source of guidance, whereas for experts, all required knowledge might be available in long-term memory. At intermediate stages, learners should be able to retrieve available schemas for dealing with previously learned elements of information and should be provided full instructional guidance for dealing with unlearned components. If, for some elements of information, no guidance is supplied by either providers of executive functions (a longterm memory knowledge base or direct instruction), learners need to use search strategies that may cause a high extraneous cognitive load. However, an overlap of a learner’s knowledge base and instructional guidance both serving as an executive function for the same units of information, could also impose an extraneous cognitive load. In this situation, learners require additional cognitive resources to establish co-references between
58
Slava Kalyuga
representations of the same information instead of using these resources in constructing new representations. The expertise principle ensures that executive function is tailored to the levels of learner expertise at each stage of instruction, thus eliminating or reducing the effect of this source of extraneous cognitive load (Kalyuga, 2005, 2007; Kalyuga, Ayres, Chandler, & Sweller, 2003). The expertise principle assumes continuous optimization of cognitive load by presenting required instructional guidance at the appropriate time and the removal of unnecessary redundant instructional support and activities as learners progress from novice to more advance levels of proficiency in a domain. For novice learners, direct instruction in the form of fully worked-out examples could be a cognitively optimal format because worked examples effectively assist in structuring information in working memory, thus providing a substitute for missing schemas (Renkl & Atkinson, 2003; Chapter 5, this volume). At intermediate levels of expertise, worked examples could support construction of higher-level schemas not yet available, and problem solving could be used for practicing retrieval and application of previously acquired lower-level schemas. Completion problems or fading worked examples may effectively combine these two different forms of cognitive support (Renkl, Atkinson, & Große, 2004; van Merri¨enboer, Kirschner, & Kester, 2003). For expert learners, most cognitive activities are based on activating previously acquired schematic knowledge structures to organize relatively new information in working memory. Problem solving and guided exploration could be effective instructional methods, whereas studying worked examples is likely to be a redundant and inefficient cognitive activity at this stage. Determining the optimal level of instructional guidance relative to levels of learner expertise in a domain is a difficult task: sufficient instructional details should be provided for students to understand the material, and redundant details that may overload working memory should be eliminated. For more advanced learners, an instructional format with redundant material eliminated may be superior to the format that includes the redundant material. For example, in computer-based environments, advanced learners may avoid unnecessary explanations by turning off the auditory or on-screen verbal explanations. Adaptive dynamic online instructional systems may present the same information in different ways to different learners or to the same learner at different stages of instruction. However, to monitor actual changes in learner levels of expertise and suggest optimal instructional formats, appropriate, rapid online diagnostic methods are needed.
Schema Acquisition and Sources of Cognitive Load
59
The Small Step-Size of Knowledge Change Principle The process of learning as a change in the long-term memory store is based on conscious processing of information within working memory. Previously acquired knowledge structures are the most important factor that influences learning new materials. To comprehend instruction, students need to instantiate appropriate familiar schemas that would allow them to assimilate new information to prior knowledge. In the absence of a relevant knowledge base, a novice learner has to deal with isolated pieces of information without an organizing structure. Available lower-level schemas could be used to partially interpret these isolated pieces of information. However, if an immediate instructional goal is too distant from the current level of the learner’s knowledge, and the learning task introduces too many new elements, constructing new organizing higher-order schemas would be problematic. In this situation, the excessive intrinsic load caused by too many new interacting elements of information may effectively become extraneous load with resultant negative learning effects. In other words, providing too much information too quickly, even if the information is essential, is an example of extraneous cognitive load. To reduce the required resources and keep this load within the learner’s cognitive capacity, the chain of instructional sub-goals and corresponding sequence of learning tasks could be defined in smaller step-sizes with manageable load within each step. The number of new interacting elements in working memory could also be reduced by sequencing instructional subgoals properly. For example, some learning elements could be developed to a high degree of automaticity first to free working memory capacity for the following changes in knowledge structures. In other cases, simplified knowledge structures could be presented first followed by details and practice with components. Pollock, Chandler, and Sweller (2002) demonstrated an isolated/interacting elements technique that artificially suspended interactions between elements and presented complex instructional material as isolated elements of information at initial stages of instruction. In this way, some partial rudimentary schemas are acquired first by novice learners, allowing the reduction of the resources required for subsequent learning of the original, highly interactive material. Thus, an excessive step-size of knowledge base change represents a potential source of extraneous cognitive load. Too many new elements of information that cannot be encapsulated into a smaller number of chunks based on available long-term memory schemas could overwhelm limited working memory capacity and cause cognitive overload. Limiting step-sizes of
60
Slava Kalyuga
incremental knowledge change could eliminate this source of extraneous load. Construction of new schematic knowledge structures based on learners’ prior experiences and available knowledge could be more efficient if it progresses gradually and is based on small revision steps. Any changes to the available knowledge base should involve a limited number of unfamiliar novel elements. The small step-size of knowledge change principle ensures that the knowledge base in long-term memory is altered in small increments without drastic and rapid changes that would exceed learner cognitive capacity, thus eliminating or reducing the effect of this source of extraneous cognitive load. According to the small step-size of knowledge change principle, it is important to gradually build new knowledge on top of students’ existing schemas or directly teach them appropriate schematic frameworks by relating them to something already known. Instructional analogies could be useful for establishing such links with existing knowledge. For example, Clement and Steinberg (2002) used a set of analogies that were based on learners’ actual experiences to design a sequence of learning tasks in which conceptual changes in learners’ knowledge of electrical circuits evolved in small increments. Each such change was based on concrete, mentally runnable explanatory schematic models of electrical circuits. Proceeding directly from students’ simplistic ideas about electricity to the operation of electrical circuits could easily exceed novice learners’ cognitive capacity limits and produce extraneous cognitive loads. Instead, the suggested multistep instructional sequence was based on small, one-at-a time, and manageable conceptual changes. Each step involved observations of a relatively new feature of the circuit that could not be explained by the available explanatory schema, and a small modification of this schema was initiated using analogies from learners’ prior everyday experiences (e.g., concepts of a container, pressure in a tire, or resistance to mechanical movements). An example of the appropriate management of instructional sequences is the four-component instructional design model (4C/ID) of van Merri¨enboer (1997). This model provides the methodology for analysis of the complex cognitive skills and knowledge structures required for performing these skills and development of appropriate sequences of wholetask learning situations that support gradual acquisition of these skills in a cognitively sustained manner. The procedure takes into account the limited processing capacity of working memory by gradually increasing the level of difficulty of whole tasks (van Merri¨enboer et al., 2003). According to this methodology, learning tasks for complex environments are organized
Schema Acquisition and Sources of Cognitive Load
61
in a simple-to-complex sequence of task classes, with gradually diminishing levels of support within each class (process of “scaffolding”). Sufficient supportive information is provided for general aspects of the learning tasks, and just-in-time (algorithmic) information is provided for invariant aspects of the learning tasks. Also, part-task practice is offered to train constituent skills that need to be performed at a very high level of automaticity (van Merri¨enboer, Clark, & de Croock, 2002).
conclusion Schemas as the units of knowledge representation allow us to treat elements of information in terms of larger higher-level chunks, thus reducing capacity demands on working memory and allowing efficient use of basic information processing features of our cognitive architecture. Cognitive mechanisms of schema acquisition and transfer from consciously controlled to automatic processing are the major learning mechanisms and foundations of our intellectual abilities and skilled performance. Many instructional materials and techniques may be ineffective because they ignore limitations of the human cognitive processing system and impose a heavy cognitive load. CLT assumes that a proper allocation of cognitive resources is critical to learning. To enhance schema acquisition, instructional designs should minimize learners’ involvement in cognitive activities that overburden their limited working memory capacity and cause excessive extraneous cognitive load. In general, we need to reduce the diversion of learners’ cognitive resources on activities and tasks that are not directly associated with schema acquisition. Extraneous cognitive load could result from an insufficient learner knowledge base or instructional guidance, an overlapping knowledge base and instructional guidance, excessive step-size of changes in the knowledge base, or interrelated instructional representations that are separated in space and/or time. To reduce or eliminate the negative effects of these sources of extraneous cognitive load, a set of general guidelines (principles) for designing instruction within the cognitive load framework has been suggested. These guidelines prescribe providing direct (worked example-based) instruction to novice learners, adapting instruction to changing levels of learner expertise, and exercising gradual knowledge base change. (Specific principles of reducing extraneous cognitive load generated by interrelated complementary representations separated in space and/or time are described in Chapter 9.) These instructional principles are directed not only to achieving
62
Slava Kalyuga
desired instructional effects, but accomplishing them efficiently and with optimal expenditure of cognitive resources and instructional time. references Ayres, P., & Sweller, J. (1990). Locus on difficulty in multi-stage mathematics problems. American Journal of Psychology, 103, 167–193. Baddeley, A. D. (1986). Working memory. New York: Oxford University Press. Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4, 55–81. Chi, M. T. H., Bassok, M., Lewis, M. W., Reimann, P., & Glaser, R. (1989). Selfexplanation: How students study and use examples in learning to solve problems. Cognitive Science, 13, 145–182. Chi, M. T. H., Feltovich, P., & Glaser, R. (1981). Categorisation and representation of physics problems by experts and novices. Cognitive Science, 5, 121–152. Chi, M. T. H., & Glaser, R. (1985). Problem solving ability. In R. Sternberg (Ed.), Human abilities: An information processing approach (pp. 227–250). San Francisco: Freeman. Chi, M., Glaser, R., & Rees, E. (1982). Expertise in problem solving. In R. Sternberg (Ed.), Advances in the psychology of human intelligence (pp. 7–75). Hillsdale, NJ: Erlbaum. Clement, J., & Steinberg, M. (2002). Step-wise evolution of models of electric circuits: A “learning-aloud” case study. Journal of the Learning Sciences, 11, 389– 452. Cooper, G., & Sweller, J. (1987). The effects of schema acquisition and rule automation on mathematical problem-solving transfer. Journal of Educational Psychology, 79, 347–362. Cooper, G., Tindall-Ford, S., Chandler, P., & Sweller, J. (2001). Learning by imagining procedures and concepts. Journal of Experimental Psychology: Applied, 7, 68–82. de Groot, A. D. (1966). Perception and memory versus thought: Some old ideas and recent findings. In B. Kleinmuntz (Ed.), Problem solving: Research, method, and theory (pp. 19–50). New York: Wiley. diSessa, A. A. (1993). Toward an epistemology of physics. Cognition and Instruction, 10, 105–225. Egan, D. E., & Schwartz, B. J. (1979). Chunking in recall of symbolic drawings. Memory and Cognition, 7, 149–158. Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102, 211–245. Glaser, R. (1990). The reemergence of learning theory within instructional research. American Psychologist, 45, 29–39. Howard, R. W. (1987). Concepts and schemata: Introduction. London: Cassel Educational. Kalyuga, S. (2005). Prior knowledge principle in multimedia learning. In R. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 325–337). New York: Cambridge University Press.
Schema Acquisition and Sources of Cognitive Load
63
Kalyuga, S. (2006). Rapid cognitive assessment of learners’ knowledge structures. Learning & Instruction, 16, 1–11. Kalyuga, S. (2007). Expertise reversal effect and its implications for learner-tailored instruction. Educational Psychology Review, 19, 509–539. Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). Expertise reversal effect. Educational Psychologist, 38, 23–31. Kalyuga, S., & Sweller, J. (2004). Measuring knowledge to optimize cognitive load factors during instruction. Journal of Educational Psychology, 96, 558–568. Kalyuga, S., & Sweller, J. (2005). Rapid dynamic assessment of expertise to improve the efficiency of adaptive e-learning. Educational Technology, Research and Development, 53, 83–93. Kotovsky, K., Hayes, J. R., & Simon, H. A. (1985). Why are some problems hard? Evidence from Tower of Hanoi. Cognitive Psychology, 17, 248–294. Larkin, J., McDermott, J., Simon, D., & Simon, H. (1980). Models of competence in solving physics problems. Cognitive Science, 4, 317–348. Low, R., & Over, R. (1992). Hierarchical ordering of schematic knowledge relating to area-of-rectangle problems. Journal of Educational Psychology, 84, 62–69. Mayer, R. E. (1989). Models for understanding. Review of Educational Research, 59, 43–64. Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97. Paas, F., & van Merri¨enboer, J. J. G. (1994). Variability of worked examples and transfer of geometrical problem-solving skills: A cognitive-load approach. Journal of Educational Psychology, 86, 122–133. Pollock, E., Chandler, P., & Sweller, J. (2002). Assimilating complex information. Learning and Instruction, 12, 61–86. Renkl, A., & Atkinson, R. K. (2003). Structuring the transition from example study to problem solving in cognitive skills acquisition: A cognitive load perspective. Educational Psychologist, 38, 15–22. Renkl, A., Atkinson, R. K., & Große, C. S. (2004). How fading worked solution steps works – a cognitive load perspective. Instructional Science, 32, 59–82. Schneider W., & Shiffrin, R. (1977). Controlled and automatic human information processing: I. Detection, search and attention. Psychological Review, 84, 1–66. Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending and a general theory. Psychological Review, 84, 127–190. Slotta, J. D., Chi, M. T. H., & Juram, E. (1995). Assessing students’ misclassifications of physics concepts: An ontological basis for conceptual change. Cognition and Instruction, 13, 373–400. Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12, 257–285. Sweller, J. (2003). Evolution of human cognitive architecture. In B. Ross (Ed.), The psychology of learning and motivation (Vol. 43, pp. 215–266). San Diego, CA: Academic Press. Sweller, J., & Chandler, P. (1994). Why some material is difficult to learn? Cognition and Instruction, 12, 185–233.
64
Slava Kalyuga
Sweller, J., Chandler, P., Tierney, P., & Cooper, M. (1990). Cognitive load and selective attention as factors in the structuring of technical material. Journal of Experimental Psychology: General, 119, 176–192. Sweller, J., & Cooper, G. A. (1985). The use of worked examples as a substitute for problem solving in learning algebra. Cognition and Instruction, 2, 59–89. Sweller, J., van Merri¨enboer, J., & Paas, F. (1998). Cognitive architecture and instructional design. Educational Psychology Review, 10, 251–296. van Merri¨enboer, J. J. G. (1997). Training complex cognitive skills: A four-component instructional design model for technical training. Englewood Cliffs, NJ: Educational Technology Publications. van Merri¨enboer, J. J. G., Clark, R. E., & de Croock, M. B. M. (2002). Blueprints for complex learning: The 4C/ID∗ model. Educational Technology Research and Development, 50, 39–64. van Merri¨enboer, J. J. G., Kirschner, P., & Kester, L. (2003). Taking the load off a learner’s mind: Instructional design for complex learning. Educational Psychologist, 38, 5–13.
4 Individual Differences and Cognitive Load Theory jan l. plass, slava kalyuga, and detlev leutner
The previous chapters discussed sources of cognitive load that are a result of the difficulty of the materials, the design of instruction, and the amount of mental effort invested by learners to process the new information. As outlined in these chapters, the major cause of cognitive load effects is the limited capacity of working memory. In this chapter, we discuss how individual differences relate to the level of cognitive load that a particular learner experiences. Individual differences in learner characteristics take many different forms, ranging from preferences for learning from different presentation formats (e.g., verbal, pictorial) or modalities (auditory, visual, haptic) and preferences for learning under different environmental conditions (e.g., lighting, noise level, or physical position) to cognitive styles (e.g., field dependency/independency), cognitive abilities (e.g., verbal, spatial ability), and intelligence (Carroll, 1993; Jonassen & Grabowski, 1993). The influence of individual differences on learning has been studied for several decades as aptitude-treatment interactions (ATIs; Cronbach & Snow, 1977; Leutner, 1992; Lohman, 1986; Mayer, Stiehl, & Greeno, 1975; Plass, Chun, Mayer, & Leutner, 1998; Shute, 1992; Snow, 1989, 1994; Snow & Lohman, 1984, 1989). Aptitude-treatment interactions occur when different instructional treatment conditions result in differential learning outcomes depending on student aptitudes, in other words, when the effect of a given treatment is moderated by a given aptitude. Different aptitudes may influence learning in specific instructional environments, and the impact of a particular aptitude on a particular condition may only be observed for a particular type of learning outcome. For example, Plass et al. (1998) found that learners with visualizer versus verbalizer learning preferences used multimedia links in a reading environment for second-language acquisition differently, resulting in different learning outcomes for text comprehension but not for 65
66
Jan L. Plass, Slava Kalyuga, and Detlev Leutner table 4.1. Categories of individual differences in learning
Information gathering Learning styles Learning preferences Personality types
Information processing
Regulation of processing
Cognitive controls Cognitive abilities Prior knowledge
Motivation Metacognition/self-regulation
vocabulary acquisition. The goal of this chapter is to focus on learner characteristics that are likely to affect the amount of available working memory and that, therefore, are expected to influence cognitive load during learning. It should be noted, however, that few studies exist that measured both individual differences and cognitive load. Which types of individual differences might be expected to have a significant effect on cognitive load that is of practical relevance? For the purpose of this chapter, we use a typology of individual differences that distinguishes between differences in information gathering, information processing, and regulation of processing (Table 4.1). Individual differences related to information gathering include learning styles, learning preferences, and personality types. This type of individual differences is characterized as value-neutral, that is, as indicators of typical performance that are not linked to outcomes in a directional sense (Jonassen & Grabowski, 1993). For example, the visual versus verbal learning style indicates a person’s preference to learn with visual (i.e., pictorial) versus verbal learning material, but does not generally predict better performance for learners with one learning style versus the other. The ATI hypothesis states that to be instructionally effective, learning environments need to match learners’ individual differences. Although confirmed under specific circumstances (Homer, Plass, & Blake, 2006; Plass, Chun, Mayer, & Leutner, 1998), to date, this hypothesis has not been empirically supported as a general principle for designing learning environments (Pashler, MacDaniel, Rohrer, & Bjork, 2009). One reason for this lack of empirical evidence may be the fact that few valid instruments exist to reliably measure individual differences variables, specifically in the field of learning styles and learning preferences (Leutner & Plass, 1998; Moreno & Plass, 2006). Individual differences in information processing include cognitive controls and cognitive abilities, including intelligence and prior knowledge, which are viewed as value directional, that is, as indicators of maximal performance and as predictors for learning success (Jonassen & Grabowski, 1993). Due to our focus on individual differences that can potentially affect
Individual Differences and Cognitive Load Theory
67
cognitive load during learning, we will examine the role of prior knowledge (Kalyuga, 2005) and spatial abilities (Mayer & Sims, 1994; Plass et al., 2003) because research has established a strong relationship between these constructs and working memory (Shah & Miyake, 1996). Differences in regulation of processing include learners’ motivation and metacognition/self-regulation. Self-regulation was found, at least in a number of studies, to be a strong predictor for learning (Graesser, McNamara, & VanLehn, 2005; Leopold, den Elzen-Rump, & Leutner, 2007; Pintrich & de Groot, 1990; White & Fredriksen, 2005; Zimmerman & Schunk, 2001) that significantly affects the level of cognitive load experienced by learners (Winne, 2001). In this chapter, we first provide a more detailed discussion of the expertise reversal effect, that is, the interaction of learners’ level of expertise and instructional design on learning outcome. Second, we extend the discussion to individual differences on spatial abilities and self-regulation. Third, we describe an adaptive approach that can be used to optimize instructional design in response to these individual differences. Finally, we outline questions and methodologies for future research on the relationship of individual learner characteristics and cognitive load.
prior knowledge (expertise reversal effect) The previous chapter described a set of general instructional principles that support processes of schema acquisition and enable learning. One of these principles, the expertise principle, reflects the primary role of learner’s organized knowledge structures (schemas) in the learning processes. Research has identified prior knowledge as one of the most important individual difference factors influencing cognitive load during learning (Kalyuga, 2005; Mayer, 2001). According to the Cognitive Load Theory (CLT), the magnitude of mental load in learning depends on the schemas that have been previously acquired by the learner. As explained in previous chapters, a learning element is a function of the level of learner expertise. What constitutes a learning element and which elements interact with each other depends on a learner’s schemas: a set of many interacting elements for one person may be a single element for another, more expert learner. Therefore, although experts in a particular domain do not possess larger working memory capacities, they experience a decreased working memory load because they have larger organized knowledge structures (or chunks of information) stored in longterm memory.
68
Jan L. Plass, Slava Kalyuga, and Detlev Leutner
In some learning scenarios, however, expertise may actually trigger additional cognitive load because experts have to process information that, given their high level of expertise in the given domain, is unnecessary for them to assure successful learning. The expertise reversal effect (see Kalyuga, Ayres, Chandler, & Sweller, 2003; Kalyuga, 2005, 2007) occurs when an instructional method that is effective for novices becomes ineffective for more knowledgeable learners (see also Lohman, 1986). Such a decline in the relative effectiveness of instruction with changes in learners’ levels of expertise can be explained within a cognitive load framework: when more experienced learners need to reconcile their available schemas with the conceptual models presented during instruction, working memory processing and storage load is likely to increase and cause an unnecessary extraneous cognitive load. Therefore, instruction that eliminates unnecessary material for a particular learner should be superior to instruction that includes such material. Detailed instructional explanations, often essential for novices to understand the learning materials, may, with increasing levels of knowledge, become unnecessary. If detailed explanations are provided for these more experienced learners, processing this information may increase cognitive load and interfere with learning instead of assisting it. For example, many diagrams require additional explanations to be comprehended by novice learners. However, if a more advanced learner has sufficient knowledge to understand a diagram by itself, any additional verbal explanations of this diagram could be unnecessary. Yet, it would be difficult to ignore text that is physically integrated into the diagram or to ignore a narrated explanation accompanying the diagram. Processing the explanation, relating it to the diagram, and, most importantly, relating it to the knowledge structures the learner has already stored in long-term memory, may result in significantly higher cognitive load than learning from the format that presents the diagram by itself (Kalyuga, 2005, 2006a; Sweller, 2005). Techniques that integrate textual explanations into diagrams, replace visual text with auditory narration, and use worked examples to increase instructional guidance, were found to be effective means of reducing working memory overload for less knowledgeable learners (see Kalyuga et al., 2003, for an overview). However, with the development of learners’ knowledge in a domain, these techniques often result in negative rather than positive or neutral effects. Subjective measures of cognitive load supported the hypothesis that processing components of instruction that were unnecessary for more knowledgeable learners increased working memory load.
Individual Differences and Cognitive Load Theory
69
In most of the original studies on the split-attention effect in learning from diagrams and text, participants were novices that did not have a sufficient schematic knowledge base (Chandler & Sweller, 1991; Sweller, Chandler, Tierney, & Cooper, 1990; Tarmizi & Sweller, 1988; Ward & Sweller, 1990). However, even at those early stages of cognitive load research, it was noticed that differences between learners in domain-specific knowledge influenced the observed effects on learning outcome. For example, Mayer and Gallini (1990) demonstrated that physically integrated parts-and-steps explanative illustrations were more effective in promoting scientific understanding (how brakes, pumps, and generators work) for low-prior-knowledge learners than for high-prior-knowledge learners. Although the results of this study did not show a complete reversal of cognitive load effects, it showed that the split-attention effect took place only for learners with low levels of expertise. Similarly, Mayer and Sims (1994) found that only novice students benefited from temporal coordination of verbal explanations with visual representations. There were no differences for high-experience learners who had already developed a sufficient long-term memory knowledge base. Longitudinal research has demonstrated that the level of learner expertise was a critical factor that influenced the occurrence of the split-attention and redundancy effects (Kalyuga, Chandler, & Sweller, 1998). Direct integration of textual explanations into diagrams was beneficial for learners with very limited experience in the domain. Such materials were easier to process and resulted in a higher level of performance. However, in subsequent stages, as the learners’ levels of expertise in the domain gradually increased, the pattern of results changed. The effectiveness of the integrated diagram-and-text condition decreased whereas the effectiveness of the diagram-alone condition increased. More experienced learners, who studied relatively new and more complex materials in the domain, reported relatively higher levels of mental load, which suggests that the text interfered with learning. The diagram-alone materials were easier to process and resulted in higher levels of performance on the subsequent tests. Similar patterns of results were obtained in other studies using different instructional formats and methods, such as dual-modality versus singlemodality presentations of text and graphics, worked-example instruction versus problem-solving practice, and worked-example instruction versus exploratory learning (Kalyuga, Chandler, & Sweller, 2000, 2001; Kalyuga, Chandler, Tuovinen, & Sweller, 2001; Tuovinen & Sweller, 1999). Additions to the original instructional text, designed to increase text coherence, were found to only benefit low-knowledge readers; high-knowledge readers benefited from using the original text only (McNamara, Kintsch, Songer, &
70
Jan L. Plass, Slava Kalyuga, and Detlev Leutner
Kintsch, 1996). Other researchers tested cognitive-load predictions regarding individual differences in learning from multiple representations. Elementary students learned how to add and subtract integers with an interactive multimedia game that included visual and symbolic representations of the procedure, with or without verbal guidance. Verbal guidance helped to minimize cognitive load only for students with low prior knowledge, low computer experience, and a less reflective cognitive style (Moreno, 2002, 2005). The previous studies concerned instructional techniques for reducing extraneous cognitive load. As mentioned in the previous chapter, when intrinsic or germane load exceeds the limits of working memory capacity for a given level of learner expertise, it could effectively become a form of extraneous load inhibiting learning processes. The expertise reversal effect was also observed with instructional methods used to manage intrinsic cognitive load. For example, to reduce the intrinsic load of some complex materials, an isolated-interactive elements instructional technique, suggested by Pollock, Chandler, and Sweller (2002), recommends first presenting separate units of information without showing the relations between them and then presenting the original material showing all the interactions. However, this instructional method did not offer any benefits to learners who already possessed basic schemas in the domain. Reduction of intrinsic load is effective only for low-knowledge learners, but not for high-knowledge learners for whom the material does not have a high degree of element interactivity. A similar expertise reversal with isolated-interactive elements effect has been demonstrated by Ayres (2005) in the domain of simple algebra transformations such as 5(3x − 4) – 2(4x − 7). A part-task, isolated-element strategy, in which the constituent elements were isolated from each other (required only a single calculation to be made), benefited learning only for students with low prior knowledge. In contrast, students with a higher level of prior knowledge learned more from whole tasks in which all elements were fully integrated (required four calculations to be completed per problem). A mixed strategy, in which students progressed from part-tasks to whole tasks, proved to be ineffective for both levels of prior knowledge. In a study on visual complexity in learning from chemistry simulations, Lee, Plass, and Homer (2006) manipulated intrinsic cognitive load of the visual display by either presenting a simulation with three variables (temperature, pressure, and volume of an ideal gas) on one screen or separating the simulation into two parts ([1] temperature and volume, and [2] pressure and volume of an ideal gas). Extraneous load was manipulated by optimizing the screen design using established cognitive load principles.
Individual Differences and Cognitive Load Theory
71
They found an expertise reversal effect for comprehension and transfer, which manifested itself in that the measures of reducing extraneous load were effective for both low- and high-prior-knowledge learners in the low intrinsic load conditions. In the high intrinsic load conditions, however, these load-reducing measures supported low-prior-knowledge learners but hindered high-prior-knowledge learners. There have also been some preliminary indications of expertise-treatment interaction effects with instructional techniques designed to enhance germane cognitive load in learning. For example, Renkl (2005) demonstrated that an instructional technique that required learners to find and fix intentionally introduced errors in worked examples to increase germane cognitive load was beneficial for high-prior-knowledge learners but not for low-prior-knowledge learners. Thus, some of the instructional design principles and techniques intended for expert learners are, as a result of the described findings, contrary to those recommended for novice learners. For example, it would be beneficial for expert learners to eliminate components of multimedia presentations that are unnecessary for them, even if the resulting format might only use a single presentation form of information (e.g., only visual diagram). Problem- and discovery-based learning environments with limited guidance could be effective for advanced learners, but would typically not be recommended for novices. Similar reversal effects are expected to be found with other cognitive load reduction methods as learners become more advanced in a domain. An expertise reversal may be expected in situations in which well-guided instructional presentations intended for novice learners are used with more advanced learners and, therefore, require an unnecessary additional expenditure of cognitive resources. If the efficiency of instructional designs depends on levels of learner prior knowledge in a domain, with learners gaining optimal benefits from different formats at different levels of expertise, a major instructional implication of the expertise reversal effect is that instructional techniques and procedures need to change as learners acquire more knowledge in a domain to minimize redundant activities at each level of expertise. According to the instructional design principles of fading (Renkl & Atkinson, 2003) and scaffolding (van Merri¨enboer, Kirschner, & Kester, 2003), which are based on CLT, novice learners should be provided with considerable instructional support that could be gradually reduced as levels of learner expertise increase. Completion tasks (van Merri¨enboer, 1990), faded worked examples (Atkinson, Derry, Renkl, & Wortham, 2000; Renkl, Atkinson, Maier, & Staley, 2002), or just varying the number and degree of details of guidelines
72
Jan L. Plass, Slava Kalyuga, and Detlev Leutner
or hints provided to students as they solve problems or explore learning environments could gradually change the levels of instructional support at intermediate and higher levels of learner expertise.
spatial abilities Spatial abilities include three basic factors related to the processes of generating, retaining, and manipulating visual images: spatial relations, the ability to mentally rotate visual images; spatial orientation, the ability to imagine how visual images might look from a different perspective; and visualization, the ability to manipulate visual patterns and identify mental images (Carroll, 1993; Lohman, 1979). According to current working memory models, spatial abilities rely heavily on working memory resources, especially on its visio-spatial sketchpad (VSSP) and executive control components (Baddeley, 1986; Miyake & Shah, 1999). Individual differences in spatial abilities are attributed to differences in spatial working memory, which is distinct from verbal working memory (Hegarty, Shah, & Miyake, 2000; Shah & Miyake, 1996). Research has also found that different spatial ability factors involve the VSSP and the executive control components of working memory to different degrees. For example, tests of spatial visualization appear to demand more involvement of executive control than tests of spatial relations, whereas the VSSP is important for all factors to maintain the visio-spatial information in memory (Hegarty & Waller, 2005). Spatial Ability and Extraneous Load Most of the studies that related individual differences and cognitive load have investigated the effect of spatial abilities under different extraneous load conditions. For example, the temporal contiguity effect, which describes learning advantages for materials with concurrent presentation of narration and animation over the successive presentation of narration and animation, was strong for students with high but not for those with low spatial abilities (Mayer & Sims, 1994). Coordinated presentation of visual and verbal explanations enhanced learning for high-spatial-ability learners and also compensated for learners’ low level of prior knowledge (Mayer & Gallini, 1990; Mayer, Steinhoff, Bower, & Mars, 1995). There is also evidence that levels of spatial abilities relate to extraneous load in a virtual learning environment. Exploring a virtual environment’s
Individual Differences and Cognitive Load Theory
73
interface requires working memory resources; therefore, high-spatial-ability students were better at exploring the interface than low-spatial-ability students. As a result, learners’ spatial abilities were highly correlated with levels of learning (Waller, 2000). Similar findings were obtained in research on audial navigation in voice-prompt systems. Untrained users were provided with four different navigation conditions: hierarchical, flexible, guided (all voice-controlled), and hierarchical (keypad-controlled; Goldstein, Bretan, Salln¨as, & Bj¨ork, 1999). The authors suggest that because of their design, the hierarchical and flexible structures offer more flexibility but require more cognitive engagement, whereas the guided condition reduced cognitive load but provided fewer options. Although no differences in the number of completed tasks, total completion time, or subjective attitudes were found across these conditions, participants who scored high on tests of spatial abilities completed their tasks more efficiently in the flexible structures than those users who obtained lower scores. Users with low spatial ability completed tasks more efficiently in the guided structures of navigation, suggesting that for these users, the guidance condition with lower cognitive load was more effective for the initial learning task, compared with high-spatial-ability users, for whom the conditions with higher cognitive load (and more flexibility) were more effective (Goldstein et al., 1999). Spatial abilities were found to differently affect various types of learning outcomes. The research by Mayer and Sims (1994), for example, found an effect of spatial abilities on transfer tasks but not on tests of retention. In research on second-language acquisition, learners read a German text with or without the following types of annotations: textual only, consisting of English translations of the selected German words; visual only, consisting of still images or video clips of the selected German words; or both text and visual annotations. These annotations were designed to aid learners’ selection of relevant information for understanding the meaning of individual vocabulary items. Learning outcomes were measured using a vocabulary and a comprehension post-test. In the vocabulary post-test, high-spatial-ability learners performed better than low-spatial-ability learners when only visual annotations were available. Low-spatial-ability learners, on the other hand, performed better than high-spatial-ability learners when no annotations were available and when both visual and verbal annotations were available. This significant interaction effect on spatial abilities and treatment conditions for the vocabulary test was not found in the comprehension test (Plass et al., 2003). These results suggest that only high-spatial-ability learners were able to focus on the main task of comprehending the text,
74
Jan L. Plass, Slava Kalyuga, and Detlev Leutner
whereas low-spatial-ability learners spent more of their cognitive resources on the low-level processing and decoding of vocabulary words and less on the comprehension of the text. Learners’ spatial abilities, and the resulting different hypothesized levels of extraneous cognitive load, may have influenced learning strategies in processing the reading text, which were differently reflected in the two outcome measures. Spatial Ability and Intrinsic Load A small number of studies included research questions that could be interpreted as providing insight into the relationship between spatial abilities and intrinsic cognitive load, even though none of them measured load directly. For example, Gyselinck, Cornoldi, Dubois, De Beni, and Ehrlich (2002) found that the beneficial effects of presenting illustrations with text disappeared when a concurrent tapping task was used to suppress visiospatial working memory. However, this pattern of results was present only in high- but not low-spatial-ability subjects. Similarly, pictorial scaffolding in a geology multimedia simulation was more beneficial for high-spatial-ability students than for low-spatial-ability students both on problem solving and transfer tests (Mayer, Mautone, & Prothero, 2002). In addition, high-spatialability students took significantly less time than low-spatial-ability students to process the learning materials. Authentic geology problems required high levels of spatial thinking, in which the pictorial-based scaffolding was particularly relevant and had a strong positive effect on high-spatial-ability students. In contrast, purely verbal scaffolding did not have a similar effect (Mayer et al., 2002). The instructional implications of these studies is that high spatial ability is typically related to better performance when instruction induces high levels of cognitive load, such as when it presents complex visio-spatial materials. Whereas learners with lower spatial ability may not be able to process such high-load materials deeply, learners with higher spatial ability have the cognitive capacity to benefit from them. However, the majority of studies examining spatial ability effects did not explicitly measure cognitive load. With the exception of those studies in which cognitive load was manipulated by design and predicted a priori (e.g., Goldstein et al., 1999), the levels of cognitive load were inferred post hoc, for example, from the analysis of study time and learning achievements. More systematic research is needed to address the relationships among spatial abilities, cognitive load, and learning outcomes and to directly measure cognitive load as well as cognitive abilities.
Individual Differences and Cognitive Load Theory
75
self-regulation skills The concept of self-regulation describes the self-directed process of monitoring and regulating one’s learning. Self-regulation is a cyclical cognitive activity that involves forethought, performance or volitional control, and reflection (Zimmerman, 1998). Evidence for the relationship between students’ self-regulation and their performance on academic tasks was found, for example, using the Motivated Strategies for Learning Questionnaire, where higher levels of reported self-regulation were associated with higher levels of academic performance (Pintrich & de Groot, 1990). Research has also shown that self-regulation strategies can be taught, and that such training can result in better learning outcomes when learning with instructional materials (see, e.g., Azevedo & Cromley, 2004, for hypermedia learning and Leopold et al., 2007, and Leutner, Leopold, & den Elzen-Rump, 2007, for learning from instructional texts). In this section, we will discuss research that can be interpreted as relating self-regulation to intrinsic load and to extraneous load, as well as research on the cognitive load impact of self-regulation scaffolds. Self-Regulation and Intrinsic Cognitive Load There is evidence that supports the notion that self-regulation is strongly related to overall cognitive load and that high cognitive load can result in failure of effective self-regulation of performance in some learners (Baumeister, Heatherton, & Tice, 1994; Vohs & Heatherton, 2000). An important determinant of learners’ self-regulation is their level of prior knowledge, which, in turn, is a determinant of intrinsic cognitive load. Experts show more metacognitive awareness and have developed better self-regulation strategies than novices (Etel¨apelto, 1993; Schoenfeld, 1987; Shaft, 1995). Learners with different levels of prior knowledge regulate their own learning by employing different learning strategies (Hmelo, Nagarajan, & Day, 2000). Variations in learners’ knowledge structures seem also to be related to differences in individual learning strategies: the higher the prior domainspecific knowledge, the deeper the learning strategy that may be preferred by the learner (Beishuizen & Stoutjesdijk, 1999). Research on expert–novice differences in performance has shown that learners’ self-regulation skills significantly influence working memory processes and the efficiency of managing cognitive resources (Moreno, 2002; Moreno & Dur´an, 2004). A study of children’s self-regulatory speech in mathematics activities, both individually in the classroom and in pairs in a laboratory setting, showed
76
Jan L. Plass, Slava Kalyuga, and Detlev Leutner
that, in the individual classroom work, high-achieving students had a statistically significantly larger frequency of regulatory speech than middle- and low-achieving students. In the laboratory setting, where children worked in pairs, these group differences disappeared, and the frequency of selfregulatory statements increased by a factor of up to five. Unlike the case of the classroom setting, where students seem to have experienced high load because of the difficulty of the assigned problems, the tasks given to students in the laboratory setting were matched to their level of prior knowledge (Biemiller, Shany, Inglis, & Meichenbaum, 1998). These results suggest that higher intrinsic load may lead to lower self-regulation activity compared with lower intrinsic load conditions. Self-Regulation and Extraneous Cognitive Load Self-regulation activities themselves can also be viewed as generating extraneous cognitive load, because the monitoring, control, and reflection activities involved in self-regulation require the investment of additional mental effort. Self-regulation demands, therefore, may result – at least for unskilled self-regulated learners – in decreased performance (Cooper & Sweller, 1987; De Bruin, Schmidt, & Rikers, 2005; Kanfer & Ackerman, 1989) and failure to engage in subsequent self-regulation (Muraven, Tice, & Baumeister, 1998). Some of the reasons for these findings were highlighted in a study by Kanfer and Ackerman (1989) that involved learning in a complex skills acquisition task (air traffic control task). Participants in one group were initially instructed to do their best to complete the task, and they received specific goals only after several trials. A second group received specific goals from the beginning. Results showed that the group who initially did not receive specific goals reported higher self-regulatory activity and outperformed the group that worked with specific goals from the beginning, suggesting that this method may have reduced extraneous cognitive load. Other studies found that positive learning outcomes depended on the types of goals given to the learners. In research on the acquisition of writing skills, Zimmerman and Kitsantas (1999) demonstrated the benefits of setting goals that facilitate self-monitoring and self-regulation: learners who were given process goals or outcome goals did not acquire writing skills as well as those who first received process goals and then shifted to outcome goals. This shifting of goals provided learners with a method to set hierarchical goals to guide their learning, a method that is suggested to lead to more independent and self-motivated learning (Bandura, 1997). Specific criteria for effective goals were identified in a study on goal setting and metacognition. This research
Individual Differences and Cognitive Load Theory
77
showed that study goals that allowed learners to derive adequate monitoring standards (e.g., specific behavioral objectives for each study session) were more effective in facilitating learning than goals that did not provide these standards (e.g., general learning goals or specific time-related goals; Morgan, 1985). Self-Regulation Scaffolds and Cognitive Load Research on the use of scaffolds to facilitate self-regulation in hypermedia learning showed that the assistance of human tutors, externally facilitating the processes of regulating students’ learning, was more effective than providing students with no scaffolding or with lists of sub-goals to guide their learning (Azevedo, Cromley, & Seibert, 2004). It also showed that metacognitive scaffolds to support low self-regulated learners can be designed in a way that does not have a negative impact on high self-regulated learners (Griffin, 2002). In Griffin’s study, scaffolds were included in an online writing course that allowed learners to reflect on the specific elements of the course that would be of value to them, allowed them to set specific learning goals for each task, and asked self-regulation questions related to the achievement of the learning goals. Results indicated that these scaffolds had little effect on low-level tasks but helped low self-regulators perform better in high-level tasks, suggesting that these scaffolds were more successful under high- than under low-load conditions. All learners receiving these scaffolds spent more time on the task than those learners who did not receive them (Griffin, 2002). Other researchers have examined the influence of learners’ reported use of self-regulated learning strategies on learning performance in learnercontrolled and program-controlled computer-based learning environments (Eom & Reiser, 2000). The Self-Regulatory Skills Measurement Questionnaire was used to measure metacognitive, cognitive, self-management, and motivational strategies prior to the study. High and low self-regulators were randomly assigned to one of the two instructional conditions: learnercontrolled and program-controlled. In the learner-controlled group, students were allowed to control the order of instructional events, whereas in the program-controlled group, the instructional sequence was predetermined. The results indicate that the performance differences between learners with high and low self-regulation skills were greater in the learnercontrolled than in the program-controlled condition. High self-regulators showed no significant differences in performance between conditions; however, low self-regulators scored higher in the program-controlled condition
78
Jan L. Plass, Slava Kalyuga, and Detlev Leutner
than in the learner-controlled condition. Low self-regulation skills might have contributed to the cognitive load experienced by learners in the learnercontrolled condition similar to the increased cognitive load experienced by low-knowledge learners in low-guidance environments (such as exploratory or discovery-based learning). In a similar study, Yang (1993) obtained a marginally significant interaction effect between levels of self-regulation and types of instructional control. High self-regulators achieved higher post-test scores in the learner-controlled condition than in the programcontrolled condition, whereas low self-regulators achieved higher scores in the program-controlled condition than in the learner-controlled one. In summary, despite the general finding that learners with higher selfregulation perform better than learners with low self-regulation, the relationship between cognitive load and self-regulation is complex and depends on several different factors that relate to both the learner and the design of the materials. Learners with higher prior knowledge usually apply deeper and more effective self-regulation strategies that use the available working memory resources more efficiently than learners with low prior knowledge. There is evidence that under high cognitive load conditions, learners use less appropriate strategies for self-regulation than under low cognitive load conditions. The cognitive processes involved in self-regulation can add to the experienced cognitive load as a function of the effectiveness of an individual’s learning strategies. However, when goals and scaffolds are well designed, this extraneous cognitive load can be reduced, and learning can be facilitated. However, many of our conclusions about self-regulation and cognitive load are interpretations that can only be inferred because none of the research on self-regulation we reviewed included a direct measure of cognitive load. Further research should therefore more systematically explore the relationships among self-regulation, cognitive load, and learning outcomes by including appropriate measures of each construct.
optimizing cognitive load in adaptive learning environments Determining the most appropriate instructional design for each individual learner is a difficult task. The decision should provide sufficient verbal and/or visual information and guidance to allow each learner to comprehend the material, yet avoid unnecessary verbal or visual information that may create extraneous cognitive overload and hinder learning. A major instructional implication of the statistical interactions found between
Individual Differences and Cognitive Load Theory
79
learner individual characteristics and learning is that instructional designs should be tailored to learners’ levels of knowledge, skills, and abilities (Leutner, 1992, 2004). To achieve the required levels of flexibility, dynamic online instructional systems might include different interactive learning modes that allow different learners to access the same information represented in different formats (Plass et al., 1998). The same instructional material may also be presented in different ways to the same individual at different stages of learning as her or his level of experience in the domain increases. For example, only selected elements of the text, graphics, and links could be displayed on the screen, and auditory explanations could be turned on or off when required by an individual learner. In such learner-adapted instructional systems, the tailoring of instructions to an individual learner can be guided by continuously assessing the person’s learning performance based on either a sophisticated computational student model, such as in intelligent tutoring systems (Anderson, Corbett, Fincham, Hoffman, & Pelletier, 1992), or using appropriate dynamic diagnostic assessment tools (Leutner, 2004). The first approach is limited to rather narrow instructional domains that need to be analyzed and described in depth on the level of elementary production rules and requires high levels of expertise in computational modeling. The second approach is more straightforward and based on repeated cycles of “test-adjust” steps. However, even this second approach requires more diagnostically powerful and rapid assessment instruments than those used in traditional educational assessment. A third approach, though pedagogically not as powerful and often not suitable for inexperienced learners, is to allow learners to make their own choices that adapt the environment to their needs. Developing suitable embedded diagnostic tools is, therefore, a major prerequisite for adapting instruction to individual learner characteristics and optimizing cognitive load. Even experienced tutors often lack sufficient diagnostic skills for adapting their level of instructional guidance to the individual needs of their students (Chi et al., 2004). As a result, instead of adapting learning tasks to student characteristics, the same uniformly prescribed “subject matter logic” is often followed (Putnam, 1987). Online learning environments usually constrain computer-mediated communication, thus making an accurate diagnosis of individual student characteristics even more difficult (N¨uckles, Wittwer, & Renkl, 2005). At the same time, these technologies offer new potentials for building adaptive learning environments based on embedded assessments of individual learners (Chang, Plass, & Homer, 2008; Leutner & Plass, 1998).
80
Jan L. Plass, Slava Kalyuga, and Detlev Leutner
For example, the empirical evidence for the expertise reversal effect described earlier indicates that instructional designs that are optimal for less knowledgeable learners might not be optimal for more advanced learners. To adapt online instructional methods to levels of learner expertise, accurate and rapid online measures of expertise are required. Because learners need to be assessed in real time during an online instructional session, traditional knowledge testing procedures may not be suitable for this purpose. Using a rapid schema-based approach to assess levels of learner expertise (Kalyuga, 2006b; Kalyuga & Sweller, 2004), recent cognitive load research demonstrated the feasibility of embedding assessment methods into online learning environments to optimize cognitive load. Two rapid diagnostic methods were investigated within this approach: the first-step method and the rapid verification method. With the first-step method, learners are presented with a task for a limited time and required to indicate rapidly their first step towards solution of the task. Depending on a person’s specific level of expertise in a domain, the first step could represent different cognitive processes. A more experienced learner may rapidly indicate a very advanced stage of the solution (or even the final answer), skipping all the intermediate solution steps, because of high levels of acquisition and automation of corresponding processes. A relatively novice learner may be able to indicate only a very immediate small change in the problem state (or start applying some random search processes). Therefore, different first-step responses would reflect different levels of expertise in a specific task area. With the rapid verification diagnostic method, learners are presented with a series of possible (correct and incorrect) steps reflecting various stages of the solution procedure for a task and are required to rapidly verify the suggested steps (e.g., by immediately clicking on-screen buttons or pressing specific keys on the computer keyboard). To successfully verify more advanced steps of a solution procedure, a learner should be able to rapidly construct and integrate more intermediate steps mentally, which is an indicator of a more advanced level of expertise. Validation studies of both methods indicated high levels of correlations between performance on these tasks and traditional measures of knowledge that required complete solutions of corresponding tasks. Test times were also reduced significantly compared with traditional test times (by up to five times in some studies). Both rapid assessment methods were combined with a measure of mental load into an integrated indicator of the efficiency of performance that was used as an online measure of expertise in adaptive learning environments (Kalyuga, 2006a; Kalyuga & Sweller, 2005).
Individual Differences and Cognitive Load Theory
81
conclusion In this chapter we argued that individual differences in learners could affect cognitive load if they influenced working memory. We then focused on the relationship of cognitive load and prior knowledge, spatial abilities, and self-regulation. However, a strong limitation of our discussion is that, with the exception of prior knowledge, the relationship among the specific individual differences, cognitive load, and learning outcomes has not yet been studied with sufficient detail, and many of our conclusions had to be inferred from indirect measures of cognitive load. Among the problems of many studies on individual differences is the way these differences are measured. The use of self-report instruments for measuring learner preferences or self-regulation seems to be a far less valid way to assess such differences than, for example, the direct observation of expressions of these differences either using log files of user behavior (Leutner & Plass, 1998) or using protocol analysis of user comments (Kalyuga, Plass, Homer, Milne, & Jordan, 2007). It should also be noted that much of the present research is based on undergraduate populations at highly selective universities, where prior knowledge, cognitive abilities, and metacognitive skills are typically high. Therefore, the participants in this research are not necessarily representative of the general population. Research needs to provide deeper insights into the effect of individual differences on working memory during learning and include reliable and valid measures for both learners’ individual differences and cognitive load during learning. Methods for measuring cognitive load are discussed in the following chapter. references Anderson, J. R., Corbett, A. T., Fincham, J. M., Hoffman, D., & Pelletier, R. (1992). General principles for an intelligent tutoring architecture. In V. Shute & W. Regian (Eds.), Cognitive approaches to automated instruction (pp. 81–106). Hillsdale, NJ: Erlbaum. Atkinson, R. K., Derry, S. J., Renkl, A., & Wortham, D. W. (2000). Learning from examples: Instructional principles from the worked example research. Review of Educational Research, 70, 181–214. Ayres, P. (2005). Impact of reducing intrinsic cognitive load on learning in a mathematical domain. Applied Cognitive Psychology, 20, 287–298. Azevedo, R., & Cromley, J. G. (2004). Does training on self-regulated learning facilitate students’ learning with hypermedia? Journal of Educational Psychology, 96(3), 523–535.
82
Jan L. Plass, Slava Kalyuga, and Detlev Leutner
Azevedo, R., Cromley, J. G., & Seibert, D. (2004). Does adaptive scaffolding facilitate students’ ability to regulate their learning with hypermedia? Contemporary Educational Psychology, 29(3), 344–370. Baddeley, A. D. (1986). Working memory. New York: Oxford University Press. Bandura, A. (1997). Self-efficacy: The exercise of control. New York: Freeman. Baumeister, R. F., Heatherton, T. F., & Tice, D. M. (1994). Losing control: How and why people fail at self-regulation. San Diego, CA: Academic Press. Beishuizen, J. J., & Stoutjesdijk, E. T. (1999). Study strategies in a computer assisted study environment. Learning and Instruction, 9, 281–301. Biemiller, A., Shany, M., Inglis, A., & Meichenbaum, D. (1998). Factors influencing children’s acquisition and demonstration of self-regulation on academic tasks. In D. H. Schunk, & B. J. Zimmerman (Eds.), Self-regulated learning: From teaching to self-reflective practice (pp. 203–224). New York: Guilford Publications. Carroll, J. (1993). Human cognitive abilities. New York: Cambridge University Press. Chandler, P., & Sweller, J. (1991). Cognitive load theory and the format of instruction. Cognition and Instruction, 8, 293–332. Chang, Y. K., Plass, J. L., & Homer, B. D. (2008). Development and Validation of a Behavioral Measure of Metacognitive Processes (BMMP). Featured Research presentation at the annual convention of the Association for Educational Communication and Technology (AECT) in October, 2008 in Orlando, FL. Chi, M. T. H., Siler, S., & Jeong, H. (2004). Can tutors monitor students’ understanding accurately? Cognition and Instruction, 22, 363–387. Cooper, G., & Sweller, J. (1987). Effects of schema acquisition and rule automation on mathematical problem-solving transfer. Journal of Educational Psychology, 79(4), 347–362. Cronbach, L. J., & Snow, R. E. (1977). Aptitudes and instructional methods: A handbook for research on interaction. New York: Irvington Publishers. de Bruin, A. B. H., Schmidt, H. G., & Rikers, R. M. J. P. (2005). The role of basic science knowledge and clinical knowledge in diagnostic reasoning: A structural equation modeling approach [Report]. Academic Medicine, 80(8), 765–773. Eom, W., & Reister, R. A. (2000). The effects of self-regulation and instructional control on performance and motivation in computer-based instruction. International Journal of Instructional Media, 27(3), 247–260. Etel¨apelto, A. (1993). Metacognition and the expertise of computer program comprehension. Scandinavian Journal of Educational Research, 37(3), 243–254. Goldstein, M., Bretan, I., Salln¨as, E.-L., & Bj¨ork, H. (1999). Navigational abilities in audial voice-controlled dialogue structures. Behaviour & Information Technology, 18(2), 83–95. Graesser, A. C., McNamara, D. S., & VanLehn, K. (2005). Scaffolding deep comprehension strategies through Point&Query, AutoTutor, and iSTART. Educational Psychologist, 40(4), 225–234. Griffin, T. (2002). Supporting students with low self-regulation through problem-based learning techniques in online education. Unpublished doctoral dissertation, New York University. Gyselinck, V., Cornoldi, C., Dubois, V., De Beni, R., & Ehrlich, M.-F. (2002). Visuospatial memory and phonological loop in learning from multimedia. Applied Cognitive Psychology, 16, 665–685.
Individual Differences and Cognitive Load Theory
83
Hegarty, M., Shah, P., & Miyake, A. (2000). Constraints on using the dual-task methodology to specify the degree of central executive involvement in cognitive tasks. Memory & Cognition, 28(3), 376–385. Hegarty, M., & Waller, D. A. (2005). Individual differences in spatial abilities. In P. Shah, & A. Miyake (Eds.), The Cambridge handbook of visuospatial thinking (pp. 121–169). New York: Cambridge University Press. Hmelo, C., Nagarajan, A., & Day, R. (2000). Effects of high and low prior knowledge on construction of a joint problem space. The Journal of Experimental Education, 69, 36–56. Homer, B. D., Plass, J. L., & Blake, L. (2006). The effects of video on cognitive load and social presence in multimedia-learning. Manuscript submitted for publication. Jonassen, D. H., & Grabowski, B. L. (1993). Handbook of individual differences, learning, and instruction. Hillsdale, NJ: Erlbaum. Kalyuga, S. (2005). Prior knowledge principle in multimedia learning. In R. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 325–337). New York: Cambridge University Press. Kalyuga, S. (2006a). Instructing and testing advanced learners: A cognitive load approach. Hauppauge, NY: Nova Science Publishers. Kalyuga, S. (2006b). Rapid cognitive assessment of learners’ knowledge structures. Learning & Instruction, 16, 1–11. Kalyuga, S. (2007). Expertise reversal effect and its implications for learner-tailored instruction. Educational Psychology Review, 19, 509–539. Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). Expertise reversal effect. Educational Psychologist, 38, 23–31. Kalyuga, S., Chandler, P., & Sweller, J. (1998). Levels of expertise and instructional design. Human Factors, 40, 1–17. Kalyuga, S., Chandler, P., & Sweller, J. (2000). Incorporating learner experience into the design of multimedia instruction. Journal of Educational Psychology, 92, 126–136. Kalyuga, S., Chandler, P., & Sweller, J. (2001). Learner experience and efficiency of instructional guidance, Educational Psychology, 21, 5–23. Kalyuga, S., Chandler, P., Tuovinen, J., & Sweller, J. (2001). When problem solving is superior to studying worked examples. Journal of Educational Psychology, 93, 579–588. Kalyuga, S., Plass, J. L., Homer, B., Milne, C., & Jordan, T. (2007). Managing cognitive load in computer-based simulations for science education. Paper presented at the UNSW Cognitive Load Theory Conference, 24–26 March 2007 in Sydney, Australia. Kalyuga, S., & Sweller, J. (2004). Measuring knowledge to optimize cognitive load factors during instruction. Journal of Educational Psychology, 96, 558– 568. Kalyuga, S., & Sweller, J. (2005). Rapid dynamic assessment of expertise to improve the efficiency of adaptive e-learning. Educational Technology Research and Development, 53, 83–93. Kanfer, R., & Ackerman, P. L. (1989). Motivation and cognitive abilities: An integrative aptitude-treatment interaction approach to skill acquisition. Journal of Applied Psychology, 74(4), 657–690.
84
Jan L. Plass, Slava Kalyuga, and Detlev Leutner
Lee, H., Plass, J. L., & Homer, B. D. (2006). Optimizing cognitive load for learning from computer-based science simulations. Journal of Educational Psychology, 89, 902–913. Leopold, C., den Elzen-Rump, V., & Leutner, D. (2007). Self-regulated learning from science texts. In M. Prenzel (Ed.), Studies on the educational quality of schools. The final report on the DFG Priority Programme (pp. 21–53). M¨unster, Germany: Waxmann. Leutner, D. (1992). Adaptive Lehrsysteme. Instruktionspsychologische Grundlagen und experimentelle Analysen [Adaptive learning systems, instructional psychology foundations and experimental analyses]. Weinheim, Germany: PVU. Leutner, D. (2004). Instructional-design principles for adaptivity in open learning environments. In N. M. Seel & S. Dijkstra (Eds.), Curriculum, plans, and processes in instructional design: International perspectives (pp. 289–308). Mahwah, NJ: Erlbaum. Leutner, D., Leopold, C., & den Elzen-Rump, V. (2007). Self-regulated learning with a text-highlighting strategy: A training experiment. Zeitschrift f¨ur Psychologie/Journal of Psychology, 215(3), 174–182. Leutner, D., & Plass, J. L. (1998). Measuring learning styles with questionnaires versus direct observation of preferential choice behavior in authentic learning situations: The Visualizer/Verbalizer Behavior Observation Scale (VV–BOS). Computers in Human Behavior, 14, 543–557. Lohman, D. F. (1979). Spatial ability: A review and reanalysis of the correlational literature (Stanford University Technical Report No. 8). Stanford, CA: Aptitudes Research Project. Lohman, D. F. (1986). Predicting mathemathanic effects in the teaching of higherorder thinking skills. Educational Psychologist, 21, 191–208. Mayer, R. E. (2001). Multimedia learning. New York: Cambridge University Press. Mayer, R., & Gallini, J. (1990). When is an illustration worth ten thousand words? Journal of Educational Psychology, 82, 715–726. Mayer, R., Mautone, P., & Prothero, W. (2002). Pictorial aids for learning by doing in a multimedia geology simulation game. Journal of Educational Psychology, 94, 171–185. Mayer, R. E., & Sims, V. K. (1994). For whom is a picture worth a thousand words? Extensions of a dual-coding theory of multimedia learning. Journal of Educational Psychology, 86, 389–401. Mayer, R., Steinhoff, K., Bower, G., & Mars, R. (1995). A generative theory of textbook design: Using annotated illustrations to foster meaningful learning of science text. Educational Technology Research and Development, 43, 31–43. Mayer, R., Stiehl, C., & Greeno, J. (1975). Acquisition of understanding and skill in relation to subjects’ preparation and meaningfulness of instruction. Journal of Educational Psychology, 67, 331–350. McNamara, D., Kintsch, E., Songer, N. B., & Kintsch, W. (1996). Are good texts always better? Interactions of text coherence, Background knowledge, and levels of understanding in learning from text. Cognition and Instruction, 14, 1–43. Miyake, A., & Shah, P. (1999). Models of working memory: Mechanisms of active maintenance and executive control. New York: Cambridge University Press.
Individual Differences and Cognitive Load Theory
85
Moreno, R. (2002). Who learns best with multiple representations? Cognitive theory predictions on individual differences in multimedia learning. World Conference on Educational Multimedia, Hypermedia and Telecommunications 2002(1), 1380– 1385. Moreno, R., & Dur´an, R. (2004). Do multiple representations need explanations? The role of verbal guidance and individual differences in multimedia mathematics learning. Journal of Educational Psychology, 96, 492–503. Moreno, R., & Plass, J. L. (2006, April). Individual differences in learning with verbal and visual representations. Paper presented at the Technology and Learning Symposium, New York. Morgan, M. (1985). Self-monitoring of attained subgoals in private study. Journal of Educational Psychology, 77(6), 623–630. Muraven, M., Tice, D. M., & Baumeister, R. F. (1998). Self-control as a limited resource: Regulatory depletion patterns. Journal of Personality and Social Psychology, 74(3), 774–789. N¨uckles, M., Wittwer, J., & Renkl, A. (2005). Information about a layperson’s knowledge supports experts in giving effective and efficient online advice to laypersons. Journal of Experimental Psychology: Applied, 11, 219–236. Pashler, H., McDaniel, M., Rohrer, D., & Bjork, R. (2009). Learning Styles: Concepts and Evidence. Psychological Science in the Public Interest, 9(3), 105–119. Pintrich, P. R., & de Groot, E. V. (1990). Motivational and self-regulated learning components of classroom academic performance. Journal of Educational Psychology, 82, 33–40. Plass, J. L., Chun, D. M., Mayer, R. E., & Leutner, D. (1998). Supporting visual and verbal learning preferences in a second-language multimedia learning environment. Journal of Educational Psychology, 90, 25–36. Plass, J. L., Chun, D. M., Mayer, R. E., & Leutner, D. (2003). Cognitive load in reading a foreign language text with multimedia aids and the influence of verbal and spatial abilities. Computers in Human Behavior, 19, 221–243. Pollock, E., Chandler, J., & Sweller, J. (2002). Assimilating complex information. Learning and Instruction, 12, 61–86. Putnam, R. T. (1987). Structuring and adjusting content for students: A study of live and simulated tutoring of addition. American Educational Research Journal, 24, 13–48. Renkl, A. (2005, August). Finding and fixing errors in worked examples: Can this foster learning outcomes? Paper presented at the 11th Biennial Conference of the European Association for Research in Learning and Instruction, Nicosia, Cyprus. Renkl, A., & Atkinson, R. K. (2003). Structuring the transition from example study to problem solving in cognitive skills acquisition: A cognitive load perspective. Educational Psychologist, 38, 15–22. Renkl, A., Atkinson, R. K., Maier, U. H., & Staley, R. (2002). From example study to problem solving: Smooth transitions help learning. Journal of Experimental Education, 70, 293–315. Schoenfeld, A. H. (1987). What’s all the fuss about metacognition? In A. H. Schoenfeld (Ed.), Cognitive science and mathematics education (pp. 189–215). Hillsdale, NJ: Erlbaum.
86
Jan L. Plass, Slava Kalyuga, and Detlev Leutner
Shaft, T. M. (1995). Helping programmers understand computer programs: The use of metacognition. Data Base Advances, 26, 25–46. Shah, P., & Miyake, A. (1996). The separability of working memory resources for spatial thinking and language processing: An individual differences approach. Journal of Experimental Psychology: General, 125(1), 4–27. Shute, V. J. (1992). Aptitude-treatment interactions and cognitive skill diagnosis. In J. W. Regian & V. J. Shute (Eds.), Cognitive approaches to automated instruction (pp. 15–47). Hillsdale, NJ: Erlbaum. Shute, V., & Towle, B. (2003). Adaptive e-learning. Educational Psychologist, 38, 105–114. Snow, R. E. (1989). Aptitude-treatment interaction as a framework for research on individual differences in learning. In P. L. Ackerman, R. J. Sternberg, & R. Glaser (Eds.), Learning and individual differences. Advances in theory and research (pp. 13–59). New York: Freeman. Snow, R. (1994). Abilities in academic tasks. In R. Sternberg & R. Wagner (Eds.), Mind in context: Interactionist perspectives on human intelligence (pp. 3–37). Cambridge, MA: Cambridge University Press. Snow, R., & Lohman, D. (1984). Toward a theory of cognitive aptitude for learning from instruction. Journal of Educational Psychology, 76, 347–376. Snow, R. E., & Lohman, D. F. (1989). Implications of cognitive psychology for educational measurement. In R. Linn (Ed.), Educational measurement (pp. 263– 331). New York: Macmillan. Sweller, J. (2005). The redundancy principle in multimedia learning. In R. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 159–167). New York: Cambridge. Sweller, J., Chandler, P., Tierney, P., & Cooper, M. (1990). Cognitive load and selective attention as factors in the structuring of technical material. Journal of Experimental Psychology: General, 119, 176–192. Tarmizi, R., & Sweller, J. (1988). Guidance during mathematical problem solving. Journal of Educational Psychology, 80, 424–436. Tuovinen, J., & Sweller, J. (1999). A comparison of cognitive load associated with discovery learning and worked examples. Journal of Educational Psychology, 91, 334–341. van Merri¨enboer, J. J. G. (1990). Strategies for programming instruction in high school: Program completion vs. program generation. Journal of Educational Computing Research, 6, 265–287. van Merri¨enboer, J. J. G., Kirschner, P. A., & Kester, L. (2003). Taking the load off a learner’s mind: Instructional design principles for complex learning. Educational Psychologist, 38, 5–13. Vohs, K. D., & Heatherton, T. F. (2000). Self-regulatory failure: A resource-depletion approach. Psychological Science, 11(3), 249–254. Waller, D. (2000). Individual differences in spatial learning from computer-simulated environments. Journal of Experimental Psychology: Applied, 6, 307–321. Ward, M., & Sweller, J. (1990). Structuring effective worked examples. Cognition and Instruction, 7, 1–39. White, B., & Frederiksen, J. (2005). A theoretical framework and approach for fostering metacognitive development. Educational Psychologist, 40(4), 211–223.
Individual Differences and Cognitive Load Theory
87
Winne, P. H. (2001). Self-regulated learning viewed from models of information processing. In B. J. Zimmerman & D. H. Schunk (Eds.), Self-regulated learning and academic achievement: Theoretical perspectives (2nd ed., pp. 153–189). Mahwah, NJ: Erlbaum. Yang, Y. C. (1993). The effects of self-regulatory learning skills and type of instructional control on learning from computer-based instruction. International Journal of Instructional Media, 20, 235–241. Zimmerman, B. J. (1998). Developing self-fulfilling cycles of academic regulation: An analysis of exemplary instructional models. In D. H. Schunk & B. J. Zimmerman (Eds.), Self-regulated learning: From teaching to self-reflective practice (pp. 1–19). New York: Guilford. Zimmerman, B. J., & Kitsantas, A. (1999). Acquiring writing revision skill: Shifting from process to outcome self-regulatory goals. Journal of Educational Psychology, 91, 241–250. Zimmerman, B. J., & Schunk, D. H. (Eds.). (2001). Self-regulated learning and academic achievement: Theoretical perspectives (2nd ed). Mahwah, NJ: Erlbaum.
part two EMPIRICAL EVIDENCE
5 Learning from Worked-Out Examples and Problem Solving alexander renkl and robert k. atkinson
One of the classic instructional effects associated with the Cognitive Load Theory (CLT) is the worked-example effect in cognitive skill acquisition (see Chapters 2 and 3, this volume; Paas & van Gog, 2006). Worked-out examples consist of a problem formulation, solution steps, and the final solution itself. They are commonplace in the instructional material pertaining to wellstructured domains such as mathematics or physics (see Figure 5.1 for an exemplary worked-out example). When CLT researchers discuss “learning from worked-out examples,” they typically mean that after the introduction of one or more domain principles (e.g., mathematical theorem, physics law), learners should be presented with several examples rather than a single example, as it is commonly the case. Despite this emphasis on learning from examples, researchers working in this area acknowledge the importance of requiring learners to solve problems later on in cognitive skill acquisition so that they can reach proficiency in the domain they are studying. In this chapter, we elaborate the theoretical assumptions and empirical findings involving the studying of worked-out examples and learning by problem solving in different phases of cognitive skill acquisition. Rather than summarizing the extensive literature on example-based learning and its implications for instructional design (for overviews, see Atkinson, Derry, Renkl, & Wortham, 2000, and Renkl, 2005), we instead focus on addressing the issues of: (a) when it is best to study worked-out solutions, (b) when it is best to solve problems, and (c) how the transition between these two learning methods should be structured. In the following sections, we discuss the effectiveness of learning from worked-out examples from a CLT perspective. Next, we describe the theoretical considerations that are relevant to the questions of when to study examples and when to move on to problem solving. Finally, we summarize 91
92
Alexander Renkl and Robert K. Atkinson
figure 5.1. A worked-out example from the domain of probability.
related experiments, propose instructional guidelines, and conclude by proposing avenues for future research.
worked-out examples: a cognitive load perspective on their effectiveness Sweller and Cooper (1985; Cooper & Sweller, 1987) performed several seminal studies that documented the worked-example effect. Across a series of experiments, they established that learning from multiple worked-out examples is more efficient (i.e., requires less learning time) and leads to better subsequent problem-solving performance on structurally identical problems (near transfer: same solution procedure) than studying a single example followed by problem solving. A caveat concerning these results was
Learning from Worked-Out Examples and Problem Solving
93
that these positive effects did not extend to far transfer performance on dissimilar problems requiring the learners to modify the example’s solution procedure to solve the novel problems. The traditional CLT explanation of the worked-example effect is relatively straightforward (see Chapter 3, this volume). In the early phases of skill acquisition, learners attempt to solve problems by using general search strategies, such as means–ends analyses. As a result, they focus their attention on specific features of the problem to reduce the difference between current and goal problem states rather than on schema-relevant principles. Moreover, the task of reducing the difference between problem states requires learners to maintain sub-goals and consider different solution options, which can result in cognitive overload. Not surprisingly, the activities connected with general search strategies do not productively contribute to the construction of problem-solving schema that would enable learners to detect relevant structural features in later problems to be solved and, on this basis, to select an appropriate solution procedure. In sum, “premature” problem solving in early phases of skill acquisition imposes a substantial amount of unnecessary (extraneous) load that does not contribute to learning. Since the publication of these seminal studies, research on learning from worked-out examples and CLT in general has flourished. Two important extensions of the worked-example effect are particularly noteworthy. First, the cognitive capacity freed up by presenting examples instead of problems to be solved is not necessarily used by all learners in a productive way (Chi, Bassok, Lewis, Reimann, & Glaser, 1989; Pirolli & Recker, 1994). In fact, most learners can be characterized as passive or superficial learners (Renkl, 1997). Thus, prompting or training the learners to use their free cognitive capacity for germane (productive) load activities is important to fully exploit the potentials of example-based learning. More specifically, learners should be prompted or trained to self-explain the presented solution steps so that they can understand their underlying rationale. Learners’ active self-explanations of worked-out examples lead not only to enhanced near transfer but also to better far transfer (dissimilar problems; e.g., Atkinson, Renkl, & Merrill, 2003; Renkl, Stark, Gruber, & Mandl, 1998). Second, the worked-example effect disappears when the learners progress through the phases of cognitive skill acquisition. For instance, if learners have high prior skill levels, then problem solving fosters learning more than studying worked-out examples. In addition, even in the course of a single learner’s cognitive skill acquisition, there should be a concerted effort to move him or her from studying worked-out examples to problem solving
94
Alexander Renkl and Robert K. Atkinson
because there is evidence that at some point, knowledge acquisition from studying worked-out examples becomes a redundant activity that contributes little or nothing to further learning. This so-called reversal of the worked-example effect is an instance of the general expertise reversal effect, as described by Kalyuga and colleagues (e.g., Kalyuga, Ayres, Chandler, & Sweller, 2003; Kalyuga, Chandler, Tuovinen, & Sweller, 2001; Chapter 4, this volume). This implies that the instructional effects introduced by CLT eventually disappear and then reverse themselves over the course of cognitive skill acquisition. In other words, studying examples is productive in the initial stages of cognitive skill acquisition but actually becomes detrimental to learning during the later phases. As learners develop a sufficient knowledge base, they are better served by engaging in independent problem solving. In line with this effect, the 4C/ID theory (e.g., van Merri¨enboer, Clark, & de Croock, 2002; van Merri¨enboer & Kester, 2005; Chapter 6, this volume) postulates that learning is best fostered by high instructional support in the beginning (e.g., by worked-out examples) that is gradually faded out (e.g., completion problems, incomplete workedout examples; see Figure 5.2) until the learners can solve problems on their own.
learning from examples and problems in the course of skill acquisition: theoretical assumptions Stages of Cognitive Skill Acquisition In the acquisition of cognitive skills, three phases are typically distinguished in psychological theories. For example, VanLehn (1996) distinguishes among early, intermediate, and late phases of skill acquisition. During the early phase, learners attempt to gain a basic understanding of the domain and its principles without necessarily striving to apply the acquired knowledge. During the intermediate phase, learners turn their attention to learning how to solve problems. Ideally, learners reflect on how abstract principles are used to solve concrete problems (e.g., by self-explanation activities). One potential outcome of this intermediate phase is that flaws or misunderstandings in the learners’ knowledge base are corrected. Finally, the learners enter the late stage, in which speed and accuracy are increased by practice. During this phase, actual problem solving rather than reflective activities, such as self-explanations, is crucial. With respect to the intermediate stage, which is the focus of this chapter, it is important to note that the construction of a sound knowledge base is not
Learning from Worked-Out Examples and Problem Solving
95
figure 5.2. An incomplete worked-out example (completion problem).
an automatic by-product of studying examples or solving problems. Rather, learners must actively self-explain the solutions of the worked-out example, that is, they must reason out the rationale of the example’s solutions (Chi et al., 1989; Neuman & Schwarz, 1998; Renkl, 1997; VanLehn, 1996). Renkl and colleagues have recently conceptualized this instructional approach as learning by self-explaining examples (e.g., Schworm & Renkl, 2007). Another important caveat about cognitive skill acquisition is that the three stages do not have precise boundaries, particularly in the case of complex cognitive skills that encompass multiple sub-components. Under these circumstances, a learner may be entering the late stage in the acquisition of one of the skill’s sub-components while simultaneously operating in the early or intermediate phase of acquiring the skill’s other sub-components.
96
Alexander Renkl and Robert K. Atkinson
Thus, a learner may be simultaneously in different stages with respect to different sub-components of a skill. Consequently, different learning activities might be germane or optimal with respect to different knowledge components. For example, when learners study the type of examples shown in Figures 5.1 and 5.2 to gain understanding in the domain of probability calculation, they may gain some understanding of the multiplication rule and its application (i.e., the principle underlying the first step in each of the examples shown in Figures 5.1 and 5.2). They may, however, still lack an understanding of the addition rule (i.e., the principle underlying the second step in each of the examples shown in Figures 5.1 and 5.2). In this case, learners will likely benefit from solving solution steps related to the multiplication rule on their own but will still need to study a worked-out solution step related to the addition rule. As previously mentioned, the fact that studying worked-out steps is more favorable in the beginning stages of cognitive skill acquisition and problem solving is superior in the later stages can be explained by the general expertise reversal effect within the framework of CLT. However, we see two drawbacks with this conceptualization. First, the term expertise is misleading because the learners who are typically participants in learning experiments do not truly gain domain expertise during the brief course of instruction (Moreno, 2006). To become a domain expert, it takes typically several years (cf. the 10-years-of-practice rule by Ericsson and colleagues; Ericsson & Charness, 1994; Ericsson, Krampe, & Tesch-R¨omer, 1993). Therefore, we prefer to describe this phenomenon as the knowledge-gain reversal effect. Second, the reversed example effect is typically ascribed to situations in which studying examples is a redundant activity that induces unnecessary (extraneous) load. However, there is no detailed account of the actual learning processes involved with this phenomenon, particularly what can account for the favorable learning outcomes from solving problems in later stages of skill acquisition. Against this background, we propose a more elaborated model of the empirically observable knowledge-gain reversal effect involved with learning from examples and problems to be solved (see Renkl, 2005). The Knowledge-Gain Reversal Effect The knowledge-gain reversal effect can be explained by the following assumptions. When learners lack an understanding of how domain principles can be instantiated for solving a certain type of problem-solving step, they typically employ shallow or general search strategies during problem solving to determine the numerical answer. For example, they can employ
Learning from Worked-Out Examples and Problem Solving
97
table 5.1. A model of the knowledge-gain reversal effect with a focus on instructional implications Understanding of how a domain principle is applied in problem solving
Typical actions during . . . . . . studying examples
. . . problem solving
Lacking understanding
Reading a worked solution step; self-explaining can be elicited by prompts productive activity leading eventually to “Given understanding”
Not principle-based, shallow, or general search strategies in problemsolving attempts unproductive activity
Given understanding
Reading a worked solution step and/or self-explaining lead to redundant elaborations unproductive activity
Principle-based problem-solving attempts; in the case of an impasse: repair of knowledge gaps; in the case of success: formation of production rules productive activity
a general search strategy such as means–ends analysis, as emphasized in the CLT account of the worked-example effect (see Chapter 3, this volume). However, they can also use shallow strategies, such as a key word strategy (i.e., selecting a procedure by a key word in the cover story of a problem; Clement & Bernhard, 2005) or a copy-and-adapt strategy (i.e., copying the solution procedure from a presumably similar problem and adapting the numbers in the procedure; VanLehn, 1998). Because of their lack of principle understanding, they cannot rely on domain strategies in their problem-solving efforts that refer to the principles to-be-learned (cf. VanLehn et al., 2005). However, employing general or shallow strategies for problem solving does not deepen domain understanding and can therefore be classified as activities inducing extraneous load (see the upper right quadrant of Table 5.1). Let us image, for example, that learners are asked to solve a probability problem, such as the one shown in Figure 5.2, but they do not understand the multiplication rule – a domain-specific rule required to solve the problem – and its application. Because the learners are not familiar with the multiplication rule, they are unable to recognize that this principle must be applied in the first solution step. In this instance, they might simply resort to revisiting a worked-out example that they had seen earlier
98
Alexander Renkl and Robert K. Atkinson
figure 5.3. A worked-out example with a self-explanation prompt.
(e.g., the example in Figure 5.1), copying the first solution step and applying the copied solution step to the current problem by updating the numerical values. Even though this approach might lead learners to solve the problem step correctly, they gain little or no new understanding of the multiplication rule and its application. In light of this possibility, we argue that it is more favorable in the beginning of skill acquisition to provide worked-out steps that require the learner to self-explain. Encouraging learners to engage in the self-explanation process can help them to gain an understanding of how the domain principles are instantiated. To overcome potentially passive learning behaviors and to assure self-explanation activities, self-explanation prompts can be used (see the upper left quadrant of Table 5.1). Figure 5.3
Learning from Worked-Out Examples and Problem Solving
99
table 5.2. Schema for a backward fading sequence from a fully worked-out example to a problem to be solved (for a solution procedure with three steps) Sequence of examples/problems Fully worked-out example Incomplete example Incomplete example Problem to be solved
Solution step 1
Solution step 2
Solution step 3
Worked Worked Worked Faded
Worked Worked Faded Faded
Worked Faded Faded Faded
shows a worked-out example with a self-explanation prompt that encourages the learner to think about the principle underlying the first solution step. It is important to note, however, that once a learner has gained an understanding of the domain principles and their instantiations, engaging in additional self-explanation activities may not support further learning (see the lower left quadrant of Table 5.1). When learners understand the relation between a type of problem-solving step and a domain principle, it is sensible to encourage them to solve a solution step requiring the application of this principle. Even in cases in which the learner’s understanding is less than perfect, for example, because of a narrowly defined understanding of when to apply the principle, a faded step could be provided (as shown in Figure 5.2’s third step). In the case of a temporary impasse, the knowledge deficit can be repaired by further reflection (self-explanation). In the case of successful problem solving, a specific production rule (Anderson & Lebiere, 1998) can be formed (see the lower right quadrant of Table 5.1). Based on this knowledge-gain reversal model, we offer the following instructional implications (see also Renkl & Atkinson, 2003, 2007). First, provide worked-out steps together with self-explanation prompts. When the learner indicates understanding (e.g., by successful self-explanations), fade the step and require the learner to solve it. More specifically, a fading method should be employed in which the worked-out steps are gradually faded from worked-out examples to problems. With a fading procedure, a completely worked example is presented initially. When the learner gains understanding, a single step can be faded, and the learner has to solve the corresponding step in the next example. After trying to solve the faded step in this incomplete example, the learner receives feedback in the form of the step’s correct solution. Then, in the following examples, the number of blanks is increased step by step until the whole problem needs to be solved. Table 5.2 shows a fading sequence for an example/problem type with three solution steps. With such a fading procedure, a smooth transition from
100
Alexander Renkl and Robert K. Atkinson
studying examples, to working on incomplete examples, to problem solving is implemented.
faded examples: empirical findings Establishing the Fading Effect We conducted a first test of the proposed fading procedure in a small-scale field study (Renkl, Atkinson, Maier, & Staley, 2002; Experiment 1). We tested whether a fading procedure was more effective than learning by exampleproblem pairs, as they are used in many studies on learning from examples (example-problem pairs consist of a fully worked-out example followed by an isomorphic problem to be solved). We compared the learning outcomes of two ninth-grade classrooms. In each classroom, a physics lesson about electricity was conducted, in which example-problem pairs or fading examples were employed, respectively. In the fading condition, the first task was to study a completely worked-out example. In the second task, students had to solve the last solution step of a problem, which was omitted. In the third task, students had to solve the last two solution steps of a problem, which were omitted. Finally, all three steps were left out for students to solve. This fading method is called backward fading of solution steps. In a post-test presented two days after the lessons, the fading classroom outperformed the example-problem pairs classroom significantly in near transfer performance, but not (significantly) on far transfer. Based on this encouraging result, we conducted two more controlled laboratory experiments to examine the efficacy of a fading procedure relative to learning by example-problem pairs. In a first laboratory experiment, psychology students worked on examples and problems from probability calculation (Renkl et al., 2002; Experiment 2). They were randomly assigned to either the fading or to the example-problem condition. In this study, we employed a forward fading procedure (omitting the first solution step first, then the second, etc.). We found that the fading procedure clearly fostered near transfer performance. This was not, however, true for far transfer performance. The effect on near transfer was mediated by the lower number of errors committed during the learning phase. We obtained converging results in the two experiments previously mentioned, even though the laboratory study and the field study differed with respect to the type of learners (school students vs. university students), the learning domain (physics/electricity vs. mathematics/probability
Learning from Worked-Out Examples and Problem Solving
101
calculation), the learning setting (school lesson vs. computer-based learning in the laboratory), and the kind of fading out worked-out solution steps (“backward” vs. “forward”). We interpreted the stability of this finding despite these very different context conditions as an indicator that the effectiveness of our fading procedure was reliable and stable. The Effects of Different Fading Procedures An open question arose from the fact that we employed two ways of fading out worked-out solution steps, a backward and forward procedure, across the two experiments. Because the context conditions in our two studies varied substantially, we could not compare the relative effectiveness of these two procedures. Such a comparison was necessary to answer the questions of whether the specific type of fading procedure significantly influences learning outcomes or whether it is of minor importance. Thus, Renkl et al. (2002; Experiment 3) implemented the condition of example-problem pairs (control) as well as the two fading procedures used in the previous experiments: forward fading and backward fading. The participants (students enrolled in educational psychology courses) were randomly assigned to one of the conditions. The positive effect of fading on near transfer was replicated. This effect was again mediated by reduced problem-solving errors during learning. In contrast to our previous studies, we found also a positive effect on far transfer. The statistically significant effect on far transfer was, however, primarily due to the backward fading condition. In addition, this type of fading procedure was more favorable compared with forward fading because it was more efficient. The learners in the backward condition spent less time on the examples without hindering their transfer performance (cf. Renkl et al., 2002; see also Renkl & Atkinson, 2003). In a subsequent study, Renkl, Atkinson, and Große (2004) examined the learning processes associated with the fading procedure to explain the observed differences between backward and forward fading performance. Perhaps the previously found difference between these fading methods might have been due to the specific order in which the domain principles were to be applied in the examples and problems that were presented for learning. In their first experiment, Renkl et al. (2004; Experiment 1) documented that the position of the faded steps did not influence learning outcomes. Instead, individuals learned most about those principles that were faded; whether a backward or forward rationale was employed did not lead to differences in learning outcomes. This finding suggested that specific self-explanation activities are triggered by faded steps.
102
Alexander Renkl and Robert K. Atkinson
In their second experiment, Renkl et al. (2004; Experiment 2) investigated this hypothesis more directly by collecting and analyzing think-aloud protocols generated by the learners during their interaction with backward faded examples or with example-problem pairs. First, in a comparison between the fading condition and the example-problem condition, we replicated a positive fading effect on near transfer and on far transfer performance. Second, we found that fading was associated with fewer unproductive learning events, which were defined as impasses that were not followed by selfexplanations. Combining Fading with Self-Explanation Prompting In an effort to optimize our fading procedure, we introduced some selfexplanation prompting at the worked-out steps in a subsequent laboratory experiment (Atkinson et al., 2003). As already mentioned, many learners do not spontaneously leverage the fact that worked-out steps afford learners sufficient cognitive capacity to generate productive self-explanations (germane load activities). The learners’ sub-optimal self-explanation activities may explain the somewhat limited effects of our fading procedure on far transfer in the previous experiments. We assumed that prompting students to self-explain the worked-out steps (not the to-be-completed steps; see also Moreno & Mayer, 2005) would render our fading procedure more effective, especially with respect to far transfer. More specifically, we again used probability examples and problems and asked the learners to determine at each worked-out step which probability rule was applied (see Figure 5.3). We compared the performance of four conditions: backward fading with and without self-explanation prompts and example-problem pairs with and without self-explanation prompts (Atkinson et al., 2003; Experiment 1). We found substantial effects of fading as well as of prompting on near transfer and on far transfer. Both effects were additive. We replicated the prompting effect in a subsequent experiment (Atkinson et al., 2003; Experiment 2). Thus, we have shown that employing instructional means to effectively use free cognitive capacity is particularly effective at fostering near transfer and far transfer performance. Are Faded Examples Superior to Well-Supported Learning by Problem-Solving? Although the previously described studies suggested that learning by selfexplaining the worked-out steps in faded examples is an effective learning
Learning from Worked-Out Examples and Problem Solving
103
method, some caveats remained. On a more general level, Koedinger and Aleven (2007) have recently argued that the effectiveness of example-based learning has been shown in relation to problem solving that included no instructional support except for corrective feedback (e.g., Sweller & Cooper, 1985). These authors argue that the worked-example effect may disappear when well-supported learning by problem solving is considered. Although Koedinger and Aleven (2007) do not reason within a CLT framework, their conjecture is compatible with CLT. This theory suggests that a high level of support is important for beginning learners, and providing worked-out examples is one way to provide such support (see Chapter 3, this volume). Hence, an empirical test of the effects of example-based learning in relation to supported problem solving is actually necessary to evaluate the generality of the worked-example effect. To date, the fading procedure’s effectiveness has been demonstrated relative to example-problem pairs. It would be informative to see whether learning from self-explaining faded examples is also superior to well-supported problem solving, as provided by teachers in the classroom or by an (intelligent) tutoring system. To address this issue, we included a faded example sequence in Cognitive Tutors (Schwonke, Renkl, Krieg, Wittwer, Aleven, & Salden, 2009), an intelligent tutoring system that has been proven to be very effective in supporting students’ learning in a variety of domains, such as mathematics and genetics (e.g., Anderson, Corbett, Koedinger, & Pelletier, 1995; Koedinger & Corbett, 2006). On the basis of a real-time assessment of the student’s learning, Cognitive Tutors provides individualized support for guided learning by problem solving. Specifically, the tutor selects appropriate problems, gives just-in-time feedback, and presents hints. Aleven and Koedinger (2002) included self-explanation prompts in Cognitive Tutors that required students to provide an explanation for each of their solution steps, by making an explicit reference to the underlying principle. This instructional method made Cognitive Tutors more effective (Aleven & Koedinger, 2002). Although Cognitive Tutors already included self-explanation prompts and many other supportive features, we were interested in examining if students’ conceptual understanding could be further improved by gradually fading worked-out examples. In addition, the empirical results on the worked-example effect also led us to the expectation that the learners would need less learning time (see Sweller & Cooper, 1985) when using an example-enriched tutor compared with the standard version. In the first experiment, there were no significant differences in the effectiveness of the standard and the example-enriched tutor versions (Schwonke
104
Alexander Renkl and Robert K. Atkinson
et al., 2009; Experiment 1). However, the example-enriched version was more efficient (i.e., students needed less learning time). A problem that was informally observed was that students had many problems in appropriately using the example-enriched tutor. As a result, in Experiment 2 (Schwonke et al., 2009), we provided students with additional instructions on how to use the tutor. In this case, the students acquired a deeper conceptual understanding when they worked with the example-enriched tutor and, as we predicted, they needed less learning time than with the standard tutor. These findings show ways in which the instructional models of tutored problem solving and example-based learning can be fruitfully combined. Adapting Fading to the Individual Learner Although there is substantial evidence to support the fading approach, we think that it can be improved. We have argued that when acquiring a complex cognitive skill, learners may be in an earlier stage with respect to one sub-component (i.e., when a principle still needs to be understood) and they may be in a later stage with respect to another sub-component (i.e., a principle is already understood). From an instructional perspective, it would be optimal to encourage a learner to study examples with self-explanations for the former sub-component while engaging them in problem solving for the latter one. However, our present fading procedure is not adaptive to an individual learner’s level of understanding of different sub-components. As it is presently structured, the problem-solving demands are gradually increased assuming a prototypical learner rather than taking into consideration the significant amount of variability among learners that we know exists (Plass et al., this volume). Kalyuga and Sweller (2004; Experiment 4) experimentally tested a realtime adaptive fading procedure when participants were learning to solve algebraic equations. To diagnose the learners’ knowledge with respect to certain steps, these authors used the rapid-assessment technique (see Chapter 3, this volume). In this technique, learners receive a (partially solved) task and are asked to indicate rapidly their next solution step. Students’ answers range from using a trial-and-error strategy to providing directly the final answer, indicating the availability of certain schemata in the domain. Kalyuga and Sweller (2004) steered the provision of worked-out or faded steps right from the beginning on the basis of the individual learner’s performance on rapid-assessment tests. The learners in the adaptive fading condition significantly outperformed their yoked counterparts with respect to knowledge gains.
Learning from Worked-Out Examples and Problem Solving
105
Salden, Aleven, Renkl, and Schwonke (2008) implemented an adaptive fading procedure using the Cognitive Tutors technology. The Cognitive Tutors assessment was based on the student’s self-explanation performance. When the self-explanation performance indicated that a student understood a principle, the solution for the steps involving that principle was faded in the next instance; thus, the learners had to solve these steps on their own. To test such an adaptation procedure, one laboratory and one classroom experiment were conducted. Both studies compared a standard Cognitive Tutors with two example-enhanced Cognitive Tutors, in which the fading of worked-out examples occurred either fixed or adaptively. Results indicate that the adaptive fading of worked-out examples leads to higher transfer performance on delayed post-tests than the other two methods.
implications for instructional design Based on the aforementioned findings, we offer the following instructional design principles that guide the use of faded worked-out examples. (1) Use a sequence of isomorphic examples. After introducing one or more domain principles, worked-out examples should be provided. Providing learners with more than one example before problem solving is the “heart” of CLT’s concept of example-based learning. (2) Elicit self-explanations. The provision of worked-out examples or steps reduces the cognitive load imposed on the learners. However, many or even most learners do not naturally take advantage of the free cognitive capacity to engage in germane load activities such as selfexplaining. To address this deficit, self-explanation prompts should be used at the worked-out steps. (3) Fade worked-out steps. After an initial completely worked-out example, incomplete examples with faded steps should be presented. As the learners gain knowledge, studying examples becomes a redundant activity. Thus, the number of blanks should be successively increased until a problem is left that has to be solved completely by the learners. In this fashion, a smooth transition from studying examples to problem solving is implemented. (4) Individualize fading. Although a non-adaptive fading procedure is effective, the effects can be further enhanced by tailoring the fading procedure to the individual trajectory of cognitive skill acquisition.
106
Alexander Renkl and Robert K. Atkinson
avenues for future research In sum, there is substantial evidence that cognitive skill acquisition can be fostered by employing the approach of learning by self-explaining faded examples. It is also important to note that not only procedural skills are fostered but also conceptual understanding and the learners’ ability to transfer knowledge to problems that require a modified solution procedure. Despite these achievements, there are still a number of open issues. One drawback of the worked-out examples investigated so far is that they typically involve well-structured domains, mostly with algorithmic solution procedures. For example, in mathematics, physics, or programming, a manageable set of solution steps can be provided that directly lead to a final answer. For activities such as cooperative learning, designing effective learning materials, scientific argumentation, finding a mathematical proof, and many other skills, solutions steps are more difficult to describe. With respect to the four previously mentioned types of skills, recent studies have shown that it is possible to extend the worked-out examples approach to illstructured, non-algorithmic domains (e.g., Hilbert, Renkl, Kessler, & Reiss, 2008: mathematical proof; Rummel, Spada, & Hauser, 2006: cooperating in a productive way; Schworm & Renkl, 2006: designing effective learning materials; Schworm & Renkl, 2007: scientific argumentation). Hence, these studies have begun to extend the applicability of the CLT concept of example-based learning to a wider range of learning domains. Nevertheless, the problem of how to structure the transition from studying examples to problem solving and how to create a fading procedure in non-algorithmic domains still needs to be addressed. references Aleven, V., & Koedinger, K. R. (2002). An effective meta-cognitive strategy: Learning by doing and explaining with a computer-based cognitive tutor. Cognitive Science, 26, 147–179. Anderson, J. R., Corbett, A. T., Koedinger, K. R., & Pelletier, R. (1995). Cognitive tutors: Lessons learned. The Journal of the Learning Sciences, 4, 167–207. Anderson, J. R., & Lebiere, C. (Eds.). (1998). The atomic components of thought. Mahwah, NJ: Erlbaum. Atkinson, R. K., Derry, S. J., Renkl, A., & Wortham, D. W. (2000). Learning from examples: Instructional principles from the worked examples research. Review of Educational Research, 70, 181–214. Atkinson, R. K., Renkl, A., & Merrill, M. M. (2003). Transitioning from studying examples to solving problems: Combining fading with prompting fosters learning. Journal of Educational Psychology, 95, 774–783.
Learning from Worked-Out Examples and Problem Solving
107
Chi, M. T. H., Bassok, M., Lewis, M. W., Reimann, P., & Glaser, R. (1989). Selfexplanations: How students study and use examples in learning to solve problems. Cognitive Science, 18, 145–182. Clement, L., & Bernhard, J. Z. (2005). A problem-solving alternative to using key words. Mathematics Teaching in the Middle School, 10, 360–365. Cooper, G., & Sweller, J. (1987). Effects of schema acquisition and rule automation on mathematical problem-solving transfer. Journal of Educational Psychology, 79, 347–362. Ericsson, K. A., & Charness, N. (1994). Expert performance: Its structure and acquisition. American Psychologist, 49, 725–747. Ericsson, K. A., Krampe, R. Th., & Tesch-R¨omer, C. (1993). The role of deliberate practice in the acquisition of expert performance. Psychological Review, 100, 363– 406. Hilbert, T. S., Renkl, A., Kessler, S., & Reiss, K. (2008). Learning to prove in geometry: Learning from heuristic examples and how it can be supported. Learning & Instruction, 18, 54–65. Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). The expertise reversal effect. Educational Psychologist, 38, 23–31. Kalyuga, S., Chandler, P., Tuovinen, J., & Sweller, J. (2001). When problem solving is superior to studying worked examples. Journal of Educational Psychology, 93, 579–588. Kalyuga, S., & Sweller, J. (2004). Measuring knowledge to optimize cognitive load factors during instruction. Journal of Educational Psychology, 96, 558–568. Koedinger, K. R., & Aleven, V. (2007). Exploring the assistance dilemma in experiments with cognitive tutors. Educational Psychology Review, 19, 239–264. Koedinger, K. R., & Corbett, A. T. (2006). Cognitive tutors: Technology bringing learning sciences to the classroom. In R. K. Sawyer (Ed.), The Cambridge handbook of the learning sciences (pp. 61–78). New York: Cambridge University Press. Moreno, R. (2006). When worked examples don’t work: Is cognitive load theory at an impasse? Learning and Instruction, 16, 170–181. Moreno, R., & Mayer, R. E. (2005). Role of guidance, reflection, and interactivity in an agent-based multimedia game. Journal of Educational Psychology, 97, 117–128. Neuman, Y., & Schwarz, B. (1998). Is self-explanation while solving problems helpful? The case of analogical problem solving. British Journal of Educational Psychology, 68, 15–24. Paas, F., & van Gog, T. (2006). Optimising worked example instruction: Different ways to increase germane cognitive load. Learning & Instruction, 16, 87–91. Pirolli, P., & Recker, M. (1994). Learning strategies and transfer in the domain of programming. Cognition & Instruction, 12, 235–275. Renkl, A. (1997). Learning from worked-out examples: A study on individual differences. Cognitive Science, 21, 1–29. Renkl, A. (2005). The worked-out-example principle in multimedia learning. In R. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 229–246). Cambridge, UK: Cambridge University Press. Renkl, A., & Atkinson, R. K. (2003). Structuring the transition from example study to problem solving in cognitive skills acquisition: A cognitive load perspective. Educational Psychologist, 38, 15–22.
108
Alexander Renkl and Robert K. Atkinson
Renkl, A., & Atkinson, R. K. (2007). An example order for cognitive skill acquisition. In F. E. Ritter, J. Nerb, E. Lehtinen, & T. M. O’Shea (Eds.), In order to learn: How the sequence of topics influences learning (pp. 95–105). New York: Oxford University Press. Renkl, A., Atkinson, R. K., & Große, C. S. (2004). How fading worked solution steps works – A cognitive load perspective. Instructional Science, 32, 59–82. Renkl, A., Atkinson, R. K., Maier, U. H., & Staley, R. (2002). From example study to problem solving: Smooth transitions help learning. Journal of Experimental Education, 70, 293–315. Renkl, A., Stark, R., Gruber, H., & Mandl, H. (1998). Learning from workedout examples: The effects of example variability and elicited self-explanations. Contemporary Educational Psychology, 23, 90–108. Rummel, N., Spada, H., & Hauser, S. (2006). Learning to collaborate in a computermediated setting: Observing a model beats learning from being scripted. In S. A. Barab, K. E. Hay, & D. T. Hickey (Eds.), Proceedings of the Seventh International Conference of the Learning Sciences (pp. 634–640). Mahwah, NJ: Lawrence Erlbaum. Salden, R., Aleven, V., Renkl, A., & Schwonke, R. (2009). Worked examples and tutored problem solving: Redundant or synergistic forms of support? Topics in Cognitive Science, 1, 203–213. Schwonke, R., Renkl, A., Krieg, K., Wittwer, J., Aleven, V., & Salden, R. (2009). The Worked-example effect: Not an artefact of lousy control conditions. Computers in Human Behavior, 25, 258–266. Schworm, S., & Renkl, A. (2006). Computer-supported example-based learning: When instructional explanations reduce self-explanations. Computers & Education, 46, 426–445. Schworm, S., & Renkl, A. (2007). Learning argumentation skills through the use of prompts for self-explaining examples. Journal of Educational Psychology, 99, 285–296. Sweller, J., & Cooper, G. A. (1985). The use of worked examples as a substitute for problem solving in learning algebra. Cognition & Instruction, 2, 59–89. VanLehn, K. (1996). Cognitive skill acquisition. Annual Review of Psychology, 47, 513–539. VanLehn, K. (1998). How examples are used during problem solving. Cognitive Science, 22, 347–388. VanLehn, K., Lynch, C., Schulze, K., Shapiro, J. A., Shelby, A., Taylor, D., et al. (2005). The Andes physics tutoring project: Five years of evaluations. International Journal of Artificial Intelligence in Education, 15, 1–47. van Merri¨enboer, J. J. G., Clark, R. E., & De Croock, M. B. M. (2002). Blueprints for complex learning: The 4C/ID-model. Educational Technology, Research and Development, 50, 39–64. van Merri¨enboer, J. J. G., & Kester, L. (2005). The four-component instructional design model: Multimedia principles in environments for complex learning. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (pp. 71–93). New York: Cambridge University Press.
6 Instructional Control of Cognitive Load in the Design of Complex Learning Environments liesbeth kester, fred paas, and jeroen j. g. van merri¨enboer
Recent instructional design theories (e.g., the case method, project-based education, problem-based learning, and competence-based education) tend to focus on authentic learning tasks that are based on real-life experiences as the driving force for complex learning (Merrill, 2002; van Merri¨enboer & Kirschner, 2001). According to these theories, authentic learning tasks have many solutions, are ecologically valid, cannot be mastered in a single session, and pose a very high load on the learner’s cognitive system. Consequently, complex learning has little to do with learning separate skills in isolation, but foremost it deals with learning to coordinate the separate skills that constitute real-life task performance. Thus, in complex learning, the whole is clearly more than the sum of its parts, because it also includes the ability to coordinate the parts. In addition, in complex learning, effective performance relies on the integration of skills, knowledge, and attitudes, where, for instance, complex knowledge structures are underlying problemsolving and reasoning skills and particular attitudes are critical to interpersonal skills or to performing safety procedures. Moreover, complex learning requires differentiation by recognizing qualitative differences among the task characteristics that influence the constituent skills that have to be applied. Figure 6.1 shows an example of a simulated, authentic learning task for novice electricians in vocational education, namely, troubleshooting electrical circuits. Some constituent skills are performed in a variable way across problem situations (e.g., troubleshooting skills, such as orienting or diagnosing). Experts can effectively perform such non-recurrent skills because they have highly complex cognitive schemata available that help them to reason about the domain and to guide their problem-solving behavior. Other constituent skills may be performed in a highly consistent way across problem situations (e.g., building or operating an electrical circuit). Experts can effectively 109
110
Liesbeth Kester, Fred Paas, and Jeroen J. G. van Merri¨enboer
figure 6.1. This example shows a malfunctioning electrical circuit (a). It contains two faults that appear when switch 1 is closed, that is, no current is flowing because the voltmeter and the ammeter are incorrectly connected (b). After this is fixed, two lamps explode because the voltage of the battery is too high for the lamps (c). The learner has to repair this circuit, and to do that, he or she has to coordinate his or her troubleshooting skills (i.e., orient, diagnose, and plan action) and circuit operating skills (i.e., execute the plan), integrate his or her knowlegde about electrical circuits and skills to correctly perform the troubleshooting task, and recognize the features of the electrical circuit that are relevant to reach a solution and those that are not. If these skills are properly executed, this will result in a well-functioning electrical circuit (d).
perform such recurrent skills because their cognitive schemata contain rules that directly associate particular characteristics of the problem situation to particular actions. The classification between non-recurrent and recurrent aspects of complex performance is particularly important because the associated learning processes are fundamentally different from each other. For non-recurrent skills, the main learning processes are related to schema construction and include induction or mindful abstraction from concrete experiences and elaboration of new information. For recurrent skills, the main
Instructional Control of Cognitive Load
111
learning processes are related to schema automation and include restricted encoding or proceduralization of new information in to-be-automated rules and compilation, and strengthening of those rules. This chapter is about the cognitive implications of focusing on authentic or complex tasks in education for the use of instructional methods. Because high cognitive load is a key characteristic of complex tasks, effective learning can only commence if the specific instructions within a complex task are properly aligned with cognitive architecture (Paas, Tuovinen, Tabbers, & Van Gerven, 2003). The notion that the human cognitive architecture should be a major consideration when choosing or designing instructional methods for meaningful learning of complex cognitive tasks is central to the Cognitive Load Theory (CLT; Paas, Renkl, & Sweller, 2003; Sweller, 1988; Sweller, van Merri¨enboer, & Paas, 1998; van Merri¨enboer & Sweller, 2005). CLT assumes that if individuals are to learn effectively in a learning environment, the architecture of their cognitive system, the learning environment, and interactions between both must be understood, accommodated, and aligned. According to CLT, well-chosen or well-designed instructional methods should decrease the load that is not necessary for learning (i.e., extraneous load, typically resulting from badly designed instruction; see Figure 6.2) and optimize the load that directly contributes to learning (i.e., germane load), within the limits of total available capacity to prevent cognitive overload. However, this chapter is about complex tasks, which implicates that even after the removal of all sources of extraneous cognitive load, the intrinsic load resulting from dealing with the element interactivity in the tasks is still too high to allow for efficient learning. CLT, therefore, recommends that instructional designers or teachers use germane-load-inducing methods only in combination with relatively simple tasks, in which the simultaneous processing of all interactive information elements leaves some spare cognitive capacity. In this chapter, however, we oppose this approach by arguing that germane-load-inducing methods can be used with complex tasks. To accomplish this, intrinsic load and germane load must be balanced by limiting the element interactivity of learning tasks while using germaneload-inducing methods. First, we discuss research findings indicating that germane-load-inducing instructional methods used for practicing simple tasks are not effective for practicing complex tasks, at the cost of transfer of learning. Second, we explain how the intrinsic load of complex tasks can be managed to allow the germane load to increase. Third, the implications of this CLT-oriented perspective on learning for instructional design theories are discussed on the basis of three instructional design models for complex
112
figure 6.2. This figure contrasts a learning task with a high extraneous load because it requires a visual search between text and circuit (i.e., split attention; [a]) and a learning task with a lower extraneous load because it does not require this search (b).
113 figure 6.2 (continued)
114
Liesbeth Kester, Fred Paas, and Jeroen J. G. van Merri¨enboer
learning. The chapter ends with a discussion of main conclusions and future research issues.
task complexity and cognitive load Research indicates that many instructional methods that work well for simple tasks do not work well for complex tasks and vice versa (for overviews, see Bainbridge, 1997; Wulf & Shea, 2002). In this section, we first discuss the differential effects of germane-load-inducing methods on learning simple and complex tasks, indicating that the positive effects of these methods decrease with task complexity. Second, we argue that for transfer of training to commence, it is essential to teach complex tasks with germane-loadinducing methods. Germane-Cognitive-Load-Inducing Instructional Methods and Task Complexity A first important germane-load-inducing method affecting learning is practice variability and, in particular, the way that different versions of a learning task are scheduled over practice trials. A common distinction is between low and high contextual interference. In a practice schedule with low contextual interference (i.e., blocked practice), one version of a task is repeatedly practiced before another version of the task is introduced. Under high contextual interference (i.e., random practice), all versions of the task are mixed and practiced in a random order. Contextual interference can be induced by varying the surface features of a task (e.g., context, representation; Quilici & Mayer, 1996) or the structural features of a task (e.g., underlying procedures). Varying the type of battery used in an electrical circuit, for example, would be varying a surface feature because it would not affect the laws of physics that apply to the circuit, whereas varying the type of circuit (i.e., series or parallel) would be varying a structural feature because it would influence the laws of physics that apply to the circuit. For simple tasks, a robust finding is that high contextual interference results in less effective performance during practice (e.g., more time and/or more trials are necessary to reach a pre-specified level of performance) but higher performance during retention tests (for a review, see Magill & Hall, 1990). Possible explanations for the beneficial effects of high contextual interference are that the different versions of a task reside together in working memory and can be compared and contrasted with each other to yield more elaborate representations in memory (Shea & Zimny, 1983) and that high contextual
Instructional Control of Cognitive Load
115
interference conditions result in repeated forgetting of the action plan, resulting in reconstructive activities that eventually yield more accessible representations in memory (Lee & Magill, 1985). What the different explanations have in common is their assumption that random practice of different versions of a task induces germane learning processes that require more effort than does blocked practice but yield cognitive representations that increase later transfer test performance. The findings for contextual interference are less clear for complex tasks, which may be partly due to the fact that learners have difficulty distinguishing surface and structural features of such tasks (Ross & Kilbane, 1997). For complex tasks in sports, beneficial effects of high contextual interference are not found at all or are only found for high-expertise learners but not for lowexpertise learners (Hebert, Landin, & Solmon, 1996). Using drawing tasks, Albaret and Thon (1999) explicitly manipulated task complexity (number of line segments to draw) and studied the effects of contextual interference. As expected, they found that the positive effects of random practice decreased with task complexity and that for the most complex task, blocked practice was even superior to random practice. These results convey the impression that complex tasks leave no processing capacity for the germane cognitive processes that help learners construct better cognitive representations. A second germane-load-inducing method relevant to the design of practice is providing limited guidance and delayed feedback. For simple tasks, reducing the amount of guidance is typically beneficial to learning. For instance, physical guidance in learning motor skills (e.g., using a mechanical stop to indicate a target position, moving the performer’s limb) is more effective when it is used for a limited number of trials than when it is used for a high proportion of trials, and guidance that focuses a learner’s attention only on the external goal of a movement is more effective than guidance that focuses attention also on the specifics of the movement itself (Schmidt, 1991). Paas, Camp, and Rikers (2001) showed that providing limited guidance by loosely indicating the goal (i.e., the end point of the maze) is more effective in maze learning tasks than giving a precise description of the goal. Results indicate that for simple tasks, extensive guidance often has strong positive effects on performance during practice, but when it is withdrawn during tests, learners who practiced with less or no guidance perform better than learners who practiced with extensive guidance. Similarly, giving feedback on some of the practice tasks or on varying aspects of performance results in more effective learning than giving feedback on all tasks or all aspects of performance. Moreover, slightly delayed feedback is more effective than concurrent or immediate feedback (Balzer, Doherty, & O’Connor, 1989).
116
Liesbeth Kester, Fred Paas, and Jeroen J. G. van Merri¨enboer
The findings for the effects of guidance and feedback on complex tasks, however, show another picture. For complex movements in sports, extensive physical assistance proved to be superior to limited physical assistance (Wulf, Shea, & Whitacre, 1998). For striking tasks, Guadagnoli, Dornier, and Tandy (1996) convincingly demonstrated that relatively long feedback summaries (i.e., delayed feedback) were most effective for teaching simple tasks to low-expertise and high-expertise learners and teaching complex tasks to high-expertise learners, but single-task feedback (i.e., immediate feedback) was most effective for teaching complex tasks to low-expertise learners (i.e., a situation with high intrinsic cognitive load). These results suggest that neither limited guidance and feedback nor alternation for the aspects of the task that receive feedback has positive effects on learning complex tasks. In contrast, it seems that the intrinsic load imposed by the complex tasks leaves no processing capacity, allowing learners to develop their own internal monitoring and feedback mechanisms or cognitive representations of how different task aspects interact with each other early in the learning process. The Transfer Paradox The research on instructional design for simple and complex cognitive tasks shows that complex tasks leave no processing capacity for the germane cognitive processes that help learners construct better cognitive representations. In general, the results indicate that the positive effects of germane-loadinducing methods (i.e., random practice, limited guidance, and delayed feedback) decrease as a function of task complexity. Therefore, it seems that instruction of complex cognitive tasks should not be based on the use of germane-load-inducing methods but on highly structured methods (i.e., blocked practice, step-by-step guidance, and immediate feedback) that primarily facilitate performance by taking over part of the cognitive processing from the learner. We do not support this conclusion, however. Highly structured methods may indeed have a positive effect on the acquisition curve and performance on retention tests, but not on problem solving and transfer of learning. Instead, we believe that if one aims at transfer of learning and the ability to show performances that go beyond given learning objectives, it is necessary to use germane-load-inducing methods. This phenomenon, in which the methods that work best for reaching specific objectives are not the methods that work best for reaching transfer of learning, has been described as the ‘transfer paradox’ (van Merri¨enboer, De Croock, & Jelsma, 1997; see also Eaton & Cottrell, 1999). This phenomenon has important implications for the selection of instructional methods for complex tasks.
Instructional Control of Cognitive Load
117
The germane-load-inducing methods that explicitly aim at transfer of learning should take two complementary dimensions of transfer into account. These dimensions are rooted in Selz’s Gestalt approach to transfer (cited in Mandler & Mandler, 1964) and Thorndike and Woodworth’s (1901) ‘identical elements’ approach to transfer. They are closely related to the high road and the low road to transfer (Salomon & Perkins, 1989), innovation and efficiency in transfer (Schwartz, Bransford, & Sears, 2005), and schemabased and rule-based transfer (van Merri¨enboer, 1997). The first approach stresses that transfer may be partly explained by general or abstract knowledge that may be interpreted in the transfer situation (i.e., other use of the same general knowledge); the second approach stresses that transfer may be partly explained by the application of knowledge elements that are shared between the practice and the transfer situation (i.e., the same use of the same specific knowledge). The germane-load-inducing methods balance both complementary dimensions and facilitate the interpretive aspects of knowing for those aspects of a complex task that are different from problem to problem situation (e.g., troubleshooting an electrical circuit) as well as facilitate the applicative aspects of knowing for those aspects of a complex task that are highly similar from situation to situation (e.g., building or operating an electrical circuit; van Merri¨enboer, 1997). Whereas both transfer dimensions need to be carefully balanced, and adaptive experts score high on both dimensions (Gentner et al., 1997), it is important to note that instructional methods that explicitly aim for one or the other can also conflict with each other. The main problem is that starting with highly structured methods that give priority to the applicative aspects of knowing (e.g., building routines) seriously hampers the later development of interpretive aspects of knowing (e.g., building general schemas). These methods constrain the problem spaces within which learners work and then make it more difficult for them to generate creative solutions or ‘think outside the box’. An example is provided in a study by Schwartz, Martin, and Pfaffman (2005), in which children learned to manipulate pieces to help solve fraction problems. One group learned with pie pieces of different sizes, with a focus on routine building because the pieces are easily seen as fractions of a whole; the other group learned with tile pieces of equal sizes, with a focus on interpretation because the pieces should be interpreted as parts of a whole rather than just units. For subsequent problem solving with new materials (beans, bars, etc.), it was found that the interpretation group was better able to use the novel materials, showed better progress, and eventually became more efficient than the routine-building group.
118
Liesbeth Kester, Fred Paas, and Jeroen J. G. van Merri¨enboer
Concluding, highly structured methods, such as blocked practice, stepby-step guidance, and immediate feedback, may help to efficiently reach pre-specified objectives but yield low transfer of learning. In addition, they may block the later development of the second, interpretive dimension of transfer. Therefore, not these germane-load reducing methods, but their counterparts – random practice, limited guidance, and delayed feedback – should be used to teach complex tasks. However, to avoid cognitive overload additional measures have to be taken. We argue that the intrinsic load of complex tasks and the germane load of instructional methods should be balanced during task performance. For a long time, intrinsic load was considered unalterable by instruction, but recently, the effects of different approaches to reduce intrinsic load on learning have been investigated (Ayres, 2006), and some techniques have been described that seem successful in reducing this load (Gerjets, Scheiter, & Catrambone, 2004; Pollock, Chandler, & Sweller, 2002).
managing intrinsic load and inducing germane load According to CLT, the complexity of a task is largely determined by its degree of element interactivity. High-element interactivity requires the learner to process several elements and their relationships simultaneously in working memory in order to learn the task. Low-element interactivity allows the learner to serially process only a few elements at a time. In the next section, we explain how intrinsic load can be managed so that germane load can be induced. Managing Intrinsic Load Instructional approaches to gradually increase the intrinsic load in a training are based on the sequencing of learning tasks from low-element to high-element interactivity. Basically, this sequencing can be operationalized in part-whole or whole-part approaches (see Figure 6.3). In a part-whole approach, the number of elements and interactions between elements may be initially reduced by simplifying the tasks, after which more and more elements and interactions are added. In a whole-part approach, the number of elements and interactions between elements may be immediately presented in their full complexity, but the learner has to take more and more interacting elements into account when performing the tasks. With regard to part-whole approaches, many studies indicate that learners benefit from learning tasks that are sequenced from simple, with
Instructional Control of Cognitive Load
119
figure 6.3. Two approaches to ordering complex tasks: the part-whole approach, which increases the number of interacting elements, and the whole-part approach, which emphasizes more and more interacting elements.
relatively few interacting elements, to complex, with all interacting elements that are necessary for complete understanding. For instance, Mayer and Moreno (2003) discuss studies that show better transfer test performance when students first had to study which components make up a system and only then how the system works. Kester, Kirschner, and van Merri¨enboer (2004a, 2004b, 2006) studied the effects of presenting information necessary to solve a complex task. They found that not presenting all information at once leads to better transfer test performance. Pollock et al. (2002) and Clarke, Ayres, and Sweller (2005) considered mathematical learning tasks and found that, especially for low-expertise learners and high-element interactivity materials, first presenting isolated elements and only then the interacting elements yields higher transfer test performance than presenting all elements simultaneously from the start. Finally, Ayres (2006) also used mathematical learning tasks and found that especially low-expertise learners benefit from the initial reduction in element interactivity, whereas high-expertise learners benefit from high-element interactivity materials used right from the start. Whole-part approaches present high-element interactivity materials in their full complexity right from the beginning, but use learning tasks that focus the learner’s attention on particular subsets of interacting elements. One way to emphasize varying interacting elements of a learning task is to constrain learners’ performance, either through forcing them to behave as an expert would do by requiring them to successfully complete a particular problem-solving phase before entering a next phase (Dufresne, Gerace, Thibodeau-Hardiman, & Mestre, 1992) or through the use of particular task
120
Liesbeth Kester, Fred Paas, and Jeroen J. G. van Merri¨enboer
formats, such as worked examples and completion tasks. Worked examples focus the learners’ attention on elements that represent correct solution steps only, so that they do not have to worry about potential solution steps that are not relevant for the task at hand. Completion tasks present a partial solution that must be completed by the learner. Like worked examples, they constrain the learner’s performance because not all potential solution steps need to be taken into consideration. Many studies indicate that low-expertise learners learn more from studying worked examples or from completing partial solutions than from independently performing the equivalent conventional tasks (for an overview, see Atkinson, Derry, Renkl, & Wortham, 2000). Furthermore, Kalyuga, Chandler, Tuovinen, and Sweller (2001) found that this effect reverses for high-expertise learners. Thus, to accommodate the learner’s increase in expertise during practice, task formats with low-element interactivity (worked examples, completion tasks) should be gradually replaced by conventional tasks with high-element interactivity. To ensure a smooth transition, one may start with worked examples, continue with completion tasks, and end with conventional tasks in an instructional strategy known as the ‘completion strategy’ (van Merri¨enboer, 1990; see also Renkl & Atkinson, 2003). Inducing Germane Load Next to a low-to-high-element interactivity sequencing strategy that lowers intrinsic load and thus frees up cognitive capacity, learning should be promoted by simultaneously implementing germane-load-inducing methods (for an example, see Figure 6.4). As discussed earlier, random practice, limited guidance, and delayed feedback are promising germane-load-inducing methods. Paas and van Merri¨enboer (1994) investigated random practice in combination with worked examples and found that learners who received a training sequence of random worked examples invested less time and mental effort in practice and attained a better transfer performance than learners who received a sequence of blocked worked examples. van Merri¨enboer, Schuurman, De Croock, and Paas (2002) obtained similar results showing that a training combining the completion strategy with random practice yielded higher transfer test performance than a training combining it with blocked practice. With regard to limited guidance and delayed feedback as methods to induce germane cognitive load, a study by Renkl (2002) indicated that using guidance in the form of a minimalist description of the probabilistic rule that was used in a worked example provided had beneficial effects on learning.
121
figure 6.4. The starting point in this example of the two-stage approach to complex learning (i.e., troubleshooting an electrical circuit) is that all sources of extraneous load are removed (see Figure 6.2b); next, intrinsic load is managed by lowering the element interactivity of the learning task (a) so that germane-load-inducing methods can be introduced (b).
122 figure 6.4 (continued).
Instructional Control of Cognitive Load
123
In addition, Renkl and Atkinson (2003) studied the use of self-explanation prompts in combination with the completion strategy in the domain of statistics (probability). During studying the worked examples, they guided the learners by asking them which probability rule was applied in each solution step. They found a strong effect on transfer test performance for learners who received the self-explanation prompts compared with learners who did not receive these prompts. Robins and Mayer (1993) presented sets of worked examples in a training ordered by type and accompanied by feedback that described the problem types. They found that learners who received sets of worked examples together with delayed feedback had superior transfer test performance. These studies suggest that once the task complexity is reduced by lowering the element interactivity as a function of learner expertise, that is, by using a low-to-high-element interactivity sequence or performance constraints, implementing germane-load-inducing methods has beneficial effects on transfer test performance. As shown in this chapter, instructional methods that attempt to balance intrinsic and germane cognitive load during complex learning have clear implications for instructional design and, in particular, the organisation of learning tasks in educational programs that are based on projects, real-life problems or cases, and other complex tasks. We first describe three example instructional design models that specifically aim at complex learning. We will indicate how these models are consistent with the presented methods that aim at balancing the intrinsic and germane cognitive load. First, we describe elaboration theory. This theory stresses the notion that working from simple to complex is a sine qua non for complex learning. Second, we examine goal-based scenarios that focus on the importance of realworld application and transfer of learning. Finally, the four-component instructional design is discussed as an example of a theory that attempts to implement all basic principles of complex learning.
implications for instructional design The basic principle of Reigeluth’s Elaboration Theory (Reigeluth, 1987, 1999; Reigeluth, Merrill, Wilson, & Spiller, 1980; Reigeluth & Stein, 1983; Van Patten, Chao, & Reigeluth, 1986) is that instruction should be organized from the simplest representation of the learning task (i.e., the ‘epitome’, which contains the most fundamental and representative ideas at a concrete level), for example, a simple electrical circuit connected in series or parallel, to increasingly more complex and elaborated representations, for instance, a complex electrical circuit connected in series and parallel. Originally, the
124
Liesbeth Kester, Fred Paas, and Jeroen J. G. van Merri¨enboer
theory focused on the sequencing of instructional content in conceptual and theoretical domains. The broadest, most inclusive concepts are taught first, including the supporting content (i.e., relevant knowledge, skills, and attitudes) related to them, and subsequently, the ever narrower, detailed concepts are taught together with related supporting content. Later, the theory also focused on sequencing interrelated sets of principles. Such a sequence first teaches the broadest, most inclusive, and most general principles along with the supporting content, and then proceeds to teach ever narrower, less inclusive, more detailed, and more precise principles and supporting content. Elaboration theory clearly reflects the presented principles of complex learning. The elaborative approach to sequencing works from simple to complex wholes, which closely resembles a whole-part approach to a lowto-high-element interactivity sequencing strategy. The combination of organizing content (conceptual, theoretical) and supporting content aims at the integration of knowledge, skills, and attitudes, which characterizes complex learning. The concept of ‘learning episodes’ is used to denote instructional units that allow for review and synthesis without breaking up the idea of a meaningful whole and can be used to incorporate limited guidance and delayed feedback. Goal-based scenarios (Schank, 1993/1994; Schank, Fano, Bell, & Jona, 1993/1994) are the backbone of learning in Schank’s learning-by-doing paradigm (Schank, Berman, & MacPherson, 1999). These goal-based scenarios represent ‘a learning-by-doing simulation in which students pursue a goal by practicing target skills and using relevant content knowledge to help them achieve their goal’ (Schank et al., 1999, p. 165). Like the learning episodes in elaboration theory, goal-based scenarios provide an opportunity to integrate knowledge, skills, and attitudes in meaningful wholes, which characterizes complex learning. Unlike the elaboration theory, however, goal-based scenarios pay far less attention to the sequencing of instruction. In contrast, there is a stronger focus on the performance of real-life tasks in authentic contexts to facilitate transfer of learning. This fits the Gestalt approach to transfer, which maintains that more general goals (i.e., integrated objectives) should drive the learning process, because highly specific learning objectives invite learners to apply strategies that do not allow for transfer of learning (see also Machin, 2002, for the role of goals in reaching transfer of learning). van Merri¨enboer’s four-component instructional design (4C/ID) model, (van Merri¨enboer, 1997; van Merri¨enboer, Clark, & De Croock, 2002;
Instructional Control of Cognitive Load
125
van Merri¨enboer et al., 2003) maintains that learning environments for complex tasks can always be described in four components: 1. Learning tasks, which are preferably based on real-life tasks and fulfill the role of a backbone for the training program. 2. Supportive information, which is made available to learners because it helps them to perform the problem-solving and reasoning aspects of learning tasks. It mainly concerns information on how the domain is organized and how problems in the domain can be systematically approached by the learner. 3. Procedural information, which is presented to learners because it helps them to perform the routine aspects of learning tasks. It mainly concerns procedural steps that precisely specify under which conditions particular actions must be taken by the learner. 4. Part-task practice, which may provide learners with additional practice for routine aspects of the complex task that need to be developed to a very high level of automaticity. Three basic prescriptions of the 4C/ID model correspond with the main principles discussed in the previous sections. First, the model suggests that learning tasks should be ordered into so-called task classes, in which earlier task classes have lower element interactivity than later task classes (i.e., a whole-part approach). Even the first task class contains whole and meaningful tasks (i.e., the most essential interacting elements) so that the learners may quickly develop a holistic vision of the whole task that is then gradually embellished in subsequent task classes. Second, when learners start to work on tasks in a new, more complex task class, it is essential to initially focus their attention on those elements that are most important for learning. This may be reached by first constraining and then increasingly relaxing their performance or by starting with worked examples, continuing with completion tasks, and ending with conventional tasks. Third, and probably most important, the combination of ordering learning tasks in simple-tocomplex task classes, with scaffolding learners within a task class, enables the use of instructional methods that evoke a germane cognitive load. Thus, learning tasks should always, right from the beginning of the training program, show random practice, give limited guidance to learners, and provide them with delayed feedback on varying aspects of performance. The three other components of the 4C/ID model explicitly take the two transfer dimensions into account. Supportive information relates to the Gestalt approach that transfer is explained by general or abstract
126
Liesbeth Kester, Fred Paas, and Jeroen J. G. van Merri¨enboer
information that may be interpreted by a task performer to solve a new problem situation. Procedural information and part-task practice mainly relate to the identical elements approach that transfer may be explained by the application of knowledge elements that are shared between the practice and the transfer situation.
discussion In this chapter, we argued that the increasing focus of instructional design theories on the use of complex ‘real-life’ tasks has important implications for the use of instructional methods. Even after removal of all sources of extraneous load, these tasks are often so cognitively demanding that it is impossible to use transfer-enhancing instructional methods right from the start of a training program. We used cognitive load theory to explain how to balance the intrinsic load imposed by a complex task and the germane load caused by instructional methods that aim for transfer. First, intrinsic load can be decreased early in learning by manipulating the element interactivity of the learning tasks. Then, learning tasks can be immediately combined with methods that induce germane cognitive load, such as random practice, limited guidance, and delayed feedback. We showed that these instructional methods can easily be implemented in contemporary instructional design models for complex learning, such as the elaboration theory (Reigeluth, 1987, 1999), Schank’s learning-by-doing paradigm (Schank et al., 1999), and the 4C/ID model (van Merri¨enboer, 1997). Our analysis points out three important directions for future research. First, the assumed interaction between intrinsic-load-reducing methods and germane-load-inducing methods has only been empirically confirmed for a limited number of concrete instructional methods. More research is needed to show that the interaction holds across a wide variety of methods. Second, more research is needed with highly complex real-life tasks performed in ecologically valid settings. Particular instructional methods such as variability might then have unexpected effects, for instance, because it is difficult for learners to distinguish between the surface and structural features of such tasks. Finally, progress must be made with regard to the measurement of cognitive load. Instruments that allow researchers to disentangle changes in cognitive load into changes in intrinsic load on the one hand and germane load on the other hand would be especially helpful to the in-depth analysis of research findings. An important point to consider in the design of training of complex tasks is that the element interactivity or intrinsic load of a task depends
Instructional Control of Cognitive Load
127
on the expertise of the learner: the higher the expertise, the lower the intrinsic load. In other words, if an individual task performer develops more expertise in a task, the functional complexity of the task decreases. In a flexible and adaptive educational program, it should be possible to take differences between individual students into account when suitable learning tasks are selected. Some students have skills acquired elsewhere that should be taken into account, and some students are better able to acquire new skills and therefore need less practice than other students. In the 4C/ID framework, this means that for each individual student, it should be possible at any given point to select the best task class to work on, as well as the amount of performance constraints applied to the selected task. Consequently, a high-ability student may quickly proceed from task class to task class and mainly work on tasks with little performance constraints, whereas a low-ability student may need many more tasks to complete the program, progress slowly from task class to task class, and work mainly on tasks with sizeable performance constraints.
references Albaret, J. M., & Thon, B. (1999). Differential effects of task complexity on contextual interference in a drawing task. Acta Psychologica, 100, 9–24. Atkinson, R. K., Derry, S. J., Renkl, A., & Wortham, D. (2000). Learning from examples: Instructional principles from the worked examples research. Review of Educational Research, 70, 181–214. Ayres, P. (2006). Impact of reducing intrinsic cognitive load on learning in a mathematical domain. Applied Cognitive Psychology, 20, 287–298. Bainbridge, L. (1997). The change in concepts needed to account for human behaviour in complex dynamic tasks. IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans, 27, 351–359. Balzer, W. K., Doherty, M. E., & O’Connor, R. (1989). Effects of cognitive feedback on performance. Psychological Bulletin, 106, 410–433. Clarke, T., Ayres, P., & Sweller, J. (2005). The impact of sequencing and prior knowledge on learning mathematics through spreadsheet applications. Educational Technology, Research and Development, 53, 15–24. Dufresne, R. J., Gerace, W. J., Thibodeau-Hardiman, P., & Mestre, J. P. (1992). Constraining novices to perform expertlike problem analyses: Effects on schema acquisition. The Journal of the Learning Sciences, 2, 307–331. Eaton, D., & Cottrell, D. (1999). Structured teaching methods enhance skill acquisition but not problem-solving abilities: An evaluation of the ‘silent run through’. Medical Education, 33, 19–23. Gentner, D., Brem, S., Ferguson, R. W., Markman, A. B., Levidow, B. B., Wolff, P., et al. (1997). Analogical reasoning and conceptual change: A case study of Johannes Kepler. Journal of the Learning Sciences, 6, 3–40.
128
Liesbeth Kester, Fred Paas, and Jeroen J. G. van Merri¨enboer
Gerjets, P., Scheiter, K., & Catrambone, R. (2004). Designing instructional examples to reduce intrinsic cognitive load: Molar versus modular presentation of solution procedures. Instructional Science, 32, 33–58. Guadagnoli, M. A., Dornier, L., & Tandy, R. D. (1996). Optimal length for summary knowledge of results: The influence of task-related experience and complexity. Research Quarterly for Exercise and Sport, 67, 239–248. Hebert, E. P., Landin, D., & Solmon, M. A. (1996). Practice schedule effects on the performance and learning of low- and high-skilled students: An applied study. Research Quarterly for Exercise and Sport, 67, 52–58. Kalyuga, S., Chandler, P., Tuovinen, J., & Sweller, J. (2001). When problem solving is superior to studying worked examples. Journal of Educational Psychology, 93, 579–588. Kester, L., Kirschner, P. A., & van Merri¨enboer, J. J. G. (2004a). Information presentation and troubleshooting in electrical circuits. International Journal of Science Education, 26(2/6), 239–256. Kester, L., Kirschner, P. A., & van Merri¨enboer, J. J. G. (2004b). Just in time presentation of different types of information in learning statistics. Instructional Science, 32, 233–252. Kester, L., Kirschner, P. A., & van Merri¨enboer, J. J. G. (2006). Just-in-time information presentation: Improving learning and troubleshooting skill. Contemporary Educational Psychology, 31, 167–185. Lee, T. D., & Magill, R. A. (1985). Can forgetting facilitate skill acquisition? In D. Goodman, R. B. Wilberg, & I. M. Franks (Eds.), Differing perspectives on memory, learning and control (pp. 3–22). Amsterdam, The Netherlands: Elsevier North Holland. Machin, M. A. (2002). Planning, managing, and optimizing transfer of training. In K. Kraiger (Ed.), Creating, implementing, and managing effective training and development (pp. 263–301). San Francisco, CA: Jossey-Bass. Magill, R. A., & Hall, K. G. (1990). A review of the contextual interference effect in motor skill acquisition. Human Movement Science, 9, 241–289. Mandler, J. M., & Mandler, G. (1964). Thinking: From association to Gestalt. New York: Wiley. Mayer, R. E., & Moreno, R. (2003). Nine ways to reduce cognitive load in multimedia learning. Educational Psychologist, 38, 43–52. Merrill, M. D. (2002). First principles of instruction. Educational Technology, Research and Development, 50, 43–59. Paas, F., Camp, G., & Rikers, R. (2001). Instructional compensation for age-related cognitive declines: Effects of goal specificity in maze learning. Journal of Educational Psychology, 93, 181–186. Paas, F., Renkl, A., & Sweller, J. (2003). Cognitive load theory and instructional design: Recent developments. Educational Psychologist, 38, 1–4. Paas, F., Tuovinen, J., Tabbers, H., & Van Gerven, P. W. M. (2003). Cognitive load measurement as a means to advance cognitive load theory. Educational Psychologist, 38, 63–71. Paas, F., & van Merri¨enboer, J. J. G. (1994). Variability of worked examples and transfer of geometrical problem-solving skills: A cognitive load approach. Journal of Educational Psychology, 86, 122–133.
Instructional Control of Cognitive Load
129
Pollock, E., Chandler, P., & Sweller, J. (2002). Assimilating complex information. Learning and Instruction, 12, 61–86. Quilici, J. L., & Mayer, R. E. (1996). Role of examples in how students learn to categorize statistics word problems. Journal of Educational Psychology, 88, 144– 161. Reigeluth, C. M. (1987). Lesson blueprints based on the elaboration theory of instruction. In C. M. Reigeluth (Ed.), Instructional theories in action: Lessons illustrating selected theories and models (pp. 245–288). Hillsdale, NJ: Erlbaum. Reigeluth, C. M. (1999). The elaboration theory: Guidance for scope and sequence decisions. In C. M. Reigeluth (Ed.), Instructional design theories and models. A new paradigm of instruction (1st ed., pp. 425–453). Mahwah, NJ: Erlbaum. Reigeluth, C. M., Merrill, M. D., Wilson, B. G., & Spiller, R. T. (1980). The elaboration theory of instruction: A model for sequencing and synthesizing instruction. Instructional Science, 9, 195–219. Reigeluth, C. M., & Stein, F. S. (1983). The elaboration theory of instruction. In C. M. Reigeluth (Ed.), Instructional design theories and models: An overview of their current status (pp. 335–381). Hillsdale, NJ: Erlbaum. Renkl, A. (2002). Worked-out examples: Instructional explanations support learning by self-explanation. Learning and Instruction, 12, 529–556. Renkl, A., & Atkinson, R. K. (2003). Structuring the transition from example study to problem solving in cognitive skill acquisition: A cognitive load perspective. Educational Psychologist, 38, 15–22. Robins, S., & Mayer, R. E. (1993). Schema training in analogical reasoning. Journal of Educational Psychology, 85, 529–538. Ross, B. H., & Kilbane, M. C. (1997). Effects of principle explanation and superficial similarity on analogical mapping in problem solving. Journal of Experimental Psychology – Learning, Memory & Cognition, 23, 427–440. Salomon, G., & Perkins, D. N. (1989). Rocky road to transfer: Rethinking mechanisms of a neglected phenomenon. Educational Psychologist, 24, 113–142. Schank, R. C. (1993/1994). Goal-based scenarios: A radical look at education. Journal of the Learning Sciences, 3, 429–453. Schank, R. C., Berman, T. R., & Macpherson, K. A. (1999). Learning by doing. In C. M. Reigeluth (Ed.), Instructional design theories and models. A new paradigm of instruction (pp. 161–181). Mahwah, NJ: Erlbaum. Schank, R. C., Fano, A., Bell, B., & Jona, M. (1993/1994). The design of goal-based scenarios. Journal of the Learning Sciences, 3, 305–345. Schmidt, R. A. (1991). Motor control and learning: A behavioral emphasis (2nd ed.). Champaign, IL: Human Kinetics. Schwartz, D. L., Bransford, J. D., & Sears, D. (2005). Efficiency and innovation in transfer. In J. Mestre (Ed.), Transfer of learning from a modern multidisciplinary perspective. Greenwich, CT: Information Age Publishing. Schwartz, D. L., Martin, L., & Pfaffman, J. (2005). How mathematics propels the development of physical knowledge. Journal of Cognition and Development, 6(1), 65–88. Shea, J. B., & Zimny, S. T. (1983). Context effects in learning movement information. In R. A. Magill (Ed.), Memory and the control of action (pp. 345–366). Amsterdam, The Netherlands: Elsevier North Holland.
130
Liesbeth Kester, Fred Paas, and Jeroen J. G. van Merri¨enboer
Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12, 257–285. Sweller, J., van Merri¨enboer, J. J. G., & Paas, F. (1998). Cognitive architecture and instructional design. Educational Psychology Review, 10(3), 251–296. Thorndike, E. L., & Woodworth, R. S. (1901). The influence of improvement in one mental function upon the efficiency of other functions. Psychological Review, 8, 247–261. van Merri¨enboer, J. J. G. (1990). Strategies for programming instruction in high school: Program completion vs. program generation. Journal of Educational Computing Research, 6, 265–285. van Merri¨enboer, J. J. G. (1997). Training complex cognitive skills: A four-component instructional design model for technical training. Englewood Cliffs, NJ: Educational Technology Publications. van Merri¨enboer, J. J. G., Clark, R. E., & De Croock, M. B. M. (2002). Blueprints for complex learning: The 4C/ID-model. Educational Technology, Research and Development, 50(2), 39–64. van Merri¨enboer, J. J. G., De Croock, M. B. M., & Jelsma, O. (1997). The transfer paradox: Effects of contextual interference on retention and transfer performance of a complex cognitive skill. Perceptual and Motor Skills, 84, 784–786. van Merri¨enboer, J. J. G., & Kirschner, P. A. (2001). Three worlds of instructional design: State of the art and future directions. Instructional Science, 29, 429–441. van Merri¨enboer, J. J. G., Kirschner, P. A., & Kester, L. (2003). Taking the load off a learners’ mind: Instructional design for complex learning. Educational Psychologist, 38, 5–13. van Merri¨enboer, J. J. G., Schuurman, J. G., De Croock, M. B. M., & Paas, F. (2002). Redirecting learners’ attention during training: Effects on cognitive load, transfer test performance, and training efficiency. Learning and Instruction, 12, 11–37. van Merri¨enboer, J. J. G., & Sweller, J. (2005). Cognitive load theory and complex learning: Recent developments and future directions. Educational Psychology Review, 17, 147–177. Van Patten, J., Chao, C., & Reigeluth, C. M. (1986). A review of strategies for sequencing and synthesizing instruction. Review of Educational Research, 56, 437–471. Wulf, G., & Shea, C. H. (2002). Principles derived from the study of simple skills do not generalize to complex skill learning. Psychonomic Bulletin & Review, 9, 185–211. Wulf, G., Shea, C. H., & Whitacre, C. A. (1998). Physical guidance benefits in learning a complex motor skill. Journal of Motor Behavior, 30, 367–380.
7 Techniques That Reduce Extraneous Cognitive Load and Manage Intrinsic Cognitive Load during Multimedia Learning richard e. mayer and roxana moreno what is multimedia learning? Suppose you open an online multimedia encyclopedia and click on the entry for “pumps.” Then, the computer presents a narrated animation describing how a pump works. Alternatively, suppose you are playing an educational science game on your computer in which you fly to a new planet and must design a plant that would survive there. An on-screen character guides you and explains how the characteristics of the roots, stem, and leaves relate to various environmental conditions. Both of these examples – multimedia lessons and agent-based simulation games – are forms of computer-based multimedia learning environments. They are multimedia learning environments because they involve words (e.g., printed or spoken words) and pictures (e.g., animation, video, illustrations, or photos). They are computer-based learning environments because they are presented via computer. Our goal in this chapter is to explore research-based principles for improving the instructional design of computer-based multimedia learning. We begin with the premise that research on multimedia learning should be theory based, educationally relevant, and scientifically rigorous. By calling for theory-based research, we mean that research on multimedia learning should be grounded in a cognitive theory of multimedia learning. In this chapter, we build on the cognitive theory of multimedia learning (Mayer, 2001, 2005a, 2005b; Mayer & Moreno, 2003), which is adapted from Cognitive Load Theory (CLT) (Paas, Renkl, & Sweller, 2003; Sweller, 1999, 2005).1 By calling for educationally relevant research, we mean that research on multimedia learning should be concerned with authentic learning 1
In this chapter, we generally concur with the editors’ definitions of key terms, such as dual channel, extraneous load, germane load, intrinsic load, limited capacity, multimedia learning, and online learning, although our wording may differ slightly.
131
132
Richard E. Mayer and Roxana Moreno
MULTIMEDIA PRESENTATION
Words
Pictures
SENSORY MEMORY
Ears
Eyes
selecting words
selecting images
LONG-TERM MEMORY
WORKING MEMORY
Sounds
Images
organizing words
organizing images
Verbal Model integrating
Prior Prior Knowledge
Pictorial Model
figure 7.1. Cognitive theory of multimedia learning.
situations and materials. In this chapter, we focus on guidelines for the design of short narrated animations, typically found in multimedia encyclopedias or as part of larger lessons, and on educational simulation games mainly dealing with topics in science and mathematics. By calling for scientifically rigorous research, we mean that research on multimedia learning should use appropriate research methods. In this chapter, our goal is to determine which features affect learning, so we focus on meta-analyses of well-controlled experiments.
how do people learn with multimedia instruction? How do people learn with multimedia presentations and simulations? We begin with three main principles based on cognitive science research: dual channels – humans possess separate channels for processing visual/ pictorial material and auditory/verbal material (Baddeley, 1999; Paivio, 1986), limited capacity – humans are limited in the amount of material they can process in each channel at one time (Baddeley, 1999; Chandler & Sweller, 1991), and active processing – meaningful learning depends on active cognitive processing during learning, including selecting relevant information for further processing, organizing selected material into a coherent mental representation, and integrating incoming material with existing knowledge (Mayer, 2001; Wittrock, 1989) In short, the challenge of meaningful learning is that people must actively process the incoming material in information processing channels that are highly limited. Figure 7.1 summarizes the process of multimedia learning. As shown on the left side, words and pictures are presented in a multimedia lesson, such as a narrated animation. The auditory material impinges on the ears,
Techniques That Reduce Extraneous Cognitive Load
133
whereas the visual material impinges on the eyes. If the learner attends to the incoming information, some of the words are transferred to working memory (indicated by the “selecting words” arrow) and some of the visual material is transferred to working memory (indicated by the “selecting images” arrow). In working memory, the learner mentally organizes the selected words into a verbal model (indicated by the “organizing words” arrow) and mentally organizes the selected images into a pictorial model (indicated by the “organizing images” arrow). Finally, the learner mentally integrates the verbal and pictorial material with each other and with relevant prior knowledge (indicated by the “integrating” arrow). The management of these processes is coordinated with existing knowledge, the learner’s goals, and the learner’s metacognitive strategies. The cognitive theory of multimedia learning presented in Figure 7.1 is based on CLT, which is a broader theory of human cognition. Although the cognitive theory of multimedia learning is a more specialized theory (i.e., dealing with learning from words and pictures), it is consistent with the major features of CLT, particularly with the focus on designing instruction that does not overload the learner’s cognitive system. Specifically, we draw on a central tenet common to CLT and the cognitive theory of multimedia learning, which can be called the triarchic theory of cognitive load. The triarchic theory of cognitive load specifies three kinds of cognitive processing demands during learning: Extraneous cognitive load (corresponding to extraneous processing in the cognitive theory of multimedia learning [CTML]) is cognitive processing that does not contribute to learning. It is caused by presenting the material in a poorly designed layout or including non-essential material in a lesson. Intrinsic cognitive load (corresponding to essential processing in CTML) is cognitive processing imposed by the inherent difficulty of the material. It is caused by having to hold a large number of elements in working memory at the same time. According to CTML, essential processing involves the learner’s initial comprehension of the material through engaging in the cognitive process of attending to the relevant material. Germane cognitive load (corresponding to generative processing in the CTML) is cognitive processing that contributes to learning. It is caused by challenging or motivating the learner to exert effort toward understanding the material. According to the CTML, generative processing involves the learner’s deep understanding of the material through engaging in the cognitive processes of organizing and integrating.
134
Richard E. Mayer and Roxana Moreno table 7.1. Three goals of instructional design
Cognitive theory of multimedia learning
Cognitive load theory
Reduce extraneous cognitive processing
Reduce extraneous cognitive load
Manage essential cognitive processing
Manage intrinsic cognitive load
Foster generative cognitive processing
Foster germane cognitive load
Description of cognitive processing Cognitive processing that does not support learning the essential material Cognitive processing aimed at mentally representing the essential material Cognitive processing aimed at mentally organizing the representation and integrating it with existing knowledge
Given the limits on each learner’s overall amount of cognitive capacity for processing information at any one time, problems occur when the total amount of extraneous, essential, and generative processing exceeds the learner’s cognitive capacity.
three goals for the design of multimedia learning environments Table 7.1 summarizes three goals for the design of multimedia learning environments – reduce extraneous cognitive processing, manage essential cognitive processing, and foster generative processing. The first column describes the goals using the terminology of the cognitive theory of multimedia learning and the second column describes the goals using the corresponding terminology of CLT. The third column provides a brief definition of each of the three goals. Reduce Extraneous Cognitive Processing First, multimedia lessons should minimize the amount of extraneous processing – processing that does not support learning – required of the learner. The most common obstacle to multimedia learning occurs when the presented material contains extraneous material or is poorly laid out. In this situation, the learner is primed to engage in extraneous cognitive processing, that is, cognitive processing that is not directly relevant to learning the essential material. Given that the amount of cognitive capacity is limited,
Techniques That Reduce Extraneous Cognitive Load
135
when learners engage in large amounts of extraneous processing they may have insufficient remaining capacity for essential and generative processing that are needed for meaningful learning. We refer to this situation as extraneous overload – extraneous processing exhausting the available cognitive capacity. Techniques for reducing extraneous cognitive processing are explored in the next section of this chapter. Manage Essential Processing Second, multimedia learning should help the learner to manage essential processing so that it does not overload the learner’s cognitive system. If one is successful in removing extraneous material and inconsiderate layout from a multimedia lesson, then the learner does not have to waste precious cognitive capacity on extraneous processing. Instead, the learner can use cognitive capacity for essential processing – that is, mentally representing the essential material – and generative processing – that is, mentally organizing the essential material and integrating it with prior knowledge. However, in some situations, the demands of essential processing may exceed the learner’s cognitive capacity, resulting in essential overload (Mayer & Moreno, 2003; Mayer, 2005b). Essential overload can occur when the essential material is complex, unfamiliar, or presented at a fast pace. First, material is complex when it contains many components that interact with one another. For example, in an explanation of lightning, there are more than a dozen key elements that interact in multiple ways, such as updrafts, downdrafts, negatively charged particles, positively charged particles, warm air, cool air, and so on. Sweller (1999) defines complexity in terms of element interactivity – that is, the number of interacting components and the nature of their interactions. Second, material is unfamiliar when the learner lacks relevant prior knowledge. Existing knowledge can be used to chunk the incoming material into larger meaningful units, effectively reducing cognitive load. For example, if a learner knows that hot air rises and negative and positive charges attract, this will help chunk an explanation of lightning formation. Sweller (1999) used the term schemas to refer to relevant prior knowledge in long-term memory. Third, material is fast-paced when the presentation rate is faster than the amount of time the learner requires for representing the material. For example, in a narrated animation on lightning formation, the learner may not be able to fully represent one step in the process before the next step is presented. Techniques for managing essential processing are described in a subsequent section of this chapter.
136
Richard E. Mayer and Roxana Moreno table 7.2. Evidence concerning the coherence principle in computer-based multimedia learning Source Moreno & Mayer (2000, Expt. 1) Moreno et al. (2000, Expt. 2) Mayer, Heiser, & Lonn (2001, Expt. 3) Mayer & Jackson (2005, Expt. 2) Median
Content
Effect size
Lightning Brakes Lightning Ocean waves
1.49 0.51 0.70 0.69 0.70
Foster Generative Processing Finally, suppose a multimedia lesson is presented in a way that eliminates extraneous processing and manages essential processing so that the learner has capacity available to engage in generative processing. How can we promote generative processing without exceeding the available cognitive capacity? This issue is discussed in Chapter 8.
research-based principles for reducing extraneous cognitive load One major instructional design problem occurs when multimedia instruction is insensitive to the information processing limitations of the learner. Cognitive overload can occur when too much extraneous material is presented or the material is displayed in confusing ways, or both. In this situation, the learner may use precious cognitive capacity for extraneous processing – cognitive processing that does not enhance learning of the essential material – which may leave insufficient remaining cognitive capacity for essential processing and generative processing. In this section, we explore five research-based principles for overcoming the insensitivity problem: coherence, redundancy, signaling, spatial contiguity, and temporal contiguity principles. Tables 7.2 to 7.6 show evidence supporting each one of these principles on measures of students’ problem-solving transfer. Different letters following experiment numbers indicate separate experimental comparisons. Coherence Principle Consider the following situation. You open a multimedia encyclopedia and click on the entry for “lightning.” On the screen, you see a 140-second
Techniques That Reduce Extraneous Cognitive Load
137
animation depicting the steps in lightning formation, and through the speakers you hear corresponding narration describing the steps in lightning formation. In an effort to spice up the lesson, we could insert several 10second video clips showing sensational lightning storms, and we could include concurrent narration describing interesting facts about the dangers of lightning. However, when Mayer, Heiser, and Lonn (2001, Experiment 3) inserted interesting video clips, students performed worse on transfer tests than when no such video clips were inserted in the lesson. In a similar effort to spice up the lesson, we could insert background music and/or appropriate environmental sounds, such as blowing wind or cracking ice. However, Moreno and Mayer (2000, Experiments 1 and 2) found across two different science lessons that students performed better on retention and transfer tests when the lessons did not have extraneous sounds and that students’ learning was hurt the most when both music and environmental sounds were combined. More recently, Mayer and Jackson (2005, Experiment 2) found that students learned better from a computer-based lesson explaining the formation of ocean waves if formulas and numerical computations were excluded. In four of four experimental tests, students performed better from a narrated animation that was concise rather than elaborated, as shown in Table 7.2. We refer to this finding as the coherence principle – delete extraneous material from multimedia instruction. The median effect size2 of the coherence principle is 0.70. Similar results were reported with paper-based lessons in which students performed better on transfer tests when interesting but extraneous facts and illustrations were excluded rather than included (Harp & Mayer, 1997, Experiment 1; Harp & Mayer, 1998, Experiments 1, 2, 3, & 4; Mayer, Bove, Bryman, Mars, & Tapangco, 1996, Experiments 1, 2, & 3). How does the coherence principle work? Cognitive processing capacity is limited and must be allocated to extraneous, essential, and generative processing. When extraneous material is excluded or inconsiderate layouts are corrected, the learner engages in less extraneous processing. This leaves more capacity for essential and generative processing (as indicated by the organizing and integrating arrows in Figure 7.1) and thus is more likely to lead to meaningful learning outcomes. 2
Effect sizes are based on Cohen’s (1988) d, in which effect sizes below 0.2 are considered negligible, effect sizes between 0.2 and 0.5 are considered small, effect sizes between 0.5 and 0.8 are considered medium, and effect sizes above 0.8 are considered large.
138
Richard E. Mayer and Roxana Moreno table 7.3. Evidence concerning the redundancy principle in computer-based multimedia learning
Source Kalyuga, Chandler, & Sweller (1999, Expt. 1) Kalyuga, Chandler, & Sweller (2000, Expt. 1) Craig, Gholson, & Driscoll (2002, Expt. 2) Mayer, Heiser, & Lonn (2001, Expt. 1) Mayer, Heiser, & Lonn (2001, Expt. 2) Moreno & Mayer (2002a, Expt. 2) Moreno & Mayer (2002b, Expt. 2a) Moreno & Mayer (2002b, Expt. 2b) Median
Content
Effect size
Electrical engineering Electrical engineering Lightning Lightning Lightning Lightning Environmental science game Environmental science game
1.38 0.86 0.67 0.88 1.21 0.72 0.19 0.25 0.79
Redundancy Principle In another attempt to improve the narrated animation explaining lightning formation, we could add captions to the bottom of the screen that mirror the narration. For example, each caption could consist of the same sentence(s) that the narrator is saying. You might think that adding captions (i.e., onscreen text that is identical to the narration) would allow people to choose their preferred mode for receiving words – either in printed or spoken form. In this way, auditory learners could listen to the narration and visual learners could read the on-screen text. However, when redundant on-screen text was added to a narrated animation on lightning, learners performed worse on transfer tests (Craig, Gholson, & Driscoll, 2002, Experiment 2; Mayer et al., 2001, Experiments 1 & 2; Moreno & Mayer, 2002a, Experiment 2). Similarly, students performed worse on transfer tests when redundant on-screen text was added to narrated animations explaining plant growth in an environmental science game (Moreno & Mayer, 2002b, Experiments 2a & 2b). In a set of similar studies, Kalyuga, Chandler, and Sweller (1999, Experiment 1; 2000, Experiment 1) presented a series of electrical diagrams on a computer screen along with an audio message containing spoken words (non-redundant group), but for some students, they added printed words to the screen that were identical to the ongoing audio message (redundant group). Consistent with the foregoing studies, students in the non-redundant group performed better on subsequent transfer tests than did students in the redundant group. Overall, in eight of eight experimental tests, students learned better when redundant on-screen text was excluded. The studies are summarized in Table 7.3 and yield a median effect size of 0.79. We refer to this finding as
Techniques That Reduce Extraneous Cognitive Load
139
the redundancy principle – exclude redundant on-screen text from narrated animations. Similar results were obtained by Mousavi, Low, and Sweller (1995, Experiments 1 & 2) in a paper-based lesson on mathematics. There may be situations in which redundant on-screen text makes pedagogic sense, such as when the students are non-native speakers or are hearing impaired or when the words are technical terms or hard to pronounce. How does the redundancy principle work? Redundant on-screen text creates extraneous processing, because learners may attempt to reconcile the two incoming verbal streams and may have to scan the animation to find elements corresponding to words at the bottom of the screen. When the redundant on-screen text is removed, the learner engages in less extraneous processing, freeing up cognitive capacity to be used for essential and generative processing (as indicated by the organizing and integrating arrows in Figure 7.1). Signaling Principle Sometimes it may not be feasible to delete extraneous material from a computer-based lesson, so a useful alternative is to provide cues that direct the learner’s attention to the essential material in the lesson. For example, Mautone and Mayer (2001, Experiments 3a & 3b) asked people to view a narrated animation on how an airplane achieves lift. For some people (signaled group), the narration included signals, such as a sentence outlining the main sections, phrases that served as headings for each section, and intonation emphasis on linking words, such as “because of this.” For others (non-signaled group), no signals were provided. On a subsequent transfer test, the signaled group outperformed the non-signaled group. Likewise, a study with pre-service teachers showed that signaling different sources of learner diversity within virtual classroom cases promoted their ability to adapt instruction to the special needs of the learners depicted in the cases (Moreno & Abercrombie, in press). This work provides preliminary evidence for the signaling principle – incorporate signals in the narration, such as outlines, headings, and pointer words. In two of two experimental tests, signaling improved transfer performance, yielding a median effect size of 0.70. Table 7.4 summarizes these findings, but more empirical research is needed. Similar results were obtained with a paper-based lesson on lightning formation (Harp & Mayer, 1998, Experiment 3a). How does the signaling principle work? When extraneous material is included in a lesson, learners engage in extraneous processing to the extent
140
Richard E. Mayer and Roxana Moreno table 7.4. Evidence concerning the signaling principle in computer-based multimedia learning Source Mautone & Mayer (2001, Expt. 3a) Mautone & Mayer (2001, Expt. 3b) Moreno & Abercrombie (in press) Median
Content
Effect size
Airplane Airplane Teaching principles
0.60 0.70 0.81 0.70
that they process that extraneous material. Signals such as outlines, headings, and highlights can help direct learners’ attention toward the essential material (as indicated by the “selecting” arrows in Figure 7.1), thus decreasing extraneous processing. This leaves more capacity for essential and generative processing, and thus is more likely to lead to meaningful learning outcomes. Temporal Contiguity Principle Sometimes extraneous processing is caused by inconsiderate layout of the instructional materials. For example, suppose you clicked on the entry for “brakes” in an electronic car manual, and on the screen appeared a speaker icon and a movie icon. When you click on the speaker icon, you hear an explanation of how a car’s braking system works; when you click on the movie icon, you see an animation of how a car’s braking system works without sound. What’s wrong with this scenario? This scenario is likely to lead to extraneous processing because the learner must use a lot of cognitive capacity to hold the entire narration in working memory until the animation is presented (or vice versa). If meaningful learning depends on holding corresponding words and pictures in working memory at the same time, then successive presentation of narration and animation can easily overload the learner’s cognitive system. For example, Mayer and Anderson (1991, Experiments 1 & 2a; 1992, Experiment 1) and Mayer and Sims (1994, Experiment 1) presented students with a short narration and animation explaining how a tire pump works. Some students received a narrated animation in which corresponding segments of the narration and animation were presented simultaneously (simultaneous group), whereas other students received the entire narration either before or after the entire animation (successive group). On a subsequent transfer test, the simultaneous group outperformed the successive group, even though
141
Techniques That Reduce Extraneous Cognitive Load table 7.5. Evidence concerning the temporal contiguity principle in computer-based multimedia learning Source Mayer & Anderson (1991, Expt. 1) Mayer & Anderson (1991, Expt 2a) Mayer & Anderson (1992, Expt. 1) Mayer & Anderson (1992, Expt. 2) Mayer & Sims (1994, Expt. 1) Mayer & Sims (1994, Expt. 2) Mayer, Moreno, Boire, & Vagge (1999, Expt. 1) Mayer, Moreno, Boire, & Vagge (1999, Expt. 2) Median
Content
Effect size
Tire pump Tire pump Tire pump Brakes Tire pump Lungs Lightning Brakes
0.92 1.14 1.66 1.39 0.91 1.22 2.22 1.40 1.31
both groups received identical material. The same pattern of results was obtained with animation and narration lessons on brakes (Mayer & Anderson, 1992, Experiment 2; Mayer, Moreno, Boire, & Vagge, 1999, Experiment 2), lungs (Mayer & Sims, 1994, Experiment 2), and lightning (Mayer et al., 1999, Experiment 1). Overall, in eight of eight experimental tests, students performed better on transfer tests after receiving simultaneous rather than successive presentations of animation and narration. The studies are summarized in Table 7.5, and yield a median effect size of 1.31. We refer to this pattern as the temporal contiguity principle – present corresponding segments of animation and narration concurrently. In a recent review of 13 experimental comparisons, Ginns (2006) reported a mean weighted effect size of d = 0.78, favoring the temporal contiguity principle. How does the temporal contiguity principle work? Extraneous cognitive load is created when the learner must use cognitive capacity to hold the entire narration in working memory until the animation is presented or vice versa (as indicated in the “WORKING MEMORY” section of Figure 7.1). By presenting corresponding segments of the animation and narration at the same time, the learner is able to mentally integrate the verbal and pictorial material within each segment, thereby eliminating the need to hold material in working memory over long periods. Spatial Contiguity Principle Another example of extraneous processing caused by inconsiderate layout occurs when corresponding words and pictures are not near one another
142
Richard E. Mayer and Roxana Moreno table 7.6. Evidence for the spatial contiguity principle in computer-based multimedia learning Source Moreno & Mayer (1999, Expt. 1)
Content
Effect size
Lightning
0.82
on the screen. For example, suppose you viewed an animation on lightning formation in which the sentences describing each step were presented at the bottom of the screen (separated group) or next to the action they described in the animation (integrated group). You might suppose that students should perform equally well in both scenarios because exactly the same information is presented in each. However, in the separated presentation, students must engage in extraneous processing – namely, scanning the screen to see what the caption at the bottom of the screen is referring to. In contrast, extraneous processing is minimized in the integrated presentation because the learner is directed where to look on the screen. Moreno and Mayer (1999, Experiment 1) found that students performed better on a transfer test when on-screen text was placed next to the corresponding element in the animation than when it was placed at the bottom of the screen. We refer to this finding as the spatial contiguity principle – place on-screen text near corresponding elements in the screen. This preliminary finding is based on one experimental comparison, with an effect size of 0.82, as shown in Table 7.6. Similar results, however, were obtained with paper-based lessons on how brakes work (Mayer, 1989, Experiment 1), lightning (Mayer, Steinhoff, Bower, & Mars, 1995, Experiments 1, 2, & 3), electrical engineering (Chandler & Sweller, 1991, Experiment 1; TindallFord, Chandler, & Sweller, 1997, Experiment 1), and mathematical problem solving (Sweller, Chandler, Tierney, & Cooper, 1990, Experiment 1). In a recent review of 37 experimental comparisons concerning spatial contiguity, Ginns (2006) reported a mean weighted effect size of d = 0.72, favoring the spatial contiguity principle. How does the spatial contiguity principle work? When learners receive separated presentations, they must scan the screen to find which part of the graphic corresponds to the words. This is a form of extraneous processing that can be eliminated when the words are placed next to the part of the graphic they describe, thereby enabling more capacity to be used for essential and generative processing (as indicated by the “organizing” and “integrating” arrows in Figure 7.1).
143
Techniques That Reduce Extraneous Cognitive Load table 7.7. Evidence concerning segmenting principle in computer-based multimedia learning Source Mayer & Chandler (2001, Expt. 2) Mayer, Dow, & Mayer (2003, Expt. 2a) Mayer, Dow, & Mayer(2003, Expt. 2b) Moreno (2007, Expt. 1) Moreno (2007, Expt. 2) Median
Content
Effect size
Lightning Electric motor Electric motor Teaching principles Teaching principles
1.13 0.82 0.98 0.39 0.61 0.82
research-based principles for managing intrinsic cognitive load We can eliminate the need for extraneous processing by removing extraneous material and laying out the presentation in a considerate way. This allows the learner to use all of his or her cognitive capacity to engage in essential processing – that is, mentally representing the essential material – and generative processing – that is, mentally organizing and integrating the essential material with existing knowledge. However, even when extraneous processing is eliminated or reduced, the demands of essential processing may be so great that little or no cognitive capacity remains for deeper learning. For example, heavy intrinsic cognitive load occurs when the essential material is complex, fast paced, or unfamiliar to the learner. In this section, we explore three principles for overcoming the difficulty problem by managing intrinsic cognitive load: segmenting, pretraining, and modality principles. Tables 7.7 to 7.9 show evidence supporting each one of these principles on measures of students’ problem-solving transfer. Different letters following experiment numbers indicate separate experimental comparisons. Segmenting Principle Suppose we ask a novice to study a concise narrated animation on how an electric motor works. The electric motor has many interacting parts and its operation is based on a complex chain of electrical and magnetic events. If learners have little prior knowledge to help in organizing the incoming information, they are likely to experience cognitive overload in their attempts to mentally represent the material. How can we overcome this threat of intrinsic overload? One technique is to break the explanation
144
Richard E. Mayer and Roxana Moreno
into bite-sized chunks whose presentation is under the learner’s control – a technique that we call segmenting. For example, Mayer, Dow, and Mayer (2003, Experiments 2a & 2b) presented some students (continuous group) with a continuous narrated animation explaining how an electric motor works. Other students (segmented group) could view meaningful segments of the narrated animation by clicking on a part of the electric motor and then choosing from a list of questions. All students saw the same narrated animation, but the segmented group saw it in bite-sized chunks with pacing under their control. On a transfer test, the segmented group outperformed the continuous group. In a related study, Mayer and Chandler (2001, Experiment 2) presented some students (continuous group) with a narrated animation on lightning formation. For other students (segmented group), after each of 16 segments a “CONTINUE” button appeared in the lower right corner of the screen. The next segment began as soon as the student clicked on the button. Although both groups received the identical narrated animation in the identical order, the segmented group could pause to digest one segment before moving on to the next, thus managing intrinsic cognitive load. As predicted, the segmented group outperformed the continuous group on a transfer test. Using the same method, a recent study (Moreno, 2007) found that college students are better able to transfer a set of teaching principles to novel teaching scenarios when learning with exemplar videos (Experiment 1) or animations (Experiment 2) that are segmented rather than non-segmented. Overall, in five of five experimental tests, segmenting had a positive effect on transfer performance. The studies are summarized in Table 7.7, and yielded a median effect size of 0.82. These findings support the segmenting principle – break narrated animations or narrated videos into learnercontrolled segments. Lee, Plass, and Homer (2006) obtained complementary results when they reduced the complexity of a science simulation on the ideal gas law by separating it from one screen into two screens. How does the segmenting principle work? It is not possible to make the to-be-learned system simpler than it actually is, but it is possible to help the learner understand it by breaking the presentation into bite-sized chunks. In this way, the learner can mentally represent one portion of the system before moving on to the next (as indicated in the “WORKING MEMORY” section of Figure 7.1). Evidence for the reduction in cognitive load resulting from segmenting videos or animations was found in Moreno’s (2007) study, where students’ cognitive load ratings were significantly lower when learning from segmented rather than non-segmented dynamic visual displays (d = 0.64 and 1.00 for Experiments 1 and 2, respectively).
Techniques That Reduce Extraneous Cognitive Load
145
Pretraining Principle Another approach to managing intrinsic cognitive load is to help learners acquire prerequisite knowledge that will help them process the narrated animation. For example, when students receive a narrated animation explaining how a car’s braking system works, they must build component models of each part (e.g., the piston in the master cylinder can be forward or back) and a causal model of the system (e.g., a change in one part affects a change in another part, and so on). If a learner is unfamiliar with cars, then the task of identifying each part may be so demanding that little capacity remains for building a causal model. The solution to this problem is to provide pretraining in the main components of the to-be-learned system, including the name, location, and behavior of the component. For example, Mayer, Mathias, and Wetzell (2002, Experiments 1 & 2) presented students with a narrated animation explaining how a car’s braking system works. Before the narrated animation, some students received a brief pretraining in which they could click on any part in an illustration of the braking system and then see the name of the part and a description of the states the part could be in (e.g., “the piston in the master cylinder can be forward or back”). The pretrained group performed better on a transfer test than did the group without pretraining. In a related study (Mayer et al., 2002, Experiment 3), students received a narrated animation describing how a tire pump works either after pretraining that emphasized the name, location, and behavior of each part (such as the inlet valve and the outlet valve) or no pretraining. On a subsequent transfer test, students in the pretrained group outperformed those who had not received pretraining. Similar results were reported in a computer-based lesson in electrical engineering (Pollock, Chandler, & Sweller, 2002, Experiments 1 & 3) and a computer-based geology simulation game (Mayer, Mautone, & Prothero, 2002, Experiments 2 & 3). Overall, in seven of seven experimental tests, pretraining in the names, location, and behavior of key components resulted in improvements in transfer test performance. These findings are summarized in Table 7.8 and yielded a median effect size of 0.92. The pretraining principle calls for providing learners with pretraining on the names, locations, and behavior of key components before presenting a narrated animation that is difficult, fast paced, or unfamiliar. How does the pretraining principle work? Knowledge in working memory can be used by the learner to help chunk the incoming material, effectively decreasing cognitive load. This is the advantage that experienced learners have in processing multimedia lessons. Pretraining is aimed at
146
Richard E. Mayer and Roxana Moreno table 7.8. Evidence concerning the pretraining principle in computer-based multimedia learning
Source Pollack et al. (2002, Expt. 1) Pollack et al. (2002, Expt. 3) Mayer, Mathias, & Wetzell (2002, Expt. 1) Mayer, Mathias, & Wetzell (2002, Expt. 2) Mayer, Mathias, & Wetzell (2002, Expt. 3) Mayer, Mautone, & Prothero (2002, Expt. 2) Mayer, Mautone, & Prothero (2002, Expt. 3) Median
Content
Effect size
Electrical engineering Electrical engineering Brakes Brakes Tire pump Geology simulation game Geology simulation game
1.22 1.15 0.79 0.92 1.00 0.57 0.85 0.92
providing relevant knowledge in long-term memory (indicated as “prior knowledge” in Figure 7.1), so learners need to allocate less processing to new incoming material. Modality Principle The most heavily researched principle of multimedia design is the modality principle – present animation and narration rather than animation and on-screen text. The modality principle may be particularly important in situations where difficult and unfamiliar material is presented at a fast pace. In this situation, presenting animation and on-screen text can create split attention (Ayres & Sweller, 2005), in which learners must divide their visual processing between the animation and the on-screen text. In this case, the learner can be overwhelmed by the demands of essential processing, or what Sweller (1999) calls intrinsic cognitive load.3 A solution to this problem is 3
It should be noted that in The Cambridge Handbook of Multimedia Learning, Mayer (2005a) lists the modality principle as an example of managing intrinsic cognitive processing, whereas Low and Sweller (2005) describe the modality principle in terms of extraneous cognitive load. The rationale for viewing the modality principle as an example of reducing extraneous cognitive processing is that presenting concurrent graphics and on-screen text constitutes poor instructional design. This poor design can be corrected by converting the printed text into spoken text. When redesigning an instructional message results in improved learning, the reason is that extraneous processing has been reduced, as is the case, for example, with the spatial contiguity principle (or split-attention principle) or the coherence principle. In short, changing from printed to spoken text is just a case of changing from poor layout to good layout, so the underlying cognitive process involved is extraneous processing. The rationale for viewing the modality principle as an example of managing intrinsic processing is that the learner’s visual channel is overloaded with essential material. When a learner receives a concise animation with printed text placed next to corresponding
Techniques That Reduce Extraneous Cognitive Load
147
to present the words as concurrent narration rather than as concurrent on-screen text, thus off-loading the processing of the words from the visual channel to the verbal channel. For example, in four experiments, students performed better on a transfer test after studying a narrated animation on lighting formation than after studying the same animation along with the same words presented as captions on the screen (Craig et al., 2002, Experiment 2; Mayer & Moreno, 1998, Experiment 1; Moreno & Mayer, 1999, Experiments 1 & 2). Moreno and Mayer (1999, Experiment 2) also found that students learned more deeply about brakes from animation and narration than from animation and onscreen text, even though exactly the same animation and the same words were presented. Similar results were reported with computer-based lessons on electrical engineering (Kalyuga et al., 1999, Experiment 1; Kalyuga et al., 2000, Experiment 1) and mathematical problem solving (Jeung, Chandler, & Sweller, 1997, Experiments 1, 2, & 3). Mayer et al. (2003, Experiment 1) reported that students learned better in an interactive simulation of how an electric motor works when they received explanations in the form of animation and narration rather than animation and on-screen text. O’Neil et al. (2000, Experiment 1) also found that students learned better in a virtual reality simulation of an aircraft’s fuel system when they received explanations in the form of animation and narration rather than animation and on-screen text. The same pattern of results was found in the context of a computer-based simulation game intended to teach environmental science in which explanations of plant growth were presented as animation and narration or animation and onscreen text (Moreno & Mayer, 2002b, Experiments 1a & 1b; Moreno et al., 2001, Experiments 4a, 4b, 5a, & 5b). Across 21 of 21 experimental comparisons, students performed better on transfer tests after receiving animation and narration rather than animation and on-screen text. These results are summarized in Table 7.9 and yielded a median effect size of 0.97. The modality principle – present words in spoken form – may help manage essential processing by distributing the cognitive processing across both information-processing channels. The modality principle is most relevant when the material is complex, unfamiliar, or fast-paced. aspects of the pictures, the material is well designed and concise, but the message simply overloads the learner’s visual channel. To better manage the processing of this essential information, the text can be off-loaded from the visual channel to the verbal channel by converting it from printed to spoken form. In short, changing from printed to spoken form is an example of managing essential processing – similar to segmenting and pretraining.
148
Richard E. Mayer and Roxana Moreno
table 7.9. Evidence concerning modality principle in computer-based multimedia learning Source Jeung et al. (1997, Expt. 1) Jeung et al. (1997, Expt. 2) Jeung et al. (1997, Expt. 3) Mayer & Moreno (1998, Expt. 1) Mayer & Moreno (1998, Expt. 2) Kalyuga et al. (1999, Expt. 1) Moreno & Mayer (1999, Expt. 1) Moreno & Mayer (1999, Expt. 2) Kalyuga et al. (2000, Expt. 1) O’Neil et al. (2000, Expt. 1) Moreno, Mayer, Spires, & Lester (2001, Expt. 4a) Moreno et al. (2001, Expt. 4b) Moreno et al. (2001, Expt. 5a) Moreno et al. (2001, Expt. 5b) Craig et al. (2002, Expt. 2) Moreno & Mayer (2002b, Expt. 1a) Moreno & Mayer (2002b, Expt. 1b) Moreno & Mayer (2002b, Expt. 1c) Moreno & Mayer (2002b, Expt. 2a) Moreno & Mayer (2002b, Expt. 2b) Mayer, Dow, & Mayer (2003, Expt. 1) Median
Content
Effect size
Math problems Math problems Math problems Lightning Brakes Electrical engineering Lightning Lightning Electrical engineering Aircraft simulation Environmental science game
0.87 0.33 1.01 1.49 0.78 0.85 1.02 1.09 0.79 1.00 0.60
Environmental science game Environmental science game Environmental science game Lightning Environmental science game Environmental science game Environmental science game Environmental science game Environmental science game Electric motor
1.58 1.41 1.71 0.97 0.93 0.62 2.79 0.74 2.24 0.79 0.97
In a recent review of 43 experimental comparisons, Ginns (2005) reported a mean weighted effect size of d = 0.72, favoring the modality principle. The modality effect is also consistent with classic research on modality reported by Penney (1989). Br¨unken, Plass, and Leutner (2004) have developed a dual-task methodology for measuring cognitive load caused by modality. How does the modality principle work? The learner is able to off-load some of the cognitive processing from the visual channel – which is overloaded – to the verbal channel – which is not overloaded. In Figure 7.1, the arrow from “words” to “eyes” is changed to an arrow from “words” to “ears,” thereby allowing the learner to use the “selecting words” and the “selecting images” arrows rather than just the “selecting images” arrow.
where do we go from here? The research summarized in this chapter has both practical and theoretical implications. On the practical side, we have been able to suggest eight
Techniques That Reduce Extraneous Cognitive Load
149
research-based guidelines for the design of computer-based multimedia instruction. The research is limited to the extent that much of the research was conducted in short-term, laboratory contexts with college students. Future research is needed that examines whether the principles apply in more authentic learning environments. Another limitation is the focus on selected topics in science and mathematics, so future research should address a broader array of the curriculum. On the theoretical side, the principles are consistent with the cognitive theory of multimedia learning and CLT, from which it is derived. This research is consistent with the premise that cognitive load issues are at the center of instructional design. In particular, the major challenge of instructional designers is to reduce extraneous cognitive load and manage essential cognitive load. This would free cognitive capacity for deep processing – which we call fostering generative processing – as described in the next chapter. Further work is needed to elaborate features of the theory such as: (1) What is the role of prior knowledge in guiding each of the five cognitive processes shown in Figure 7.1? (2) How can we measure the cognitive load experienced by the learner in each channel? (3) How can we measure the level of complexity of the presented material? (4) How can we calibrate the amount of extraneous and essential processing required in a computerbased presentation? and (5) What is the nature of the mental representation created by integrating verbal and pictorial material? Overall, research on multimedia learning has made significant progress in the past 15 years. Future research is needed to meet the practical demands for offering guidance to multimedia instructional designers and the theoretical demands for crafting a cognitive theory of how people learn from words and pictures.
author note This chapter is based on chapters 11 and 12 in The Cambridge Handbook of Multimedia Learning (Mayer, 2005a, 2005b). Preparation of this chapter was supported by Grant No. N000140810018 from the Office of Naval Research, entitled “Research-Based Principles for Instructional Games and Simulations.” references Ayres, P., & Sweller, J. (2005). The split-attention principle in multimedia learning. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 135–146). New York: Cambridge University Press.
150
Richard E. Mayer and Roxana Moreno
Baddeley, A. D. (1999). Human memory. Boston: Allyn & Bacon. Br¨unken, R., Plass, J. L., & Leutner, D. (2004). Assessment of cognitive load in multimedia learning with dual-task methodology: Auditory load and modality effects. Instructional Science, 32, 115–132. Chandler, P., & Sweller, J. (1991). Cognitive load theory and the format of instruction. Cognition and Instruction, 8, 293–332. Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Mahwah, NJ: Erlbaum. Craig, S. D., Gholson, B., & Driscoll, D. M. (2002). Animated pedagogical agents in multimedia educational environments: Effects of agent properties, picture features, and redundancy. Journal of Educational Psychology, 94, 428–434. Ginns, P. (2005). Meta-analysis of the modality effect. Learning and Instruction, 15, 313–331. Ginns, P. (2006). Integrating information: A meta-analysis of the spatial contiguity and temporal contiguity effects. Learning and Instruction, 16, 511–525. Harp, S. F., & Mayer, R. E. (1997). The role of interest in learning from scientific text and illustrations: On the distinction between emotional interest and cognitive interest. Journal of Educational Psychology, 89, 92–102. Harp, S. F., & Mayer, R. E. (1998). How seductive details do their damage: A theory of cognitive interest in science learning. Journal of Educational Psychology, 90, 414–434. Jeung, H., Chandler, P., & Sweller, J. (1997). The role of visual indicators in dual sensory mode instruction. Educational Psychology, 17, 329–343. Kalyuga, S., Chandler, P., & Sweller, J. (1999). Managing split-attention and redundancy in multimedia instruction. Applied Cognitive Psychology, 13, 351–371. Kalyuga, S., Chandler, P., & Sweller, J. (2000). Incorporating learner experience into the design of multimedia instruction. Journal of Educational Psychology, 92, 126–136. Lee, H., Plass, J. L., & Homer, B. D. (2006). Optimizing cognitive load for learning from computer-based science simulations. Journal of Educational Psychology, 98, 902–913. Low, R., & Sweller, J. (2005). The modality principle in multimedia learning. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (pp. 147–158). New York: Cambridge University Press. Mautone, P. D., & Mayer, R. E. (2001). Signaling as a cognitive guide in multimedia learning. Journal of Educational Psychology, 93, 377–389. Mayer, R. E. (1989). Systematic thinking fostered by illustrations in scientific text. Journal of Educational Psychology, 81, 240–246. Mayer, R. E. (2001). Multimedia learning. New York: Cambridge University Press. Mayer, R. E. (2005a). Principles for managing essential cognitive processing in multimedia learning: Segmenting, pretraining, and modality principles. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 169–182). New York: Cambridge University Press. Mayer, R. E. (2005b). Principles for reducing extraneous processing in multimedia learning: Coherence, signaling, redundancy, spatial contiguity, and temporal contiguity principles. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 183–200). New York: Cambridge University Press.
Techniques That Reduce Extraneous Cognitive Load
151
Mayer, R. E., & Anderson, R. B. (1991). Animations need narrations: An experimental test of a dual-coding hypothesis. Journal of Educational Psychology, 83, 484–490. Mayer, R. E., & Anderson, R. B. (1992). The instructive animation: Helping students build connections between words and pictures in multimedia learning. Journal of Educational Psychology, 84, 444–452. Mayer, R. E., Bove, W., Bryman, A., Mars, R., & Tapangco, L. (1996). When less is more: Meaningful learning from visual and verbal summaries of science textbook lessons. Journal of Educational Psychology, 88, 64–73. Mayer, R. E., & Chandler, P. (2001). When learning is just a click away: Does simple user interaction foster deeper understanding of multimedia messages? Journal of Educational Psychology, 93, 390–397. Mayer, R. E., Dow, G., & Mayer, S. (2003). Multimedia learning in an interactive selfexplaining environment: What works in the design of agent-based microworlds? Journal of Educational Psychology, 95, 806–813. Mayer, R. E., Heiser, H., & Lonn, S. (2001). Cognitive constraints on multimedia learning: When presenting more material results in less understanding. Journal of Educational Psychology, 93, 187–198. Mayer, R. E., & Jackson, J. (2005). The case for coherence in scientific explanations: Quantitative details can hurt qualitative understanding. Journal of Experimental Psychology: Applied, 11, 13–18. Mayer, R. E., Mathias, A., & Wetzell, K. (2002). Fostering understanding of multimedia messages through pre-training: Evidence for a two-stage theory of mental model construction. Journal of Experimental Psychology: Applied, 8, 147– 154. Mayer, R. E., Mautone, P., & Prothero, W. (2002). Pictorial aids for learning by doing in a multimedia geology simulation game. Journal of Educational Psychology, 94, 171–185. Mayer, R. E., & Moreno, R. (1998). A split-attention effect in multimedia learning: Evidence for dual processing systems in working memory. Journal of Educational Psychology, 90, 312–320. Mayer, R. E., & Moreno, R. (2003). Nine ways to reduce cognitive load in multimedia learning. Educational Psychologist, 38, 43–52. Mayer, R. E., Moreno, R., Boire, M., & Vagge, S. (1999). Maximizing constructivist learning from multimedia communications by minimizing cognitive load. Journal of Educational Psychology, 91, 638–643. Mayer, R. E., & Sims, V. K. (1994). For whom is a picture worth a thousand words? Extensions of a dual-coding theory of multimedia learning? Journal of Educational Psychology, 86, 389–401. Mayer, R. E., Steinhoff, K., Bower, G., & Mars, R. (1995). A generative theory of textbook design: Using annotated illustrations to foster meaningful learning of science text. Educational Technology Research and Development, 43, 31–43. Moreno, R. (2007). Optimizing learning from animations by minimizing cognitive load: Cognitive and affective consequences of signaling and segmentation methods. Applied Cognitive Psychology, 21, 1–17. Moreno, R., & Abercrombie, S. (in press). Promoting awareness of learner diversity in prospective teachers: Signaling individual and group differences within virtual classroom cases. Journal of Technology and Teacher Education.
152
Richard E. Mayer and Roxana Moreno
Moreno, R., & Mayer, R. E. (1999). Cognitive principles of multimedia learning: The role of modality and contiguity. Journal of Educational Psychology, 91, 358–368. Moreno, R., & Mayer, R. E. (2000). A coherence effect in multimedia learning: The case for minimizing irrelevant sounds in the design of multimedia messages. Journal of Educational Psychology, 92, 117–125. Moreno, R., & Mayer, R. E. (2002a). Verbal redundancy in multimedia learning: When reading helps listening. Journal of Educational Psychology, 94, 156–163. Moreno, R., & Mayer, R. E. (2002b). Learning science in virtual reality multimedia environments: Role of methods and media. Journal of Educational Psychology, 94, 598–610. Moreno, R., Mayer, R. E., Spires, H. A., & Lester, J. C. (2001). The case for social agency in computer-based teaching: Do students learn more deeply when they interact with animated pedagogical agents? Cognition and Instruction, 19, 177–213. Mousavi, S. Y., Low, R., & Sweller, J. (1995). Reducing cognitive load by mixing auditory and visual presentation modes. Journal of Educational Psychology, 87, 319–334. O’Neil, H. F., Mayer, R. E., Herl, H. E., Niemi, C., Olin, K., & Thurman, R. A. (2000). Instructional strategies for virtual aviation training environments. In H. F. O’Neil & D. H. Andrews (Eds.), Aircrew training and assessment (pp. 105–130). Mahwah, NJ: Erlbaum. Paas, F., Renkl, A., & Sweller, J. (2003). Cognitive load theory and instructional design: Recent developments. Educational Psychologist, 38, 1–4. Paivio, A. (1986). Mental representations: A dual coding approach. New York: Oxford University Press. Penney, C. G. (1989). Modality effects and the structure of short-term memory. Memory & Cognition, 17, 398–442. Pollock, E., Chandler, P., & Sweller, J. (2002). Assimilating complex information. Learning and Instruction, 12, 61–86. Sweller, J. (1999). Instructional design in technical areas. Camberwell, Australia: ACER Press. Sweller, J. (2005). Implications of cognitive load theory for multimedia learning. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 19–30). New York: Cambridge University Press. Sweller, J., Chandler, P., Tierney, P., & Cooper, M. (1990). Cognitive load and selective attention as factors in the structuring of technical material. Journal of Experimental Psychology: General, 119, 176–192. Tindall-Ford, S., Chandler, P., & Sweller, J. (1997). When two sensory modalities are better than one. Journal of Experimental Psychology: Applied, 3, 257–287. Wittrock, M. C. (1989). Generative processes of comprehension. Educational Psychologist, 24, 345–376.
8 Techniques That Increase Generative Processing in Multimedia Learning: Open Questions for Cognitive Load Research roxana moreno and richard e. mayer In Chapter 7, we defined multimedia learning, described how people learn from verbal and pictorial information according to the Cognitive Theory of Multimedia Learning (CTML; Mayer, 2005), and examined the relationship between CTML and cognitive load theory (CLT; Sweller, 1999). Specifically, we offered a triarchic theory of cognitive load according to which there are three kinds of cognitive processing demands during learning: extraneous, essential, and generative. We defined extraneous processing as the cognitive processes that are not necessary for making sense of the new information, essential processing as the cognitive processes that are required to mentally select the new information that is represented in working memory, and generative processing as the processes of mentally organizing the new information into a coherent structure and integrating the new knowledge representations with prior knowledge. As explained in the previous chapter, the different nature of the three cognitive demands suggests three goals for the design of multimedia learning environments, namely, to reduce extraneous cognitive processing, to help students manage essential cognitive processing, and to foster generative processing. In the present chapter, we focus on the third of these goals by reviewing techniques that are aimed at increasing generative processing in multimedia learning. As in Chapter 7, the methods reviewed in the present chapter have been distilled from the research program of the authors, which is aimed at better understanding how aspects of media design correspond to the cognitive processes that affect knowledge acquisition. According to the active processing principle of CTML, even when the learning environment is carefully designed to exclude unnecessary processing and to reduce the complexity of the materials to-be-learned, students may fail to learn unless instruction includes methods aimed at engaging the learner in investing mental effort in the construction of knowledge. In CLT 153
154
Roxana Moreno and Richard E. Mayer
terms, generative processing increases students’ germane cognitive load, the load that results from cognitive activities that are relevant to the processes of schema acquisition and automation. According to CLT, germane cognitive load is the only type of load that should be increased during learning because “it contributes to, rather than interferes with learning” (Sweller, van Merri¨enboer, & Paas, 1998, p. 264). Germane cognitive load is the result of exerting effort toward understanding the material. Therefore, germane cognitive load should be the result of actively engaging in the cognitive processes of organizing and integrating relevant instructional information. Because CLT posits that intrinsic, extraneous, and germane loads are additive, generative processing should also result in increased overall cognitive load. Do students who engage in the active organization and integration of instructional materials experience higher cognitive load levels? In this chapter we summarize the findings of our work on generative processing methods and suggest directions for future cognitive load research that can help answer this question. In the past, we emphasized the distinction between behavioral activity and cognitive activity (Moreno & Mayer, 2005, 2007). This distinction becomes relevant to our discussion of generative processing methods. According to CTML, deep learning depends on cognitive activity, such as selecting relevant information from a lesson, mentally organizing it into a coherent structure, and integrating the new knowledge with existing knowledge. In contrast, behavioral activity is not necessary or sufficient to achieve deep learning in the CTML model. As we discuss in the forthcoming sections, student behavior may lead to deep learning only when the activity is designed to prime the learner to be cognitively active. In other words, hands-on activity needs to be combined with minds-on activity. Thus, an important goal of our research program has been to examine how interactive multimedia environments (which ask students to become behaviorally active during learning) should be designed to promote appropriate cognitive processing (for a review of interactive multimedia, see Moreno & Mayer, 2007). In this chapter, we review research on five methods for fostering generative processing in non-interactive and interactive learning environments.
how can we promote generative processing in multimedia learning? Suppose a multimedia lesson is presented in a way that eliminates extraneous processing and manages essential processing so that the learner has capacity
Techniques That Increase Generative Processing
155
available to engage in generative processing. How can we promote generative processing without exceeding the learner’s available cognitive capacity? In this section, we explore five research-based design principles that can be used to increase generative processing: multimedia, personalization, guided activity, feedback, and reflection. Tables 8.1 to 8.5 show evidence from our research program supporting each one of these principles on measures of students’ problem-solving transfer. Different letters following experiment numbers indicate separate experimental comparisons. As will be seen in the following sections, for each principle, we cite additional support from individual studies conducted by researchers outside of our workgroup, discuss our theoretical interpretation of the principle and limitations of the research, and offer suggestions for future cognitive load research. Multimedia Principle Consider the following scenario. A teacher who is interested in teaching her students about the process of lightning formation provides them with a one-page description of the causal chain of events that leads to the phenomenon. After studying the text, several students raise their hands with questions such as “Where are the clouds with respect to the freezing level? How do the negative and positive charges interact within the cloud? Does the step leader make it all the way to the ground?” In this situation, students’ questions are suggesting that they are having a difficult time organizing the elements described in the text, which, in turn, will reduce the likelihood of meaningfully integrating the new information with their prior knowledge. To promote the cognitive processes of organization and integration, the teacher may choose to show them a computer animation depicting each one of the steps in lightning formation and how the elements involved in the phenomenon (i.e., clouds, temperature, electric charges, and step leader) interact with one another. In 9 of 9 experimental studies, Mayer and colleagues found that college students who learned science by receiving text and illustrations or narration and animations performed better on transfer tests than did learners who received text alone or narration alone, respectively (Mayer, 2001). This finding was replicated in a set of experiments in which students who learned science with a combination of picture frames and text outperformed those who received text alone or the picture frames alone on tests of transfer (Moreno & Valdez, 2005). Furthermore, Moreno and colleagues found a similar pattern of results in teacher education and art education. In three of four experimental studies, pre-service teachers who learned with a multimedia program that included a video or animation
156
Roxana Moreno and Richard E. Mayer table 8.1. Evidence concerning the multimedia principle in computer-based multimedia learning
Source Mayer & Anderson (1991, Experiment 2a) Mayer & Anderson (1992, Experiment 1) Mayer & Anderson (1992, Experiment 2) Mayer et al. (1996, Experiment 2) Mayer & Gallini (1990, Experiment 1) Mayer & Gallini (1990, Experiment 2) Mayer & Gallini (1990, Experiment 3) Mayer (1989b, Experiment 1) Mayer (1989b, Experiment 2) Moreno & Valdez (2005, Experiment 1a) Moreno & Valdez (2005, Experiment 1b) Moreno & Valdez (2007, Experiment 1) Moreno & Valdez (2007, Experiment 2) Moreno & Ortegano-Layne (2008, Experiment 1) Moreno & Morales (2008) Median
Content
Effect size
Pumps Pumps Brakes Lightning Brakes Pumps Generators Brakes Brakes Lightning Lightning Learning principles Learning principles Learning principles Painting techniques
2.43 1.90 1.67 1.39 1.19 1.00 1.35 1.50 1.71 0.73 1.27 1.01 1.15 1.95 1.51 1.39
illustrating how an expert teacher applied principles of learning to her teaching practice performed better on transfer tests than did those who learned with an identical program that included a narrative corresponding to the video or animation (Moreno & Ortegano-Layne, 2008; Moreno & Valdez, 2007). Additionally, in a recent study in which middle-school students were asked to learn about a set of masterpiece paintings by studying a reproduction of the artwork (visual group), a description of the artwork (verbal group), or both (dual group), the dual group outperformed the verbal and visual groups on a transfer task (Moreno & Morales, 2008). Several studies outside of our laboratory replicated these findings by showing that the combination of pictorial and verbal information can increase reading comprehension (Duquette & Painchaud, 1996; Kost, Foss, & Lenzini, 1999) and the learning of a second language (Al-Seghayer, 2001; Chun & Plass, 1996). We refer to the previously discussed pattern of findings as the multimedia principle – instruction that includes verbal and pictorial representations of knowledge are more likely to lead to meaningful learning than those that present verbal information alone (Fletcher & Tobias, 2005; Mayer, 2001). Table 8.1 shows the mean effect sizes resulting from comparing the transfer scores for students who received words and pictures (multimedia
Techniques That Increase Generative Processing
157
group) with students who received words only (verbal group) for each of the 14 experimental comparisons conducted in our laboratory. In each comparison, the multimedia group performed better than the verbal group on the transfer test. As can be seen in the table, the effect sizes are medium to large, with a median of 1.39, which is a large effect (Cohen, 1988). How does the multimedia principle work? The multimedia principle is based on dual coding theory, or the idea that different coding systems (such as those for words and pictures) reinforce each other (Paivio, 1986). According to CTML’s dual channel assumption, humans possess separate channels for processing pictorial material and verbal material (Mayer & Moreno, 2003). When learners are presented with verbal and pictorial representations of the system to-be-learned, they become more cognitively active because they need to organize each representation and build connections between them. The act of building connections between verbal and pictorial representations is an important step in conceptual understanding (Schnotz, 2005; Seufert, J¨anen, & Br¨unken, 2007). Therefore, students who receive well-constructed multimedia instruction should perform better on transfer tests, which are designed to measure understanding, than students who receive verbal explanations alone. It is important to note that the experimental studies that included a measure of students’ interest showed consistent significant differences favoring the multimedia groups (Moreno & Ortegano-Layne, 2008; Moreno & Valdez, 2007). This finding supports a Cognitive-Affective Theory of Learning with Media (CATLM; Moreno, 2005, 2006, 2009) by suggesting that the improved transfer for multimedia groups may rely not only on additive coding but also, at least in part, on increased student interest. The CATLM was proposed to integrate affective and metacognitive factors into the more traditional, cognitively driven CTML model. Limitations and suggestions for future cognitive load research. Despite its robustness, the multimedia principle needs to be reconsidered under the light of the individual differences reviewed in Chapter 4 of this volume. For example, Mayer and Gallini (1990) found support for the multimedia principle for low-prior-knowledge students but high-prior-knowledge students showed much less difference in performance between multimedia and text-only presentations. This finding suggests that applying the multimedia principle is especially important for fostering learning when learners have little or no prior knowledge in a domain (i.e., novice learners). According to CLT, high-prior-knowledge students already have appropriate schematic knowledge structures from long-term memory that they can retrieve to guide their understanding of the verbal information contained in the
158
Roxana Moreno and Richard E. Mayer
presentation. Consequently, the added visual code may not be necessary to promote deeper understanding of the system to-be-learned. On the other hand, because low-prior-knowledge students do not have appropriate schemas to guide the processing of new information, instruction that includes additional visual coding can help structure information in working memory, thus providing a substitute for missing schemas. A caveat, however, is that not all novices will benefit equally from multimedia instruction. Specifically, Mayer and Sims (1994) found that students with high spatial ability benefit significantly more from the simultaneous presentation of animations and narrations than their counterparts. This finding suggests that students with low spatial ability may need additional guidance to support the processing of dynamic visual materials. It is important to note that the reviewed research was not designed to test CLT’s assumptions. Because the amount and/or type of cognitive load experienced by students in each experimental condition were not measured, the implications for CLT are limited and only indirectly derived from students’ learning outcomes. Therefore, future cognitive load research on the multimedia principle should investigate the relationship between students’ prior knowledge, spatial abilities, and cognitive load. In addition, future research should test our hypothesis that the benefits of multimedia instruction are indeed the result of investing more effort during the encoding of the two sources of information. Although this hypothesis is consistent with CLT and supported by our evidence that students report higher interest during multimedia learning, an alternative hypothesis is that the transfer benefits or multimedia learning are the result of the stronger retrieval produced by dual rather than single representations of the learned system (Paivio, 1986). Personalization Principle Suppose you are a student playing an educational science game called Design-a-Plant (Lester, Stone, & Stelling, 1999), in which you fly to a new planet and must design a plant that would survive there. An on-screen pedagogical agent interacts with you by giving you hints, feedback, and explanations about how the characteristics of the roots, stem, and leaves relate to various environmental conditions. In a non-personalized version of the game, the agent speaks in a formal, monologue style, addressing you as an observer. In a personalized version of the game, the agent speaks in an informal, conversational style, addressing you as if you were both sharing the learning experience. For example, in the non-personalized version of the game the agent might say, “The goal of this program is to design a plant that will survive, maybe even flourish, in an environment of heavy rain. The
159
Techniques That Increase Generative Processing
table 8.2. Evidence concerning the personalization principle in computer-based multimedia learning Source Moreno & Mayer (2000, Experiment 1) Moreno & Mayer (2000, Experiment 2) Moreno & Mayer (2000, Experiment 3) Moreno & Mayer (2000, Experiment 4) Moreno & Mayer (2000, Experiment 5) Moreno & Mayer (2004, Experiment 1a) Moreno & Mayer (2004, Experiment 1b) Mayer et al. (2004, Experiment 1) Mayer et al. (2004, Experiment 2) Mayer et al. (2004, Experiment 3) Median
Content
Effect size
Lightning (narration) Lightning (text) Botany (narration) Botany (text) Botany (narration) Botany (narration) Botany (narration, virtual reality) Lungs (narration) Lungs (narration) Lungs (narration)
1.05 1.61 1.92 1.49 1.11 1.58 1.93 0.52 1.00 0.79 1.30
leaves need to be flexible so they won’t be damaged by the heavy rain.” In the personalized version of the game, the agent might say instead, “Your goal here is to design a plant that will survive, maybe even flourish, in this environment of heavy rain. Your leaves need to be flexible so they are not damaged by the heavy rain.” In ten of ten experimental studies, learners who received personalized messages (as narrated or written text) performed better on transfer tests than did learners who received non-personalized messages. We refer to this finding as the personalization principle – instruction that includes personalized messages is more likely to lead to more meaningful learning than those that use non-personalized messages (Moreno & Mayer, 2000, 2004). The positive effects of personalization, however, have been found to extend to reading comprehension (Reeder, McCormick, & Esselman, 1987) and the comprehension of mathematical word problems (Anand & Ross, 1987; d’Ailly, Simpson, & MacKinnon, 1997; Davis-Dorsey, Ross, & Morrison, 1991). Table 8.2 shows the mean effect sizes resulting from comparing the transfer scores of students who learned in personalized and non-personalized learning conditions for each of the ten experimental studies conducted in our laboratory. In each comparison, the personalized group performed better than the non-personalized group on the transfer test. As can be seen in the table, the effect sizes are medium to large and consistent, with a median of 1.30, which is considered a large effect. How does the personalization principle work? Similar to the case of other self-referential effects found in experimental psychology (Symons & Johnson, 1997), the personalization principle is thought to be based on the
160
Roxana Moreno and Richard E. Mayer
idea that personalization promotes more active processing of the new information by having students relate the material to themselves, thus creating deeper memories of the learning experience. An additional interpretation is that when students are induced to believe that they are participants rather than observers of the learning environment, they become more engaged in making sense of the learning materials. This last interpretation is consistent with the CATLM (Moreno, 2005; Moreno & Mayer, 2007), which proposes that motivation and affect determine how much of the available cognitive resources will be assigned to the learning task. Thus, personalized messages may help learning by influencing students to spend more effort on the task (Pintrich & Schunk, 2002). Limitations and suggestions for future cognitive load research. The personalization principle has been supported for written and auditory messages, for non-interactive and interactive multimedia learning environments, and for desktop and virtual reality displays. However, similar to the multimedia principle, research on the personalization principle very seldom included measures of cognitive load. In one of the more recent experiments (Moreno & Mayer, 2004), students in personalized groups reported significantly lower levels of perceived cognitive load than did students in the non-personalized groups (effect size, d = 0.67, a medium-to-large effect). However, the mechanism underlying students’ cognitive load perceptions needs further investigation. It is not clear why self-referencing might more actively engage students in the learning experience and simultaneously promote lower cognitive load perceptions. Does personalization promote the elaboration of the instructional materials as suggested by past self-referential studies (Symons & Johnson, 1997) or could it be that “personalized messages are more consistent with our schemas for communicating in normal conversations, therefore requiring less cognitive effort to process” (Moreno & Mayer, 1999, p. 725)? A number of studies on narrative comprehension suggest that narrative discourse is easier to comprehend and remember than other, more formal discourse genres (Graesser, Golding, & Long, 1998). Future cognitive load research on the personalization principle should include differentiated measures of intrinsic, extraneous, and germane cognitive load to advance our understanding of this phenomenon. Guided Activity Principle An alternative to presenting a multimedia explanation to teach how a complex system works may consist of asking students to engage in mixedinitiative problem solving with a pedagogical agent (Lester et al., 1999).
Techniques That Increase Generative Processing
161
Guided activity occurs when learners are able to interact in multimedia environments and receive guidance about their actions during learning. Therefore, at the heart of the guided activity principle are the following two ideas: interactivity and feedback. In recent work, we distinguished among five common types of interactivity: manipulating, dialoguing, controlling, searching, and navigating (Moreno & Mayer, 2007). The guided activity principle reviewed in this section focuses mainly on manipulating and dialoguing types of interactivity. In interactivity by manipulating, learners experiment with instructional materials, such as when they set parameters before a simulation runs to test a hypothesis. In interactivity by dialoguing, learners ask questions and receive answers or give answers and receive feedback. In our research program, we tested the guided activity principle with two instructional programs. Using the Design-A-Plant learning environment, we found in three of three experimental studies that learners who were allowed to make choices about the characteristics that plants needed to survive in different weather conditions, outperformed learners who received direct instruction (Moreno, Mayer, Spires, & Lester, 2001). In the first two experiments, middle-school students (Experiment 1) and college students (Experiment 2) who learned with direct instruction could see the same set of roots, stems, and leaves, and received the same instructional words as in the guided activity condition, but were not able to design the plants before listening to the explanations of the program. The third experiment was identical to the second experiment with the exception that college students in both conditions received instructional messages from an on-screen pedagogical agent. A later study showed that elementary-school children who learned with the verbal guidance of an on-screen agent as they independently practiced the addition and subtraction of integers outperformed those who learned without guidance on a transfer measure (Moreno & Dur´an, 2004). Many studies outside of our own workgroup provide evidence for the benefits of presenting students with guided activity. For instance, early research has documented that students learn rules and principles significantly better from guided-discovery rather than pure-discovery methods (Shulman & Keisler, 1966) and that kindergarten children are able to successfully solve conservation tasks when given adult guidance (Gelman, 1969). Students who attempt to learn LOGO programming language with extensive hands-on experience and no guidance are no better than those who receive no programming experience on learning post-tests (Pea & Kurland, 1984), and pure-discovery hinders programming performance compared with learning with guided-discovery methods (Fay & Mayer, 1994;
162
Roxana Moreno and Richard E. Mayer table 8.3. Evidence concerning the guided activity principle in computer-based multimedia learning Source
Content
Effect size
Moreno et al. (2001, Experiment 1) Moreno et al. (2001, Experiment 2) Moreno et al. (2001, Experiment 3) Moreno & Dur´an (2004) Median
Botany Botany Botany Math
0.95 1.20 0.70 0.50 0.83
Lee & Thompson, 1997). Sweller and colleagues demonstrated that students learn how to solve problems significantly better when they are presented with a worked-out problem followed by a practice problem rather than by the traditional method of solving problems with no guidance (Mwangi & Sweller, 1998). More recent studies have found that learning is improved when students are guided to map different sources of information (Seufert & Br¨unken, 2006) or guided to actively integrate multiple sources of information (Bodemer, Pl¨otzner, Bruchm¨uller, & H¨acker, 2005). Table 8.3 shows the mean effect sizes resulting from comparing the transfer scores of students who learned in guided activity and no guided activity (control group) conditions for each of the experimental studies conducted in our laboratory. In each comparison, the guided activity group performed better than the control group on the transfer test. We refer to this finding as the guided activity principle – instruction that allows students to interact by dialoguing and manipulating the learning materials is more likely to lead to meaningful learning than instruction that does not allow for dialoguing (i.e., pure discovery) or for manipulating the learning materials (i.e., direct instruction). As can be seen in the table, the effect sizes are medium to large, with a median of 0.83, which is considered large. How does the guided activity principle work? The theoretical rationale that we offer for the guided activity principle is that prompting students to actively engage in the selection, organization, and integration of new information, encourages essential and generative processing. Guided activity leads to deeper understanding than having students passively process identical instructional materials (Mayer & Moreno, 2003). Yet, meaningful learning may not occur if despite the ability to interact by manipulating the instructional materials, there are no opportunities to engage in dialoguing interactivity with a pedagogical agent, such as the case of pure-discovery learning (Mayer, 2004). Although it may be argued that pure discovery is
Techniques That Increase Generative Processing
163
a way to facilitate active learning by allowing students to explore, manipulate, and test hypotheses (Bruner, 1961; Piaget, 1954; Wittrock, 1966), when novice students are prevented from receiving feedback from a pedagogical agent, they often become lost and frustrated, and their confusion may reinforce or compound existing misconceptions (Brown & Campione, 1994; Garnett, Garnett, & Hackling, 1995). In short, guided activity increases the likelihood that learners who lack proper schemas will select and organize the new information successfully (Mayer, 2004). CLT explains the positive effects of guidance as the result of the more efficient use of students’ limited cognitive resources (Sweller, 1999). Specifically, the unguided search for meaning demands a substantial portion of students’ cognitive capacity, thus, leaving relatively little capacity available to engage in the development of new schemas (Sweller, van Merri¨enboer, & Paas, 1998). “Cognitive load theory suggests that search imposes an extraneous cognitive load that interferes with learning” (Tuovinen & Sweller, 1999, p. 335). Limitations and suggestions for future cognitive load research. Interestingly, across the first three experiments reported in Table 8.3, participants who learned with guided activity did not differ from those who learned with direct instruction on self-reported measures of cognitive load. Therefore, a puzzling result of our research is that students’ cognitive load reports fail to support our hypothesis that providing generative learning activities will result in higher levels of perceived cognitive load as a result of the increased germane load that these activities induce. Moreover, these data do not support CLT’s assumption that unguided search imposes extraneous cognitive load either. A possible interpretation of this finding is that the self-reported measures used in our study, although typical in cognitive load research, are not sensitive enough to capture differences in cognitive load. However, we should not discard the possibility that guided activity may reduce extraneous cognitive load and increase germane cognitive load simultaneously. According to this alternative hypothesis, extraneous and germane cognitive load effects may cancel each other out and lead to similar levels of total cognitive load in guided and unguided instruction. Future research should examine more carefully the relationships among guidance, activity, and measures of the three types of cognitive load. Feedback Principle As pointed out in the previous section, meaningful learning may not occur if students are not given appropriate feedback about their understanding. For example, in a meta-analysis of the effects of feedback in computer-based
164
Roxana Moreno and Richard E. Mayer table 8.4. Evidence concerning the feedback principle in computer-based multimedia learning Source
Content
Effect size
Moreno & Mayer (1999) Moreno (2004, Experiment 1) Moreno (2004, Experiment 2) Moreno & Mayer (2005) Median
Math Botany Botany Botany
0.47 1.16 1.58 1.31 1.24
instruction, Azevedo and Bernard (1995) conclude that feedback messages, to be effective, should stimulate the cognitive processes necessary to gain deep understanding. In our research program, we investigated the role of feedback in discovery-based multimedia learning across four experiments. First, we asked middle-school children to solve a set of sixty-four addition and subtraction practice problems over four training sessions with two feedback methods: corrective feedback (CF), consisting of information about the correctness of their response, or CF plus explanatory feedback (EF), consisting of a verbal explanation relating the arithmetic procedure to a visual metaphor for the procedure to-be-learned (Lakoff & Nunez, 1997). Students who received CF and EF showed greater gains on solving difficult problems than those who learned with CF alone (Moreno & Mayer, 1999). In the next three experiments, we used the Design-A-Plant learning environment. Experiments 2 and 3 showed that students who learned about botany with CF and EF produced higher transfer scores and perceived the program as being less difficult than students who learned with CF alone (Moreno, 2004). Finally, in the fourth study (Moreno & Mayer, 2005), students who learned with EF produced higher transfer scores and showed a greater reduction of their misconceptions over time (effect size, d = 1.88) than those who learned with CF alone. Many other studies, in both classrooms and technology-based environments, support the idea that offering any form of explanatory feedback in combination with corrective feedback is preferable to offering corrective feedback alone (Butler & Winne, 1995; Hattie & Timperley, 2007; Kulhavy & Wager, 1993; Mory, 2004; Pashler, Cepeda, Wixted, & Roher, 2005). Table 8.4 shows the mean effect sizes resulting from comparing the transfer scores of students who learned in EF and CF learning conditions for each of four experimental studies conducted in our laboratory. As can be seen in the table, the effect sizes are medium to large, with a median of 1.24, which is a large effect. We refer to this finding as the feedback principle – novice
Techniques That Increase Generative Processing
165
students learn better when presented with explanatory feedback during learning. How does the feedback principle work? According to CTML, the effectiveness of multimedia learning will depend on the relationship between the amount of feedback given by the system and student’s prior knowledge (Mayer, 2004). EF encourages essential and generative processing by guiding students’ selection and organization of new information when no mental model is available (Schauble, 1990). Although the studies reviewed in this section did not include affective measures, we cannot discard the potential motivational effects of presenting students with EF. Feedback that provides students with information about how their performance can be improved is found to lead to greater intrinsic motivation, task engagement, and persistence than performance feedback (Pressley et al., 2003). Therefore, according to the CATLM, the feedback principle may rely, at least in part, on the facilitative effect that increased motivation has on learning (Moreno, 2009). This hypothesis, however, should be empirically tested in future research. Limitations and suggestions for future cognitive load research. The contribution of the feedback principle to CLT is limited in that most studies did not include direct measures of students’ cognitive load during learning. The only two studies that presented students with a self-report measure of cognitive load suggest that EF helps students by reducing the difficulty that arises from learning with no internal or external guiding schemas (i.e., EF groups reported lower cognitive load than CF groups). Nevertheless, we chose to include the feedback principle among methods aimed at increasing germane cognitive load because according to both CTML and CLT, principle-based explanations promote learning by engaging the learner in cognitive processes that are necessary for schema construction. To test the two alternative CLT hypotheses – EF reduces extraneous load versus EF increases germane load – would require having valid and reliable measures for each load type. Moreover, because of the potential mediation effect that motivation may have on learning, future research should include not only measures of learning but other self-reported and behavioral measures of students’ motivation (Moreno, 2009). In addition, a profitable venue for future research is to extend this work by examining the role of students’ individual differences when learning with different feedback types. For instance, students who have high prior knowledge are likely to experience an expertise reversal effect (see Chapter 3, this volume), and perform better when receiving less rather than more elaborated feedback during problem solving. A cognitive load hypothesis to test
166
Roxana Moreno and Richard E. Mayer
in the future is that experts will experience higher extraneous load when receiving EF and higher germane load when receiving CF alone. Likewise, self-regulated students may use their metacognitive skills to compensate for their lack of knowledge and require less feedback from the system than their counterparts. According to the CATLM, when students are aware of the strengths and limitations of their knowledge, strategies, and motivation, they are better able to regulate their own learning by planning and monitoring the cognitive processes needed for understanding (Moreno, 2005, 2006, 2009). In sum, future research should examine the relationships among feedback, individual differences that may have an impact on cognitive processing (see Chapter 4, this volume), and the three cognitive load types. Reflection Principle Reflection can be implemented in many different ways in multimedia learning environments. For example, a popular reflection method to promote reading comprehension is elaborative interrogation, which consists of asking students to answer “why” questions about information they have just read (Moreno et al., 2001; Seifert, 1993). Another reflection method that can be used for math and science learning is called self-explanations, which consists of asking students to explain their answers to problems during learning (Chi, de Leeuw, Chiu, & La Vancher, 1994). Overall, reflection methods are based on the idea that even when multimedia learning environments are designed with the guided activity and feedback principles in mind, deep learning will depend on the degree to which students invest their cognitive resources to reflect on their actions and feedback. To examine the role of reflection in multimedia learning, we conducted the following two experiments using the Design-A-Plant program. First, we used a two-factor design in which students learned with or without EF for their choices (feedback factor) and with or without elaborative interrogation (reflection factor). In the elaborative interrogation condition, students were asked to provide a rationale for the choices they made as they attempted to discover botany principles. For example, a student in this condition may have been asked “Why did you choose a plant with deep roots for this environment?” There was no reflection effect on transfer and no interaction between reflection and feedback (Moreno & Mayer, 2005, Experiment 1). Although we expected a main reflection effect similar to the one found in the reading comprehension literature, we hypothesized that the effects of elaborative interrogation were diminished in the botany game because
Techniques That Increase Generative Processing
167
students were already primed to actively engage in cognitive activity when asked to decide on a particular plant design. To test this hypothesis, we conducted a follow-up study in which students were asked to manipulate or not manipulate the instructional materials (guided activity factor) and to reflect or not reflect on the principles underlying their plant designs (reflection factor). As expected, the findings showed a significant interaction between the two factors (Moreno & Mayer, 2005). For groups who were not allowed to manipulate the materials, students who were prompted to reflect on worked-out examples presented by a pedagogical agent had higher transfer scores than those who were not prompted to reflect on the examples. Similar to the first experiment, there were no significant differences between reflective and non-reflective treatments on transfer for groups who were allowed to manipulate the materials (Moreno & Mayer, 2005, Experiment 2). A third experimental study revealed that, for reflection to be effective, students must be asked to reflect on correct models of the new information (Moreno & Mayer, 2005, Experiment 3). Specifically, we used a self-explanation method to test the hypothesis that the beneficial effects of reflection are contingent on the quality of the elaboration made by students. The results showed that students who were asked to reflect on correct solutions (i.e., worked-out examples) performed better on transfer tests than did those who were asked to reflect on solutions that presented errors. The positive effects of promoting student reflection have been replicated using a variety of methods. For example, student teachers who were verbally prompted to make connections between learned principles and classroom animations took less time to study the animations, produced higher scores on a transfer test, and showed higher motivation to learn than those who were not prompted (Moreno, 2009); students who were prompted to explain the reasons for choosing each navigation step in a hypermedia system outperformed those who learned without reflection prompting on transfer tests (Bannert, 2006); and students who produced self-explanations of workedout examples (see Chapter 5, this volume) outperformed those who did not on near and far transfer measures (Atkinson, Renkl, & Merrill, 2003; Renkl, Stark, Gruber, & Mandl, 1998; Schworm & Renkl, 2007). How does the reflection principle work? Reflection is at the heart of CTML’s generative processing assumption. Encouraging students to provide principle-based explanations for their thinking promotes the organization and integration of new information with students’ prior knowledge. Moreover, we interpret the findings to support CATLM’s metacognitive mediation assumption by showing that, when learning environments are not interactive, it might be necessary to prompt students to become more
168
Roxana Moreno and Richard E. Mayer
table 8.5. Evidence for the reflection principle in computer-based multimedia learning Source Moreno & Mayer (2005; Experiment 2) Moreno & Mayer (2005; Experiment 3) Moreno & Valdez (2005; Experiment 3) Moreno, Reisslein, & Ozogul (2009) Median
Content
Effect size
Botany Botany Botany Electrical engineering
0.98 0.80 0.71 0.74 0.77
mentally active (Azevedo, 2005; Moreno, 2009). However, an additional study warns us about the potential mindless processing that some interactive learning environments may unintentionally promote (Moreno & Valdez, 2005). Specifically, asking students to organize the steps corresponding to a causal chain of events in an interactive multimedia program was not sufficient to improve their understanding about the topic compared with having students study the organized chain of events. Only when the program was modified to prompt students to evaluate their responses before submitting them for feedback did it promote students’ transfer (Moreno & Valdez, 2005, Experiment 3). A similar phenomenon occurred in a recent study in engineering education (Moreno, Reisslein, & Ozogul, 2009). A fading method (see Chapter 5, this volume) failed to promote transfer until students were asked to reflect on their solutions by comparing them with those of a worked-out example. In sum, it is most important to carefully examine whether the design of students’ interaction promotes superficial or deep processing of the instructional materials (Bangert-Drowns, Kulik, Kulik, & Morgan, 1991). Table 8.5 shows the mean effect sizes resulting from comparing the transfer scores of students who learned with and without reflection methods. In four of four comparisons, the reflection group performed better than the control group on the transfer test, yielding a median effect size of 0.77. We refer to this finding as the reflection principle – instruction that prompts students to reflect on correct solutions is more likely to lead to meaningful learning than instruction that does not present such opportunities, especially when learning environments are not interactive. Limitations and suggestions for future cognitive load research. Similar to the previously discussed principles, the contribution of the reflection principle to CLT is limited because most reported studies were not designed to test CLT assumptions directly. Nevertheless, two of the four studies included a self-report measure of cognitive load (Moreno & Valdez, 2005, Experiment. 3; Moreno et al., 2009). In both studies, however, students in reflective and
Techniques That Increase Generative Processing
169
non-reflective treatments did not differ on their perceived cognitive load during learning. Thus, the reflection principle presents another set of open questions that are ripe for future cognitive load research. For instance, still open is the question of whether methods aimed at promoting germane processing, such as those that prompt students to reflect or evaluate their own learning, will result in greater germane cognitive load than those that do not. An additional area for extending this research is that of examining the role of students’ individual differences in self-regulation ability. An interesting hypothesis to be tested is that the reflection principle will not hold for self-regulated learners because such students already possess the ability to control all aspects of their learning, from planning and monitoring to performance evaluation (Bruning, Schraw, Norby, & Ronning, 2004). Therefore, according to CLT, it is likely that reflection prompts will be redundant to self-regulated learners and therefore will hinder rather than promote learning by increasing extraneous cognitive load. In sum, future research should examine the relationships among reflection methods, selfregulation, and the three cognitive load types.
conclusion Theoretical and Practical Implications The research summarized in this chapter has theoretical and practical implications similar to those discussed in Chapter 7 in this volume. On the practical side, we have been able to suggest five additional research-based guidelines for the design of computer-based multimedia instruction. The goal of the present chapter was to summarize our research on methods that have the potential to promote generative processing when learning from verbal and pictorial information and to propose some future directions for cognitive load research. On the theoretical side, the principles that we offer are consistent with CTML (Mayer, 2005) and with the more recent CATLM, which integrates assumptions about the role of affect, motivation, and metacognition in learning with media (Moreno, 2009). In particular, our empirical work supports the active processing principle defined in Chapter 7 – meaningful learning depends on active cognitive processing during learning including selecting relevant information for further processing, organizing selected material into a coherent mental representation, and integrating incoming material with existing knowledge (Mayer, 2001; Wittrock, 1989). We suggest that generative processing can be promoted: (1) by asking students to build
170
Roxana Moreno and Richard E. Mayer
referential connections between verbal and pictorial representations (multimedia principle); (2) by inducing the feeling that students are participants rather than observers of a learning environment (personalization principle); (3) by scaffolding the exploration of interactive games (guided activity principle); (4) by providing principle-based feedback to students’ responses (feedback principle); and (5) by prompting students to explain and evaluate their understanding (reflection principle). On the other hand, as can be seen from the sections that offer suggestions for future cognitive load research, it is difficult to derive strong theoretical implications for CLT from our work because most of the reviewed studies did not measure students’ cognitive load. Although a few studies included a reliable self-reported measure of overall mental effort (Paas, Tuovinen, Tabbers, & Van Gerven, 2003), this measure does not discriminate between the different cognitive load sources. Furthermore, in some studies, students who learned with active processing methods reported similar or even lower ratings on the effort scales than those who learned more passively, a finding that is counter to CLT’s assumption that germane, extraneous, and intrinsic loads are additive. Future Research Directions in Germane Cognitive Load and Generative Processing A challenge for cognitive load research is to establish what type of cognitive load is being affected when students engage in generative processing. Therefore, one productive direction for future research would entail replicating some of the reviewed experimental studies using valid and reliable measures for the three load types. These measures should help advance the theory of instructional design by testing the interpretations offered in this chapter against those offered by alternative theoretical models. Nevertheless, the generative methods reviewed in this chapter are limited and future research could also extend the present work to other methods aimed at promoting the active selection, organization, and integration of new information with students’ prior knowledge. Promising environments that allow investigating issues of germane cognitive load further are inquiry- and problem-based learning scenarios because they offer students the opportunity to select and manipulate instructional materials, explore multiple representations, test hypotheses, self-assess learning, and reflect on the outcome of knowledge construction. Because these environments are highly complex, the potential sources of extraneous and germane cognitive load can be manipulated to shed light on their interaction and learning
Techniques That Increase Generative Processing
171
effects. For example, an experimental study in an inquiry-based environment may consist of the following cognitive load conditions: low extraneous/low germane, high extraneous/low germane, low extraneous/high germane, and high extraneous/high germane, with extraneous cognitive load being induced by spatial or temporal discontiguity and germane cognitive load being induced by generative processing methods. In addition, because CLT predicts different learning and cognitive load outcomes for learners of different expertise and ability levels, it would be important to test each examined method using a variety of learners. For instance, according to CLT, learners’ level of prior knowledge is extremely diagnostic in categorizing instructional materials and/or methods as imposing intrinsic, extraneous, or germane loads, and to predict learning outcomes accordingly (Sweller, van Merri¨enboer, & Paas, 1998). However, the learning benefits of methods aimed at increasing active learning are not only dependent on students’ prior knowledge. In this regard, it is important to note that the active processing principle underlying the CATLM rests on the following three assumptions (Moreno, 2005, 2006, 2009). First, it assumes that the learner has enough capacity available to engage in generative processing. Second, it assumes that the learner has the necessary skills to be successful in the required mental activity. Third, it assumes that the learner is willing to spend his/her available cognitive resources and relevant skills on generative processing. These assumptions, however, may not hold for all learners. In fact, recent research shows that adding cognitive activities aimed at increasing germane load does not necessarily increase student learning (Moreno, 2006). To mention a few examples of this phenomena: some learners do not spontaneously engage in germane cognitive activities such as example elaboration or comparison (Chi, Bassok, Lewis, Reimann, & Glaser, 1989; Gerjets, Scheiter, & Catrambone, 2004), learners vary in the degree of elaboration requested by the learning environment (Renkl, Atkinson, Maier, & Staley, 2002), and learners may show an “illusion of understanding” by reporting lower ratings on effort scales but no increased learning when engaged in more active processing methods (Renkl, 2002). Whether these findings are the result of insufficient cognitive resources, skills, motivation, or metacognition is an open question in cognitive load research. In sum, to advance our understanding about who learns from generative processing methods and how, new CLT developments should specify the mediating effects of students’ individual differences, especially those related to their abilities, motivation, and self-regulation on the three load types (Moreno, in press).
172
Roxana Moreno and Richard E. Mayer
Finally, we should note that similar to the research reported in the previous chapter, most of the reported studies in this chapter were conducted in short-term, laboratory contexts using low-prior-knowledge college students as participants. Future research should examine whether the reviewed principles apply to more authentic learning environments and to a variety of learners and content domains.
author note This material is based upon work supported by the U.S.A. National Science Foundation under Grant No. 0238385. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author and do not necessarily reflect the views of the funding agency. Corresponding author’s address and email: Educational Psychology Program, Simpson Hall 123, University of New Mexico, Albuquerque, NM 87131,
[email protected]. references Al-Seghayer, K. (2001). The effect of multimedia annotation modes on L2 vocabulary acquisition: A comparative study. Language Learning and Technology, 5, 202–232. Anand, P., & Ross, S. M. (1987). Using computer-assisted instruction to personalize math learning materials for elementary school children. Journal of Educational Psychology, 79, 72–79. Atkinson, R. K., Renkl, A., & Merrill, M. M. (2003). Transitioning from studying examples to solving problems: Combining fading with prompting fosters learning. Journal of Educational Psychology, 95, 774–783. Azevedo, R. (2005). Computer environments as metacognitive tools for enhancing learning. Educational Psychologist, 40, 193–197. Azevedo, R., & Bernard, R. M. (1995). A meta-analysis of the effects of feedback in computer-based instruction. Journal of Educational Computing Research, 13(2), 111–127. Bangert-Drowns, R. L., Kulik, C. C., Kulik, J. A., & Morgan, M. T. (1991). The instructional effect of feedback in test-like events. Review of Educational Research, 61(2), 213–238. Bannert, M. (2006). Effects of reflection prompts when learning with hypermedia. Journal of Educational Computing Research, 35(4), 359–375. Bodemer, D., Pl¨otzner, R., Bruchm¨uller, K., & H¨acker, S. (2005). Supporting learning with interactive multimedia through active integration of representations. Instructional Science, 33, 73–95. Brown, A. L., & Campione, J. C. (1994). Guided discovery in a community of learners. In K. McGilley (Ed.), Classroom lessons: Integrating cognitive theory and classroom practice (pp. 229–272). Cambridge, MA: MIT Press. Bruner, J. S. (1961). The art of discovery. Harvard Educational Review, 31 (1), 21–32.
Techniques That Increase Generative Processing
173
Bruning, R. H., Schraw, G. J., Norby, M. M., & Ronning, R. R. (2004). Cognitive psychology and instruction. Upper Saddle River, NJ: Prentice-Hall. Butler, D. L., & Winne, P. H. (1995). Feedback and self-regulated learning. Review of Educational Research, 65, 245–281. Chi, M. T. H., Bassok, M., Lewis, M. W., Reimann, P., & Glaser, R. (1989). Selfexplanations: How students study and use examples in learning to solve problems. Cognitive Science, 13, 145–182. Chi, M. T. H., de Leeuw, N., Chiu, M. H., & La Vancher, C. (1994). Eliciting selfexplanations improves understanding. Cognitive Science, 18, 439–477. Chun, D., & Plass, J. (1996). Effects of multimedia annotations on vocabulary acquisition. The Modern Language Journal, 80(2), 183–198. Cohen, J. (1988). Statistical power analysis for the social sciences (2nd ed.). Hillsdale, NJ: Erlbaum. d’Ailly, H. H., Simpson, J., & MacKinnon, G. E. (1997). Where should “you” go in a math compare problem? Journal of Educational Psychology, 89, 562–567. Davis-Dorsey, J., Ross, S. M., & Morrison, G. R. (1991). The role of rewording and context personalization in the solving of mathematical word problems. Journal of Educational Psychology, 83, 61–68. Duquette, L., & Painchaud, G. (1996). A comparison of vocabulary acquisition in audio and video contexts. The Canadian Modern Language Review, 54(1) 143–172. Fay, A. L., & Mayer, R. E. (1994). Benefits of teaching design skills before teaching LOGO computer programming: Evidence for syntax independent learning. Journal of Educational Computing Research, 11, 187–210. Fletcher, J. D., & Tobias, S. (2005). The multimedia principle. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 117–133). New York: Cambridge University Press. Garnett, P. J., Garnett, P. J., & Hackling, M. W. (1995). Students’ alternative conceptions in chemistry: A review of research and implications for teaching and learning. Studies in Science Teaching, 25, 69–95. Gelman, R. (1969). Conservation acquisition: A problem of learning to attend to relevant attributes. Journal of Experimental Child Psychology, 7, 167–187. Gerjets, P., Scheiter, K., & Catrambone, R. (2004). Designing instructional examples to reduce intrinsic cognitive load: Molar versus modular presentation of solution procedures. Instructional Science, 32, 33–58. Graesser, A. C., Golding, J. M., & Long, D. L. (1998). Narrative representation and comprehension. In R. Barr, M. L. Kamil, P. B. Mosenthal, & P. D. Pearson (Eds.), Handbook of reading research (Vol. II, pp. 171–205). White Plains, NY: Longman Publishing Group. Hattie, J., & Timperley, H. (2007). The power of feedback. Review of Educational Research, 77, 81–112. Kost, C., Foss, P., & Lenzini, J. (1999). Textual and pictorial gloss: Effectiveness on incidental vocabulary growth when reading in a foreign language. Foreign Language Annals, 32(1), 89–113. Kulhavy R. W., & Wager, W. (1993). Feedback in programmed instruction: Historical content and implications for practice. In J. V. Dempsey & G. C. Sales (Eds.), Interactive instruction and feedback (pp. 3–20). Engelwood Cliffs, NJ: Educational Technology.
174
Roxana Moreno and Richard E. Mayer
Lakoff, G., & Nunez, R. E. (1997). The metaphorical structure of mathematics: Sketching out cognitive foundations for a mind-based mathematics. In L. D. English (Ed.), Mathematical reasoning: Analogies, metaphors, and images (pp. 21– 89). Mahwah, NJ: Erlbaum. Lee, M., & Thompson, A. (1997). Guided instruction in LOGO programming and the development of cognitive monitoring strategies among college students. Journal of Educational Computing Research, 16, 125–144. Lester, J. C., Stone, B. A., & Stelling, J. D. (1999). Lifelike pedagogical agents for mixed-initiative problem solving in constructivist learning environments. User Modeling and User-Adapted Interaction, 9, 1–44. Mayer, R. E. (2001). Multimedia learning. New York: Cambridge University Press. Mayer, R. E. (2004). Should there be a three-strikes rule against pure discovery learning? American Psychologist, 59, 14–19. Mayer, R. E. (2005). Cognitive theory of multimedia learning. In R. Mayer (Ed.), The Cambridge handbook of multimedia learning (pp. 31–48). New York: Cambridge University Press. Mayer, R. E., & Anderson, R. B. (1991). Animations need narrations: An experimental test of a dual-coding hypothesis. Journal of Educational Psychology, 83, 484–490. Mayer, R. E., & Anderson, R. B. (1992). The instructive animation: Helping students build connections between words and pictures in multimedia learning. Journal of Educational Psychology, 84, 444–452. Mayer, R. E., Bove, W., Bryman, A., Mars, R., & Tapangco, L. (1996). When less is more: Meaningful learning from visual and verbal summaries of science textbook lessons. Journal of Educational Psychology, 88, 64–73. Mayer, R. E., Fennell, S., Farmer, L., & Campbell, J. (2004). A personalization effect in multimedia learning: Students learn better when words are in conversational style rather than formal style. Journal of Educational Psychology, 96, 389–395. Mayer, R. E., & Gallini, J. K. (1990). When is an illustration worth ten thousand words? Journal of Educational Psychology, 82, 715–726. Mayer, R. E., & Moreno, R. (2003). Nine ways to reduce cognitive load in multimedia learning. Educational Psychologist, 38, 43–52. Mayer, R. E., & Sims, V. K. (1994). For whom is a picture worth a thousand words? Extensions of a dual-coding theory of multimedia learning? Journal of Educational Psychology, 86, 389–401. Moreno, R. (2004). Decreasing cognitive load for novice students: Effects of explanatory versus corrective feedback on discovery-based multimedia. Instructional Science, 32, 99–113. Moreno, R. (2005). Instructional technology: Promise and pitfalls. In L. PytlikZillig, M. Bodvarsson, & R. Bruning (Eds.), Technology-based education: Bringing researchers and practitioners together (pp. 1–19). Greenwich, CT: Information Age Publishing. Moreno, R. (2006). When worked examples don’t work: Is cognitive load theory at an impasse? Learning and Instruction, 16, 170–181. Moreno, R. (2009). Learning from animated classroom exemplars: The case for guiding student teachers’ observations with metacognitive prompts. Journal of Educational Research and Evaluation, 15(5), 487–501.
Techniques That Increase Generative Processing
175
Moreno, R. (in press). Cognitive load theory: More food for thought. Instructional Science. Advance online publication. doi: 10.1007/s11251-009-9122-9. Moreno, R., & Dur´an, R. (2004). Do multiple representations need explanations? The role of verbal guidance and individual differences in multimedia mathematics learning. Journal of Educational Psychology, 96, 492–503. Moreno, R., & Mayer, R. E. (1999). Multimedia-supported metaphors for meaning making in mathematics. Cognition and Instruction, 17, 215–248. Moreno, R., & Mayer, R. E. (2000). Engaging students in active learning: The case for personalized multimedia messages. Journal of Educational Psychology, 92, 724–733. Moreno, R., & Mayer, R. E. (2004). Personalized messages that promote science learning in virtual environments. Journal of Educational Psychology, 96, 165– 173. Moreno, R., & Mayer, R. E. (2005). Role of guidance, reflection, and interactivity in an agent-based multimedia game. Journal of Educational Psychology, 97, 117–128. Moreno, R., & Mayer, R. E. (2007). Interactive multimodal learning environments. Educational Psychology Review, 19, 309–326. Moreno, R., Mayer, R. E., Spires, H. A., & Lester, J. C. (2001). The case for social agency in computer-based teaching: Do students learn more deeply when they interact with animated pedagogical agents? Cognition and Instruction, 19, 177– 213. Moreno, R., & Morales, M. (2008). Studying master paintings to promote painting skills: The role of visualization, copying from memory, and spatial ability. Empirical Studies of the Arts, 27(2), 131–154. Moreno, R., & Ortegano-Layne, L. (2008). Using cases as thinking tools in teacher education: The role of presentation format. Educational Technology Research and Development, 56, 449–465. Moreno, R., Reisslein, M., & Ozogul, G. (2009). Optimizing worked-example instruction in electrical engineering: The role of fading and feedback during problem-solving practice. Journal of Engineering Education, 98, 83–92. Moreno, R., & Valdez, A. (2007). Immediate and delayed learning effects of presenting classroom cases in teacher education: Are video cases or case narratives more effective? Journal of Educational Psychology, 99, 194–206. Moreno, R., & Valdez, A. (2005). Cognitive load and learning effects of having students organize pictures and words in multimedia environments: The role of student interactivity and feedback. Educational Technology Research and Development, 53, 35–45. Mory, E. H. (2004). Feedback research revisited. In D. H. Jonassen (Ed.), Handbook of research on educational communications and technology (2nd ed., pp. 745–783). Mahwah, NJ: Erlbaum. Mwangi, W., & Sweller, J. (1998). Learning to solve compare word problems: The effect of example format and generating self-explanations. Cognition and Instruction, 16, 173–199. Paas, F., Tuovinen, J. E., Tabbers, H., & Van Gerven, P. W. M. (2003). Cognitive load measurement as a means to advance cognitive load theory. Educational Psychologist, 38, 63–71.
176
Roxana Moreno and Richard E. Mayer
Paivio, A. (1986). Mental representations: A dual coding approach. New York: Oxford University Press. Pashler, H., Cepeda, N. J., Wixted, J. T., & Roher, D. (2005). When does feedback facilitate learning of words? Journal of Experimental Psychology, Memory, and Cognition, 31, 3–8. Pea, R. D., & Kurland, D. M. (1984). On the cognitive effects of learning computer programming. New Ideas in Psychology, 2, 137–168. Piaget, J. (1954). The construction of reality in the child. New York: Basic Books. Pintrich, P., & Schunk, D. (2002). Motivation in education: Theory, research, and applications (2nd ed.). Upper Saddle River, NJ: Prentice-Hall. Pressley, M., Dolezal, S. E., Raphael, L. M., Welsh, L. M., Bogner, K., & Roehrig, A. D. (2003). Motivating primary-grade students. New York: Guilford. Reeder, G. D., McCormick, C. B., & Esselman, E. D. (1987). Self-referent processing and recall of prose. Journal of Educational Psychology, 79, 243–248. Renkl, A. (2002). Worked-out examples: Instructional explanations support learning by self-explanations. Learning and Instruction, 12, 529–556. Renkl, A., Atkinson, R. K., Maier, U. H., & Staley, R. (2002). From example study to problem solving: Smooth transitions help learning. The Journal of Experimental Education, 70, 293–315. Renkl, A., Stark, R., Gruber, H., & Mandl, H. (1998). Learning from workedout examples: The effects of example variability and elicited self-explanations. Contemporary Educational Psychology, 23, 90–108. Schauble, L. (1990). Belief revision in children: The role of prior knowledge and strategies for generating evidence. Journal of Experimental Child Psychology, 49, 31–57. Schnotz, W. (2005). An integrated model of text and picture comprehension. In R. E. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 49–69). New York: Cambridge University Press. Schworm, S., & Renkl, A. (2007). Learning argumentation skills through the use of prompts for self-explaining examples. Journal of Educational Psychology, 99, 285–296. Seifert, T. L. (1993). Effects of elaborative interrogation with prose passages. Journal of Educational Psychology, 85, 642–651. Seufert, T., & Br¨unken, R. (2006). Cognitive load and the format of instructional aids for coherence formation. Applied Cognitive Psychology, 20, 321– 331. Seufert, T., J¨anen, I., & Br¨unken, R. (2007). The impact of intrinsic cognitive load on the effectiveness of graphical help for coherence formation. Computers in Human Behavior, 23, 1055–1071. Shulman, L. S., & Keisler, E. R. (1966). Learning by discovery. Chicago: Rand McNally. Sweller, J. (1999). Instructional design in technical areas. Camberwell, Australia: ACER Press. Sweller, J., van Merri¨enboer, J. J. G., & Paas, F. W. C. (1998). Cognitive architecture and instructional design. Educational Psychology Review, 10, 251–296. Symons, C. S., & Johnson, B. T. (1997). The self-reference effect in memory: A meta-analysis. Psychological Bulletin, 121, 371–394.
Techniques That Increase Generative Processing
177
Tuovinen, J. E., & Sweller, J. (1999). A comparison of cognitive load associated with discovery learning and worked examples. Journal of Educational Psychology, 91, 334–341. Wittrock, M. C. (1966). The learning by discovery hypothesis. In L. S. Shulman & E. R. Keislar (Eds.), Learning by discovery: A critical appraisal (pp. 33–75). Chicago: Rand McNally. Wittrock, M. C. (1989). Generative processes of comprehension. Educational Psychologist, 24, 345–376.
part three DISCUSSION
9 Measuring Cognitive Load ¨ roland brunken, tina seufert, and fred paas
the problem of cognitive load measurement: what are good cognitive load indicators? The previous chapters have outlined the basic theoretical assumptions for cognitive load theory (Chapter 2), described how cognitive load affects the process of schema acquisition (Chapter 3), and discussed the role that learners’ individual differences play in the process of knowledge construction (Chapter 4). The central problem identified by Cognitive Load Theory (CLT) is that learning is impaired when the total amount of processing requirements exceeds the limited capacity of human working memory. In addition to the fundamental assumption that learning is a function of available cognitive resources, CLT makes some additional assumptions with respect to the relation among cognitive resources, demands, and learning. The first of these additional assumptions is that instructional design and/or methods may induce either a useful (germane) or a wasteful (extraneous) consumption of cognitive capacity. The second assumption is that the source of cognitive load can also vary depending on the complexity of the task to-be-solved (intrinsic cognitive load defined by element interactivity). There is a large body of empirical research supporting the assumptions of CLT by analyzing the relation between the factors influencing cognitive load and learning outcomes. For example, several empirically well-established instructional design principles (Mayer, 2005) were identified in that line of research, which are discussed in other chapters of this book (see Chapters 7, 8). However, can CLT’s assumptions be verified directly? Are there valid and reliable methods to test the complex interactions between the learner’s cognitive resources, demands, and learning? In this chapter, we describe and discuss the various measures of cognitive load currently available. 181
182
Roland Br¨unken, Tina Seufert, and Fred Paas
Cognitive load research has produced several methods and approaches within the last fifteen years – yet, we have neither a single standardized method for cognitive load measurement nor a common measurement paradigm. In the following section, we organize these currently used approaches based on the source of information about the resource consumption (subjective, objective, or combined; Br¨unken, Plass, & Leutner, 2003).
measurement of cognitive load The basic question of cognitive load measurement is whether there are valid, reliable, and practical methods to measure cognitive load. Generally, there are two different approaches to assess cognitive load: (1) learners can be asked to rate their perceived cognitive load subjectively, and (2) objective measures of cognitive load, such as physiological measures, can be used. For both approaches, several measures exist and are discussed in the following section. Subjective Measures of Cognitive Load Probably the most common way of measurement in CLT is the use of subjective self-reported rating scales for the assessment of perceived mental effort, often combined with a measure of subjectively perceived task difficulty. The benefits and problems of both approaches are discussed next. Subjective Rating of Perceived Mental Effort In this type of cognitive load measure, learners are asked to rate their perceived cognitive load with items such as “I invested . . . mental effort” on a semantically differential scale varying from “very, very low” to “very, very high” (Paas & van Merri¨enboer, 1993, 1994; Paas, Tuovinen, Tabbers, & Van Gerven, 2003). The scale method is based on the assumption that learners can make a reliable and valid estimation of the amount of load they were confronted with in a specific situation. Several studies using subjective rating scales demonstrate the utility of this method for CLT research. Learners in general report different amounts of perceived load depending on different instructional designs of learning materials (for an overview, see Paas et al., 2003). Several rating scales are used in cognitive load research (Whelan, 2005), most of them using 7- to 9point Likert scales. Moreover, the subjective load rating usually is combined with subjective ratings of variables indirectly related to cognitive load, such
Measuring Cognitive Load
183
as difficulty (see the next section) or fatigue, to form a multidimensional assessment tool. However, the different assessments usually are highly correlated so that a unidimensional scale is also able to assess cognitive load in valid and reliable ways (Paas & van Merri¨enboer, 1994). Perhaps the most obvious benefit of assessing cognitive load with subjective rating scales is its simplicity. However, they also have some serious limitations. First, they usually deliver a one-point post hoc assessment of cognitive load imposed by a learning task or working situation. Usually, learners are asked to assess the cognitive load they experienced after they have finished the learning task. Thus, the resulting assessment is a global scaling across different parts of the learning situation and different tasks. It remains unclear, however, which specific aspects of a learning situation caused the level of cognitive load reported. On the other hand, this does not seem to be a major limitation for using subjective rating scales because they can be applied repeatedly during the learning situation (e.g., using a pop-up window in a multimedia scenario) to get a time series measurement that could be easily synchronized with the learning task or material presented. Yet, to date, there are only a few empirical studies using this approach (e.g., Tabbers, Martens, & van Merri¨enboer, 2004). A second, more serious problem of the subjective rating scales is related to their content validity, namely, to the question of which type of cognitive load the scale measures and how this assessment is related to learning. Although learners are assumed to be able to be introspective about their own cognitive processes and quantify their perceived mental load during learning (Paas et al., 2003), these measures are unable to provide useful information regarding which processes have caused the perceived amount of mental load. Furthermore, it is not possible to conclude which of the three types of load originated the reported mental effort level. Was it caused by extraneous load (i.e., poor instructional design), germane load (i.e., learners’ problem-solving activities), a combination of both, or perhaps simply by the intrinsic load of the material? Currently, only one study from Ayres (2006) analyzed an approach to subjectively rate the intrinsic cognitive load by measuring task difficulty. In his research design, the global rating of difficulty can be clearly ascribed to intrinsic cognitive load because germane load was controlled by using a task that required students to use their existing knowledge rather than to learn new knowledge, and extraneous load was controlled by excluding instructional materials. Nevertheless, the question of how different types of load can also be measured in domains where new information has to be learned is crucial because (as defined in the prior sections of this book) different types of load are related to learning
184
Roland Br¨unken, Tina Seufert, and Fred Paas
in different ways. Hence, the question of how mental load and learning are related needs to be analyzed in more detail in future research. Subjective Rating of Perceived Task Difficulty As mentioned earlier, the cognitive load rating is often combined with a rating of perceived task difficulty (Paas et al., 2003). If the same content is rated as more or less difficult by the learner, depending on the form of its presentation, then this would serve as an additional indicator for the extraneous load concept of CLT. However, task complexity ratings can also be an indicator for the intrinsic load of the material. One could argue that task complexity depends on the amount of interrelated elements (i.e., the element interactivity; see Chapter 2) in a presented concept. Moreover, element interactivity is not only a function of the presented learning materials but also a function of the interaction between learning material and the learner’s prior domain-specific knowledge. Based on these assumptions, we should expect different ratings for task complexity depending on the learner’s level of expertise. Indeed, in our current research, we found evidence for two cognitive load factors (i.e., intrinsic and extraneous) within students’ cognitive load ratings. In a 2-factor experimental design (Seufert, J¨anen, & Br¨unken, 2007) with 142 high school students, we compared the learning results from two presentation modes for two prior-knowledge levels. With respect to the subjective rating of task complexity, an analysis of variance revealed main effects for both factors but no interaction. The results clearly showed two different sources for the task complexity assessment, which, in our view, reflect two different aspects of cognitive load: the intrinsic part (demonstrated by the generally lower scores of the higher knowledge group) and the extraneous part (demonstrated by the differences in the two presentation modes independent of the expertise factor). Overall, the rating of task difficulty comes along with the same benefits and problems as the rating of mental effort, which suggests the need to supplement the subjective measures with additional objective measures discussed in the following sections. Objective Measures of Cognitive Load Besides the subjective measurement approaches, several objective indicators for cognitive load have been addressed in cognitive load research. They can be distinguished by their relation to the learning process in outcome variables (learning outcomes), input variables (task difficulty), and
Measuring Cognitive Load
185
process-related behavioral variables (e.g., physiological parameters or timeon-task). Learning Outcomes as a Measure of Cognitive Load At first sight, the most obvious objective indicator of the level of cognitive load is the learning outcome itself. CLT predicts differences in learning outcomes based on the different amounts of cognitive load induced by a specific situation. Hence, if we find the proposed differences within controlled experimental learning situations, we assume these differences are caused by the differences in cognitive load (e.g., Chapters 7 and 8). Although this was the most common way of argumentation in the early years of cognitive load research, most researchers have by now acknowledged the serious methodological problems of this approach. Assuming that we make no mistakes in our experimental design and procedure, we can explain differences in the dependent variables (learning) by the experimental variation of the independent variable (e.g., the instructional design of the material), but we cannot be sure that the different experimental variations really caused different amounts of cognitive load and did not, at least additionally, cause differences in other variables relevant to learning (i.e., arousal or motivation). Therefore, from a viewpoint of theory validation, we need a manipulation check to ensure that the differences in learning outcomes are indeed caused by differences in cognitive load, as CLT would suggest. Although learning outcomes are the central variables for assessing the fundamental effect of instruction, thus making them indispensable measures in cognitive load research (as in any research on learning and instruction), by themselves they are not valid measures for cognitive load measurement. However, in combination with other cognitive load measures, they can deliver convergent validity for the underlying theoretical assumptions. Task Complexity as a Measure of Cognitive Load Another indicator for cognitive load, and at least implicitly addressed in recent research, is task complexity (which can be defined by the number of interacting elements needed to solve the task) or task difficulty (which usually is defined as the mean probability of task solution). Task complexity affects cognitive processing; the more complex the task, the more cognitive resources needed. In most research on cognitive load, learning outcomes are measured by different tasks, varying in complexity. Usually, these tasks are integrated in scales, for example, labeled “retention,” “problem solving,”
186
Roland Br¨unken, Tina Seufert, and Fred Paas
and “transfer” (Mayer, 2001). The usual pattern found with respect to these variables is that cognitive load effect sizes are stronger as task complexity increases (for a review of modality effects and the moderating effect of task difficulty, see Ginns, 2005). According to CLT, the higher the complexity of the learning task is, the higher the intrinsic load imposed on the learner will be. In the previously mentioned study by Ayres (2006), the complexity, hence, the intrinsic cognitive load, was measured by subjective ratings. However, the exact relation between task complexity and intrinsic load remains unclear. Currently, cognitive load research does not assess intrinsic cognitive load objectively (e.g., by cognitive task analysis). Nevertheless, it seems logical to conclude that if the learning goal is to understand the functionality of a complex technical system, intrinsic load must be higher than in cases where the learning goal is to memorize the names of the system elements. Therefore, it might be more useful to define intrinsic cognitive load not in terms of element interactivity, but in terms of the amount of information that has to be extracted from the information source with respect to a specific learning goal. Behavioral Data as Indicators of Cognitive Load during the Learning Process In contrast to learning outcomes as an outcome-related measure of cognitive load or task difficulty as an input measure, the learners’ behavior is more directly related to the learning process. Thus, several behavioral parameters can serve as an indicator of cognitive load, such as neuro-physiological parameters, invested time-on-task, information retrieval patterns, and resource consumption measured by a dual-task approach. Neuro-physiological measures for cognitive load. Cognitive neuroscience research provides direct, basic measures of resource consumption. The most common technique in this fast-growing field of research is functional magnetic resonance imaging (fMRI), a technique that depicts metabolic activity, for example, in the human brain. It provides qualitative information about the location of activation and can be supplemented with other imaging techniques in which the quantity of activation can be depicted. Based on a research review, Whelan (2007) argues that fMRI studies make it possible to measure the different types of cognitive load specifically. However, because such techniques require a highly sophisticated technological apparatus, they can only be used in very specialized laboratory contexts and are not yet appropriate for typical learning situations. Thus, other physiological parameters can serve as indicators for cognitive load. General psychology
Measuring Cognitive Load
187
approaches suggest a number of potential candidates for this type of measure, such as heart rate, galvanic skin reaction, and eye-tracking behavior. There is a limited body of research that focuses on the issue of cognitive load measurement using behavioral data (for a review, see Paas et al., 2003). The study by Paas and van Merri¨enboer (1994) provides the first example of a physiological method to measure cognitive load within the cognitive load framework. The spectral-analysis technique of heart-rate variability is based on the assumption that changes in cognitive functioning are reflected in physiological functioning. The spectral analysis of heart-rate variability offers a measure of the intensity of mental effort. Spectral analysis is a method for investigating whether a signal contains periodic components. Aasman, Mulder, and Mulder (1987) and Mulder (1992), among others, have validated this technique with several cognitive tasks (e.g., multidimensional classification and sentence comprehension). Although these studies suggest that the 10-Hz component of heart-rate variability exhibits a reliable reflection of mental effort associated with different tasks (Tattersall & Hockey, 1995), the heart-rate variability was not found to be sensitive to the subtle fluctuations in cognitive load that are typically investigated in cognitive load research within or between instructional conditions. In another study, Van Gerven, Paas, van Merri¨enboer, and Schmidt (2004) measured taskevoked pupillary responses as a function of different levels of cognitive load in both young and old participants. They found that mean pupil dilation, as measured by a remote eye-tracking device, is a highly sensitive indicator of tracking fluctuating levels of cognitive load, especially for young adults (for an overview of task-evoked pupillary responses, see Beatty & Lucero-Wagoner, 2000). Although most of the physiological techniques discussed previously are highly reliable, they raise questions of construct validity. For example, although the number and duration of eye movements and fixations on different parts of visually presented information can be observed with high precision by using modern eye-tracking cameras, if differences are found, for example, in the duration of eye fixation, there could be several causes for such a change, such as information complexity, interest, or readability. As we discussed earlier with respect to the subjective ratings of cognitive load, global measures always have the problem of multi-causality: They can be caused by various factors and, at least in experimental settings, it sometimes remains unclear which of the potential causes have been addressed with the experimental variation. Therefore, behavioral data analysis requires extremely careful experimental designs to assure that the measure is not only reliable but also valid.
188
Roland Br¨unken, Tina Seufert, and Fred Paas
With respect to CLT, one could ask, “For which cognitive load effects is eye-tracking observation a useful tool?” Although it might be extremely useful for observing behavior related to the integration of different visual information sources (as addressed in the split-attention effect or the colorcoding effect in Chapters 4 and 7), it might be less useful to obtain cognitive load differences caused by different information modes (such as the multimedia effect) or different information modalities (such as the modality effect). However, used within sensitive experimental settings, the observation of behavioral data can offer interesting and highly relevant results that are closely related to the assumptions of CLT. For example, Folker, Ritter, and Sichelschmidt (2005) compared two variants of a multimedia presentation of learning materials in the domain of cell biology, one with and one without color coding. From a cognitive load perspective, color coding should reduce the imposed extraneous load of the material by giving hints about how to integrate different representations and therefore facilitating coherence formation. Using eye-tracking observation, Folker et al. found different fixation patterns for the two learning conditions, with fewer fixations and fewer changes between the textual and pictorial information in the color-coded condition, which suggests that color coding indeed reduces integration costs. This (relative) reduction of cognitive costs, for example, by fewer fixation changes, can be seen as a direct indicator for different cognitive load caused by the learning materials, which leads to different amounts of knowledge acquisition. Time-on-task as a measure of cognitive load. All cognitive processes take time. The amount of time needed to reach a solution is affected by several factors, including the complexity of the task, the learners’ prior knowledge, the time needed to search information, and so forth. Nevertheless, time-ontask is directly related to cognitive processing and in a good experimental setting, it is possible to control for most of these factors by measuring them or by randomly assigning participants to different conditions. In this way, it is possible to come closer to the basic processing variables causing differences in time-on-task. Recent research by Tabbers et al. (2004) illustrates this issue. In a series of experiments, they found that the modality effect in multimedia learning (Mayer & Moreno, 1998) vanishes when learners are given the opportunity to control the pace of the presentation: Students who learn with text and pictures spend more time-on-task to reach the performance level of students who learn with narration and pictures. These results are highly interesting for basic cognitive load research because they demonstrate that differences in learning outcomes caused by different instructional designs can be compensated for by additional time-on-task. This finding
Measuring Cognitive Load
189
highlights the relation of instructional design and cognitive processing as it is assumed by CLT: Because the verbal material (which is intrinsically equally demanding) is presented in different modalities, differences in learning time are necessary to come to the same learning results. This is strong evidence for what CLT calls extraneous load: the additional consumption of cognitive resources caused by the instructional design without a corresponding learning benefit. However, the theoretical relation between time-on-task and cognitive load remains indirect. It is the timeframe in which cognitive processes take place, and the size of the timeframe that affects the efficiency of cognitive processing. Moreover, little is known about exactly how timeon-task and cognitive load are related. Usually, we would expect a linear relationship (the more time needed, the more the load imposed by the material), but it might also be plausible to argue that very little time-ontask indicates high cognitive load because the load might be so high that the learner stops investing effort on the learning situation. Information retrieval behavior as a measure of cognitive load. The way in which learners search and select information can also serve as an indicator of cognitive load within the learning process. For example, the analysis of navigation pathways in hypertext environments could serve as an indicator for schema construction, because it highlights aspects of individual information search, which is related to the learners’ prior knowledge (M¨oller & M¨uller-Kalthoff, 2000). Imagine a large, well-structured hypertext learning environment containing information about a specific domain, such as a complex medical topic. A novice learner, unfamiliar with the domain, might start reviewing hypertext pages containing basic information. A domain expert, however, might directly navigate to pages containing specific information, for example, about alternative medications or diagnostic strategies. This will result in two different navigation patterns that are typical for the respective level of prior knowledge. Therefore, with regard to a specific task, the navigation pathway – the pages visited as well as the pages skipped – allows for insights into the structure of the underlying domain-specific knowledge of the learner. Other examples could be the amount of helpseeking or the use of optional explanatory annotations (Wallen, Plass, & Br¨unken, 2005), which could indicate differences in the amount of help needed and therefore could be argued to indirectly measure cognitive load. But, again, the relation between these measures and cognitive load is not straightforward. Differences in navigation behavior can result from various causes, and careful experimental designs are needed to control for possible influencing factors not related to cognitive load. Despite the fact that all these measures are easy to administer, are objective, and produce fairly
190
Roland Br¨unken, Tina Seufert, and Fred Paas
reliable scores, especially when using technology-based learning environments, we found few examples for using these cognitive load measurements in the literature (see also Paas et al., 2003). Dual-task measures. Another type of cognitive load measurement based on the analysis of objective behavioral data recently used by Br¨unken et al. (2003) is based on the dual-task paradigm (see also Chandler & Sweller, 1996; Van Gerven, Paas, van Merri¨enboer, & Schmidt, 2006). This approach is based on a simple hypothesis derived from the “limited capacity assumption” of CLT (see Chapters 2 and 3, this volume). If the total cognitive capacity for a certain learner at a certain time is limited in its amount, it has to be distributed to all cognitive processes that have to be carried out at that point in time. If a learner has to process two different tasks at the same time, which in addition, consume his complete capacity, then the performance on the one task can be used as a measure of the capacity consumption of the other task. The assumption of complete resource consumption by the two tasks is crucial for dual-task research in general. The secondary task can only be sensitive and reliable if it is exhaustive (Wickens, 1984). Imagine two groups of students learning the same information from two different learning environments in a dual-task scenario. If the capacity requirements of the two different learning environments vary because of the instructional design, then the performance on a simultaneously processed secondary task should vary accordingly. In a series of studies (Br¨unken, Plass, & Leutner, 2004; Br¨unken, Steinbacher, Plass, & Leutner, 2002), we demonstrated the utility of using this secondary task method as a direct estimation of cognitive load. Other studies confirmed these results (e.g., Renkl, Gruber, Weber, Lerche, & Schweitzer, 2003). Moreover, the results of all of these studies were in line with the predictions of CLT. However, the dual-task method also has some serious limitations. First, it is limited to a comparative measurement of the cognitive load induced by different instructional designs. It fails to provide an absolute estimation of resource consumption. However, it should be noted that this is not only a problem of dual-task measurement. We will discuss this as an important issue of cognitive load measurement in the next section. Second, because of the complexity of the experimental setting of dual-task measures, especially when using reaction times as a secondary task measure, their use is usually limited to laboratory research. The third and perhaps most problematic issue is that, because of the nature of shared cognitive resources in this approach, the secondary task may interfere with or hinder the primary task (which is usually learning). A final point to be addressed in dual-task
Measuring Cognitive Load
191
research is the modality specificity of the secondary task. Our research shows that the consumption of cognitive resources is modality specific, that is, the sensitivity of the secondary task measure depends on the specific modality chosen for the primary and secondary tasks in the learning environment (Br¨unken et al., 2002, 2004). For example, an auditory secondary task might be insensitive for assessing extraneous cognitive load caused by a splitattention effect because this effect is primarily related to visual processing. However, if well designed, the dual-task method is a powerful approach for cognitive load measurement and applicable to the evaluation of learning scenarios (e.g., in multimedia learning) and the usability of software in general. Combined (Efficiency) Measures of Cognitive Load A well-established approach to model the relation between mental effort and performance has been introduced by Paas and van Merri¨enboer (1993, see also Paas et al., 2003). Imagine two groups of students learning about the same content material but with different instructional designs: The first group reports a lower amount of invested mental effort, but the second group outperforms the first one with respect to knowledge acquisition. Which instructional variant should be favored? Such a question cannot be answered easily because the answer may depend on the concrete learning task as well as on the available alternatives. For example, for a task of monitoring critical technical systems, in which the prevention of errors is crucial, the imposed cognitive load might be less important because users have to be able to prevent errors even under heavy load conditions. For other tasks, such as introducing new knowledge to novice learners, it might be more beneficial to avoid cognitive load that is too high. To resolve this ambiguous relation, Paas and van Merri¨enboer (1993) proposed an approach that combines the two dimensions of mental effort and performance. The method consists of first transforming both variables to a comparable scale by normalizing the respective scores and then calculating the relationship between mental effort and performance using a simple equation: the difference between the standardized scores of performance and mental effort divided by the square root of two. This calculation yields an efficiency score, which indicates high instructional efficiency when the perceived cognitive load of a specific instructional design is lower than should be expected by the performance results or low instructional efficiency when the perceived level of cognitive load is higher than should be expected by the performance results.
192
Roland Br¨unken, Tina Seufert, and Fred Paas
Recently, Tuovinen and Paas (2004) introduced an extension of the original two-dimensional instructional efficiency measure. This new threedimensional approach can either combine two measures of effort (i.e., learning effort and test effort) and test performance, or one effort measure and one time measure, and test performance. Each of these approaches, with their associated insights and analyses, may be useful for instructional researchers as diagnostic instruments to identify different aspects of efficient or inefficient instructional conditions. The efficiency approach is appealing because it is easy to calculate (see Paas et al., 2003; van Gog & Paas, 2008) and independent from the specific way of measuring load or performance. Moreover, it can easily be adapted to situations in which one of the dimensions is more important than the other by including weight terms in the formula. However, the approach is limited to the comparison of different instructional variants of the same material and provides only a measure of the relative efficiency of the investigated variants. Each time a new variant has to be tested, the efficiency has to be recalculated, and the relationships among the variants might change. Moreover, one could question (Paas et al., 2003) whether efficiency is only affected by cognitive load and performance or if other learning-related variables, such as time-on-task or motivation, should also be included in the calculation. Although the issue of conceptualizing efficiency is highly important, it does not challenge the principle of efficiency calculation. Conceptually, all these variables could (a valid and reliable measurement presumed) be included in an efficiency formula. This was recently recognized by Paas, Tuovinen, van Merri¨enboer, and Darabi (2005), who introduced a similar procedure to combine mental effort and performance to compute the differential effects of instructional conditions on learner motivation. From a theoretical perspective, the discussion about instructional efficiency highlights a more general aspect of cognitive load measurement: its relation to the theoretical concepts of CLT and the impact of various internal and external factors on cognitive load. We address this point in the next part of the chapter. Table 9.1 summarizes the classification of cognitive load measures we just reviewed. To summarize, most recent empirical studies on CLT incorporate cognitive load measures, indicating that next to learning performance measures, they are the second method used to assess the effectiveness of instruction. However, although most of the studies include subjective rating scales for practical reasons, other, more objective methods are of increasing interest, especially when they can be incorporated into computer-based
193
Measuring Cognitive Load table 9.1. Classification of cognitive load measures Type of cognitive load measure
Measure
Subjective
Subjective rating scales
Objective
Learning outcomes
Time-on-task Navigation behavior, helpseeking behavior Task complexity
Behavioral data (heart rate, pupil dilation) Secondary task analysis Eye-tracking analysis
Combined
Efficiency measures
Main research question Learner’s subjective assessment of task demands Relation between instructional design and knowledge acquisition Learner’s investment in the learning process Learner’s information need
Relation among affordances, instructional design, and knowledge acquisition Global or specific physiological reactions of the organism involved in a learning process Mental load induced by the (primary) learning task Basic behavioral aspects of information processing and their relation to learning outcomes Optimizing instructional design decisions by calculating the relation of invested effort and learning outcome
Research examples Paas & van Merri¨enboer, 1993 Mayer, 2005; Mayer & Moreno, 1998 Tabbers et al., 2004 M¨oller & M¨ullerKalthoff, 2000 Seufert et al., 2007
Van Gerven et al., 2004
Br¨unken et al., 2002 Folker et al., 2005
Paas et al., 2003
environments for learning, such as multimedia or simulation-based learning. In addition, efficiency calculations are of growing interest, especially for guiding instructional design decisions.
discussion: open questions and future directions In the previous sections, we showed that cognitive load can be measured by various methods more or less easily by using subjective ratings as well
194
Roland Br¨unken, Tina Seufert, and Fred Paas
as by conducting objective observations of learners’ behavior. In our view, it is not useful to ask which type of measurement is the best one because they vary in their objectives as well as in their practicality. They all have benefits and disadvantages, and the decision regarding which one to use highly depends on the specific research questions raised. Open Questions in Cognitive Load Measurement Going back to the initial theoretical concepts of CLT, we will try to reconceptualize cognitive load from a measurement point of view. In this view, cognitive load is related to all processes performed in working memory independent of their relation to the task. Cognitive load simply means that something non-automatic happens in the mind, which causes the consumption of mental resources. One might argue that this definition is trivial, but with respect to the limited nature of mental resources, it is crucial whether the activity performed in the mind is actually related to a specific goal or not. Following are some of the current challenges to the measurement of cognitive load during learning. Cognitive Load Is Ambiguous with Respect to Its Learning Outcomes CLT proposes that cognitive load can foster learning as well as hinder it. An aspect directly related to learning is the task orientation of the actual cognitive activity. So in our view, cognitive load is related to learning by differentiating between task-relevant (germane) and task-irrelevant (extraneous) consumption of cognitive resources. However, both types of load consume the same resources, which causes serious measurement problems, because it is not possible to predict the impact of cognitive load on learning without distinguishing between the two types of load. Cognitive Load Depends on the Learner’s Characteristics Coming to a more detailed concept of cognitive load, which could guide us to more sensitive measures, we are confronted with a second characteristic of cognitive load that causes serious measurement problems – its relativity. Cognitive load is not only ambiguous with respect to learning outcomes but it is also relative in several ways. First, and maybe most obvious, cognitive load depends on the characteristics of the task. The more complex a task, the more cognitive capacity is necessary to solve it. This aspect is integrated
Measuring Cognitive Load
195
in CLT by the concept of intrinsic load. Usually, intrinsic load is conceptualized by element interactivity, which is an indicator of task complexity based on the number of conceptual elements that have to be held in the mind simultaneously to solve a specific task (e.g., to understand a complex concept). However, this definition is questionable because it assumes that intrinsic load only depends on task characteristics. Analyzing this concept in more detail, we will find, for example, that the differences in learning success might also depend on the learner’s prior domain-specific knowledge. But what exactly is prior knowledge? In terms of CLT, we can define prior knowledge as the amount of task-relevant elements already available in the learner’s mind. So the intrinsic load of a specific learning situation not only depends on the number of elements needed for understanding that are presented by the material, but also on the number of externally presented elements needed for understanding and not already available in the learner’s mind. The fact that intrinsic load depends on the learner’s prior knowledge suggests that cognitive load is also dependent on learner aptitudes (Seufert, J¨anen, & Br¨unken, 2007). Maybe the most compelling evidence for this idea is the expertise reversal effect (Kalyuga, Ayres, Chandler, & Sweller, 2003). Similarly, other learner aptitudes may be seen as influential to cognitive load, such as students’ memory span or speed of information processing, as well as specific cognitive abilities, such as verbal or spatial abilities. Several empirical studies support this idea (e.g., Mayer & Sims, 1994; Plass, Chun, Mayer, & Leutner, 1996; Wallen et al., 2005). Paas, Camp, and Rikers (2001; see also Paas, Van Gerven, & Tabbers, 2005) showed that the cognitive load and learning effects resulting from different instructional designs are also highly dependent on learner age. According to cognitive aging research, people experience declines in working memory capacity, processing speed, inhibition, and integration as they age. Consequently, working memory processing becomes less efficient for older adults. Studying the differential effects of goal specificity on maze learning between younger (20 years old) and older (65 years old) adults, Paas and colleagues (2001) showed that instruction designed according to CLT principles compensates for these age-related cognitive declines. More specifically, the researchers found that the absence of a specific goal disproportionately enhanced the elderly participants’ learning and transfer performance, almost up to the level of the younger adults. It is becoming clearer from these results that learner aptitudes and individual differences should be taken into consideration not only when trying to define the concept of intrinsic load, but also when trying to measure CL.
196
Roland Br¨unken, Tina Seufert, and Fred Paas
Cognitive Load Is Not Constant during Learning In addition, our engagement in learning tasks is not always constant but varies depending on other affective factors, such as motivation and selfconcept. To date, neither theoretical nor empirical cognitive load developments take these factors into account. This may be one of the main shortcomings of the present state of the art in CLT, which is only based on the analysis of cognitive factors of learning. Nevertheless, in terms of cognitive load measurement, we can assume that cognitive engagement is the total amount of cognitive resources that can be allocated to a specific task in a specific situation. Last but not least, cognitive load also varies, both with respect to the imposed load by the learning situation and the resources available (for a conceptual discussion of this point, see Paas et al., 2003). The transient nature of cognitive load should be taken into account for measurement purposes. It should be clear from this review that whichever type of measurement is used, cognitive load cannot be seen as a constant factor that only depends on objective attributes of the learning material or design. Cognitive load measurement is relative, and the efficiency of a learning situation has to be seen in light of relevant personal and environmental variables. However, measurement of cognitive load can be more or less sensitive to several of the discussed aspects. For instance, the impact of load distributed over time can be investigated by a repeated real-time measurement rather than by a one-time, ex post assessment. The role of learner aptitudes can be studied in experimental aptitude-treatment-interaction analyses (Cronbach & Snow, 1977) as well as by correlation analyses in field experiments. Moreover, we do not only need more and better experimental studies on the relation of cognitive load and learner characteristics, but we also have to adapt existing measures to capture these potential interactions. A critical issue that is currently under discussion is the relation of cognitive load and the perceived amount of invested mental effort, which either can be a result of differences in knowledge, abilities, motivation (see Paas et al., 2005), or time-on-task. Future Directions in Cognitive Load Measurement Research in the area of cognitive load measurement methods is currently concerned with two additional issues, highly relevant to both the theory and measurement of cognitive load: (1) the differential measurement of the three sources of cognitive load, and (2) the additivity assumption of cognitive load.
Measuring Cognitive Load
197
The Differential Measurement of Intrinsic, Extraneous, and Germane Loads CLT proposes three different sources of cognitive load – intrinsic, extraneous, and germane – differing in their origin and their relation to performance. However, at the moment, there is no measure available to empirically distinguish between these three load types. Coming to a more precise description of the relation between these loads and learning performance may be currently the most important challenge for measurement research. There is surely more than one definition of this relation in use by cognitive load researchers, depending on the chosen measurement approach, but there are some general requirements that should be addressed by each method. For example, different types of load cannot be measured simultaneously on a one-dimensional scale or by a one-dimensional value. This requires either a multidimensional measurement approach or a sequential or isolated measurement of the different load types. A multidimensional approach could most easily be employed by using multiple subjective rating scales, each of them related to one type of load. For instance, “How easy or difficult was it for you to work with the learning environment?” could be used as a question specifically related to extraneous cognitive load, or “How easy or difficult do you consider the specific learning content ‘x’ at this moment?” as an item related to intrinsic load, or “How easy or difficult was it to understand the specific concept ‘y’?” as a question to assess the germane load. Although the dimensional structure of such a questionnaire can be evaluated using factor analysis, a more fundamental question is whether learners are able to distinguish the contribution of each cognitive load type during learning. For example, can learners assess the amount of germane load in valid and reliable ways? The experiences with self-assessment in the domain of learning style research are not very encouraging (Leutner & Plass, 1998). Although it seems to be possible to use questionnaires to assess the amount of learning strategy use (Wild & Schiefele, 1994), it seems to be much more problematic to assess the quality of strategy use. This problem of lacking metacognitive reflection is usually reflected in low or zero correlations between self-reported strategy use and performance (Leutner & Plass, 1998). However, the use of learning strategies can be seen in terms of CLT as cognitive activities, which cause germane load and which may lead to similar problems with the assessment of germane activities. A possible way to ease introspection might be to refer closely to concrete activities directly related to the information to be learned instead of referring
198
Roland Br¨unken, Tina Seufert, and Fred Paas
to the use of global strategies. For example, the question “Did you underline the basic concept ‘x’ in the presented material?” can be used instead of the question “How frequently did you use underlining?” Similarly, to assess the intrinsic load with respect to the individual’s expertise level, it seems more useful to ask students to assess the perceived difficulty of a concrete concept than the perceived difficulty of the overall learning experience. This would demand a repeated measurement of cognitive load during learning. The continuous observation of cognitive demands in a learning situation could, at least for research objectives, be more easily accomplished using appropriate objective measurement techniques, such as dual-task analysis or eye movement tracking. However, this method requires the development of better, less invasive measures as well as the application of adequate statistical methods, such as time series analysis, which have not been used in cognitive load research to date. The Additivity Hypothesis A second issue not addressed in current research on cognitive load measurement is related to a fundamental but empirically unproven assumption of CLT: the additivity hypothesis. According to this hypothesis, the different sources of load (intrinsic, extraneous, and germane) are additive elements of the total cognitive load imposed by a specific learning situation. This assumption is crucial because it implies that a reduction/increase in one type of load automatically produces an identical reduction/increase in total amount of cognitive load. For example, reductions of intrinsic cognitive load should free up cognitive resources that can be used to engage in other cognitive activities, such as germane learning processes (Pollock, Chandler, & Sweller, 2002; see also Chapter 2). However, recent research shows that this is not always the case (Seufert & Br¨unken, 2006). Specifically, a reduction of extraneous load does not always lead to increased learning. Moreover, the additivity assumption seems questionable from a theoretical point of view. First, a large body of research is concerned with the phenomenon called “illusion of understanding,” in which learners stop knowledge acquisition before they fully understand a concept because they overestimate their level of understanding (Glenberg, Wilkinson, & Epstein, 1982; Renkl & Atkinson, 2003). What triggers this illusion? A possible explanation is that instructional design with low extraneous load may be perceived as “an easy task” and therefore may result in cognitive disengagement, an effect also shown in Salomon’s early research on media effects (Salomon, 1984). However, a complex task with high extraneous load could be perceived as
Measuring Cognitive Load
199
challenging, leading the learner to invest more mental effort in learning. Second, the assumption that intrinsic and extraneous loads are independent seems problematic. Is a specific instructional design for learners of different expertise levels equally “extraneously loading”? The expertise reversal effect suggests that this is not the case (Kalyuga et al., 2003). Whereas some instructional material seems to be beneficial for novice learners, the same material is redundant for experts, resulting in decreased knowledge acquisition. All these questions concerning the additivity hypothesis require groundbreaking experimental research on the relation of the different load types that in turn need differential measures of cognitive load. To summarize, cognitive load measurement is still in its infancy. Most currently available cognitive load measures are less than perfect measures of students’ overall cognitive load. However, even these imprecise measures contribute to CLT by providing empirical evidence for the global relation of instructional design, cognitive processes, and performance. The development of better measures will therefore be beneficial for CLT in general, as well as for its application to the instructional design of learning materials. Nevertheless, three basic aspects of cognitive load have to be taken into account, both to optimize measurement techniques as well as to clarify the basic concepts of CLT: (1) the relative nature of cognitive load, (2) the dependency among the three cognitive load types, and (3) the dimensionality of cognitive load measurement. As we shed more light on these crucial issues, we should be able to bridge the gap between CLT and research. references Aasman, J., Mulder, G., & Mulder, L. J. M. (1987). Operator effort and the measurement of heart-rate variability. Human Factors, 29, 161–170. Ayres, P. (2006). Using subjective measures to detect variations of intrinsic cognitive load within problems. Learning & Instruction, 16, 389–400. Beatty, J., & Lucero-Wagoner, B. (2000). The pupillary system. In J. T. Cacioppo, L. G. Tassinary, & G. G. Berntson (Eds.), Handbook of psychophysiology (2nd ed., pp. 142–162). Cambridge, MA: Cambridge University Press. Br¨unken, R., Plass, J. L., & Leutner, D. (2003). Direct measurement of cognitive load in multimedia learning. Educational Psychologist, 38, 53–61. Br¨unken, R., Plass, J. L., & Leutner, D. (2004). Assessment of cognitive load in multimedia learning with dual-task methodology: Auditory load and modality effects. Instructional Science, 32, 115–132. Br¨unken, R., Steinbacher, S., Plass, J. L., & Leutner, D. (2002). Assessment of cognitive load within multimedia learning by the dual task methodology. Experimental Psychology, 49, 109–119.
200
Roland Br¨unken, Tina Seufert, and Fred Paas
Chandler, P., & Sweller, J. (1996). Cognitive load while learning to use a computer program. Applied Cognitive Psychology, 10, 151–170. Cronbach, L. J., & Snow, R. E. (1977). Aptitudes and instructional methods. New York: Irvington. Folker, S., Ritter, H., & Sichelschmidt, L. (2005). Processing and integrating multimodal material. The influence of color-coding. In B. G. Bara, L. Barsalou, & M. Bucciarelli (Eds.), Proceedings of the 27th annual conference of the Cognitive Science Society 2005 (pp. S. 690–695). July 21–23. Stresa, Italy. Ginns, P. (2005). Meta-analysis of the modality effect. Learning and Instruction, 15, 313–331. Glenberg, A. M., Wilkinson, A. C., & Epstein, W. (1982). The illusion of knowing: Failure in the self-assessment of comprehension. Memory & Cognition, 10, 597– 602. Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). The expertise reversal effect. Educational Psychologist, 38, 23–31. Leutner, D., & Plass, J. L. (1998). Measuring learning styles with questionnaires versus direct observation of preferential choice behavior in authentic learning situations: The visualizer/verbalizer behavior observation scale (vv-bos). Computers in Human Behavior, 14, 543–557. Mayer, R. E. (2001). Multimedia learning. Cambridge, UK: Cambridge University Press. Mayer, R. E. (2005). (Ed.). The Cambridge handbook of multimedia learning. Cambridge UK: Cambridge University Press. Mayer, R. E., & Moreno, R. (1998). A split-attention effect in multimedia learning: Evidence for dual processing systems in working memory. Journal of Educational Psychology, 90, 312–320. Mayer, R., & Sims, V. (1994). For whom is a picture worth a thousand words? Journal of Educational Psychology, 86, 389–401. M¨oller, J., & M¨uller-Kalthoff, T. (2000). Lernen mit Hypertext: Effekte von Navigationshilfen und Vorwissen [Learning with hypertext: The impact of navigational aids and prior knowledge]. Zeitschrift f¨ur P¨adagogische Psychologie [German Journal of Educational Psychology], 14, 116–123. Mulder, L. J. M. (1992). Measurement and analysis methods of heart rate and respiration for use in applied environments. Biological Psychology, 34, 205–236. Paas, F., Camp, G., & Rikers, R. (2001). Instructional compensation for age-related cognitive declines: Effects of goal specificity in maze learning. Journal of Educational Psychology, 93, 181–186. Paas, F., Tuovinen, J., Tabbers, H., & Van Gerven, P. W. M. (2003). Cognitive load measurement as a means to advance cognitive load theory. Educational Psychologist, 38, 63–71. Paas, F., Tuovinen, J., van Merri¨enboer, J. J. G., & Darabi, A. (2005). A motivational perspective on the relation between mental effort and performance: Optimizing learner involvement in instruction. Educational Technology, Research & Development, 53, 25–33. Paas, F., Van Gerven, P. W. M., & Tabbers, H. K. (2005). The cognitive aging principle in multimedia learning. In R. Mayer (Ed.), Cambridge handbook of multimedia learning (pp. 339–354). New York: Cambridge University Press.
Measuring Cognitive Load
201
Paas, F., & van Merri¨enboer, J. J. G. (1993). The efficiency of instructional conditions: An approach to combine mental-effort and performance measures. Human Factors, 35, 737–743. Paas, F., & van Merri¨enboer, J. J. G. (1994). Variability of worked examples and transfer of geometrical problem-solving skills: A cognitive-load approach. Journal of Educational Psychology, 86, 122–133. Plass, J. L., Chun, D. M., Mayer, R. E., & Leutner, D. (1998). Supporting visual and verbal learning preferences in a second language multimedia learning environment. Journal of Educational Psychology, 90, 25–35. Pollock, E., Chandler, P., & Sweller, J. (2002). Assimilating complex information. Learning and Instruction, 12, 61–86. Renkl, A., & Atkinson, R. K. (2003). Structuring the transition from example study to problem solving in cognitive skills acquisition: A cognitive load perspective. Educational Psychologist, 38, 15–22. Renkl, A., Gruber, H., Weber, S., Lerche, T., & Schweitzer, K. (2003). Cognitive Load beim Lernen aus L¨osungsbeispielen [Cognitive load during learning from worked-out examples]. Zeitschrift f¨ur P¨adagogische Psychologie [German Journal of Educational Psychology], 17, 93–101. Salomon, G. (1984). Computers in education: Setting a research agenda. Educational Technology, 24, 7–11. Seufert, T., & Br¨unken, R. (2006). Cognitive load and the format of instructional aids for coherence formation. Applied Cognitive Psychology, 20, 321–331. Seufert, T., J¨anen, I., & Br¨unken R. (2007). The impact of intrinsic cognitive load on the effectiveness of graphical help for coherence formation. Computers in Human Behavior, 23, 1055–1071. Tabbers, H. K., Martens, R. L., & van Merri¨enboer, J. J. G. (2004). Multimedia instructions and cognitive load theory: Effects of modality and cueing. British Journal of Educational Psychology, 74, 71–81. Tattersall, A. J., & Hockey, G. R. (1995). Level of operator control and changes in heart rate variability during simulated flight maintenance. Human Factors, 37, 682–698. Tuovinen, J., & Paas, F. (2004). Exploring multidimensional approaches to the efficiency of instructional conditions. Instructional Science, 32, 133–152. Van Gerven, P. W. M., Paas, F., van Merri¨enboer, J. J. G., & Schmidt, H. G. (2004). Memory load and task-evoked pupillary responses in aging. Psychophysiology, 41, 167–175. Van Gerven, P. W. M., Paas, F., van Merri¨enboer, J. J. G., & Schmidt, H. G. (2006). Modality and variability as factors in training the elderly. Applied Cognitive Psychology 20(3), 311–320. van Gog, T., & Paas, F. (2008). Instructional efficiency: Revisiting the original construct in educational research. Educational Psychologist, 43, 16–26. Wallen, E., Plass, J. L., & Br¨unken, R. (2005). The function of annotations in the comprehension of scientific texts – Cognitive load effects and the impact of verbal ability. Educational Technology, Research & Development, 53, 59–72. Whelan, R. R. (2005). The multimedia mind: Measuring cognitive load in multimedia learning. Unpublished doctoral dissertation. New York University, New York.
202
Roland Br¨unken, Tina Seufert, and Fred Paas
Whelan, R. R. (2007). Neuroimaging of cognitive load in instructional multimedia. Educational Research Review, 2, 1–12. Wickens, C. D. (1984). Processing resources in attention. In R. Parasuraman & D. R. Davies (Eds.), Varieties of attention (pp. 63–102). London: Academic Press. Wild, K., & Schiefele, U. (1994). Lernstrategien im Studium: Ergebnisse zur Faktorenstruktur und Reliabilit¨at eines neuen Fragebogens [Learning strategies during studies: Results on factoral structure and reliability of a new questionnaire]. Zeitschrift f¨ur Differentielle und Diagnostische Psychologie [German Journal of Differential and Diagnostical Psychology], 15, 185–200.
10 From Neo-Behaviorism to Neuroscience: Perspectives on the Origins and Future Contributions of Cognitive Load Research richard e. clark and vincent p. clark historical perspectives on cognitive load research and theory European and American psychology may have developed in a way that prevented or delayed the development of Cognitive Load Theory (CLT) until George Miller’s (1956) classic paper on working memory capacity appeared a half century ago. At the beginning of the twentieth century and fifty years before Miller’s paper kick-started the field of cognitive science, Charles Hubbard Judd (1908) lost an important argument with Edward Thorndike (1903) about the role of mental effort in the transfer of learning. The loss helped to sidetrack psychology into emphasizing behaviorism over cognitive processing. Judd, an American who was Wilhelm Wundt’s student in Leipzig at the end of the nineteenth century, hypothesized that internal cognitive processes and external instructional strategies supported the mental work necessary to transfer knowledge between different problem contexts and settings. Judd had learned from Wundt to emphasize a version of scientific psychology that favored the study of consciousness, problem solving, thinking, and sensations. Judd’s (1908) famous bow and arrow experiment demonstrated that effortful cognitive processes could support the generalization of a principle about the diffractive properties of water and so allow people to adjust their aim with the bow to hit an underwater target The authors want to acknowledge their debt to a number of colleagues who reviewed previous drafts of this chapter and gave advice, including Mary Helen Immordino-Yang, Robert Rueda, and the three editors of this volume. Any errors that remain are our responsibility. The first author wants to acknowledge that his contribution to this chapter has been partially sponsored by the U.S. Army Research, Development, and Engineering Command (RDECOM). Statements and opinions expressed do not necessarily reflect the position or the policy of the U.S. Government, and no official endorsement should be inferred.
203
204
Richard E. Clark and Vincent P. Clark
that appeared to be somewhere else. Thorndike (1903) focused his research on animal maze learning and proposed an “identical elements” transfer theory, arguing that it was positive reinforcement that led to learning and transfer – and not cognitive processing. Because Thorndike was a student of the powerful William James, who supported his work, Judd’s theory and evidence were largely ignored. William James’s support for Thorndike’s view of transfer marked a turning point in psychology. James’s earlier work had emphasized the role of mental effort in cognition when, for example, he described attention as “the taking possession by the mind, in clear and vivid form, of one out of what seem several simultaneously possible objects or trains of thought. Focalization and concentration of consciousness are of its essence. It implies withdrawal from some things in order to deal effectively with others” (James, 1890, pp. 403–404). In an 1898 lecture that mirrors some of the arguments made recently about the possible evolutionary selection advantage offered by limitations on working memory by John Sweller (Chapter 2), James gave a series of lectures at Johns Hopkins University in which he claimed that consciousness had an evolutionary function or it would not have been naturally selected in humans. A few years later, James (1904) reversed himself and expressed strong misgivings in an article titled “Does ‘Consciousness’ Exist?” Judd (1910) later protested and argued for a selection bias for consciousness, but at the same time, Thorndike (1903) and others were more successfully arguing that learning was “not insightful” but instead was incremental. Thorndike’s claim essentially denied any important role for consciousness or working memory in learning or problem solving. A number of historians have proposed that the transition in psychology during James and Thorndike’s era was due in large measure to an increasing interest by the American public in the development of the physical and biological sciences and a distrust of the introspective approach in philosophy and imprecise psychological research methods. This may have been the reason that American psychologists such as James, Thorndike, and others at that time were attracted to the learning research of 1904 Nobel Prize winner Ivan Pavlov and supported the use of animal experiments and the careful control of observable and measurable events favored in medical research. This exclusive focus on animal learning and connectionism was not reflected in European psychology, where researchers continued to be concerned with experimental work as well as introspection, Gestalt studies of consciousness, physiology, experimentation, and case study methods. The more flexible approach taken by European researchers may be the reason many of the prime movers in CLT have been trained in the
From Neo-Behaviorism to Neuroscience
205
European psychological tradition. The irony is that behaviorism resulted in important advances in measurement, the specification of instructional method variables, and precise experimental methods while it discouraged hypotheses based on cognitive processing during learning and transfer. It also became increasingly obvious that behaviorism focused primarily on motivation to learn through reinforcement and emphasized very simple forms of learning. That recognition eventually made it possible for neobehaviorists to hypothesize internal cognitive processes to explain complex learning. One of the very early attempts to deal with complaints that behaviorism only focused on simple learning tasks was the neo-behaviorist research on complexity by Canadian psychologist Daniel Berline (1960). In the 1960s, information processing theory was developing, and Berline offered a model for representing cognitive stimulus and response bonds to describe the cognitive processing required for handling uncertainty and novelty. He proposed a method of measuring individual uncertainty about any stimulus and hypotheses that guided research on the relationship of problem uncertainty and learning. His internationalism and his neo-behaviorist theories made early attempts at cognitive science more acceptable to behaviorists in North America. During this time, cognitive science was developing slowly, forced to swim upstream against powerful behaviorists who resisted change. In addition to Miller’s (1952) classic “Magical Number Seven” article, Ulric Neisser’s (1967) book Cognitive Psychology also had a major impact on the development of CLT. Neisser proposed a computer processing metaphor for cognition and urged psychologists to study the function of working memory in daily activities. Although many cognitive psychologists now avoid the restrictive computer metaphor for cognition, educational psychology benefitted from the analogy during a formative stage. A decade after Neisser’s book was published, the article by Schneider and Shiffrin (1977) on controlled and automated processing had a huge impact on our view of complex learning, memory, and problem solving. With these events in the background, a decade later, John Sweller’s (1988) article in Cognitive Science laid the groundwork for CLT. An important lesson to be learned from the history of psychology is that education and psychology must permit more diversity in theoretical and methodological approaches. With a more interdisciplinary approach, we might have started to develop CLT a half-century earlier and so would have been considerably more advanced at this point. Yet, it may also be the case that one of the benefits of the historical delay caused by the dominance of behavioral theories was the development of a clear focus on pragmatic
206
Richard E. Clark and Vincent P. Clark
instructional research. Behaviorists such as B. F. Skinner encouraged psychologists to conduct careful instructional research in schools. CLT researchers have retained the behavioral focus on instruction and as a result, CLT has made significant contributions to instructional design.
clt contributions and challenges to instructional design An emphasis on the application of research findings to instruction requires that we understand the conditions necessary for selecting and implementing the most efficient and effective instructional design for different learning tasks, learners, and delivery media. This decision has worked to the benefit of instructional design in at least two ways. First, we are no longer inclined to make quick inferences about how to support learning by reasoning from a descriptive theory of learning or from empirical studies unsupported by theoretical insights. Learning can accurately be described as a process in which people construct new knowledge by drawing on their prior experience and blending it with new information about a task (Mayer, 2004). We also have clear evidence that asking students to construct what they must learn without guidance is consistently less effective and efficient than worked examples that demonstrate how to perform a task or solve a problem (Mayer, 2004; Kirschner, Sweller, & Clark, 2006). CLT accurately predicts that when students are required to construct or discover how to solve problems or perform complex tasks, the cognitive effort most often overloads working memory and inhibits learning for students who have novice to intermediate levels of germane prior knowledge. Most of the chapters in this book and the research on the use of CLT for instructional design that preceded this book are clearly focused on helping those who design, develop, and present all types of instruction to learners at every age and level of expertise. Recently published examples are Richard Mayer’s (2001, 2005) edited handbooks on multimedia design, his book with Ruth Colvin Clark (Clark & Mayer, 2007) on designing e-learning instruction, and the systematic instructional design strategy for teaching complex knowledge published by Jeroen van Merri¨enboer and Paul Kirschner (2007). These developments can be viewed as attempts to use CLT to identify the many ways that common instructional practices cause overload and suggest concrete and systematic ways to avoid them. Because many of the researchers who are committed to CLT development are also interested in instructional design, some of the most important educational contributions serve to define and clarify the role of instructional methods.
From Neo-Behaviorism to Neuroscience
207
CLT and Instructional Methods Another important advantage of the behaviorism that preceded the development of CLT may be CLT researchers’ adaptation of the goal to provide specific, evidence-based operational definitions of “instructional methods” and welcome explanations of how different methods serve to maximize germane cognitive load and so lead to more learning. Most instructional design systems suggest that those who are developing instruction should “select appropriate instructional methods” without providing adequate guidance about the definition, design, or selection of effective methods. When a young cognitive science was developing in the early 1970s, Lee Shulman famously complained that an obsessive emphasis on aptitude in learning theories had led to the situation in which instructional methods “are likely to remain an empty phrase as long as we measure aptitudes with micrometers and instructional methods with divining rods” (Shulman, 1970, p. 374). Cronbach and Snow (1977) reviewed all instructional research conducted for approximately four decades and recommended that we invest much more emphasis on understanding instructional methods. Until CLT, our failure to focus adequate attention on the specification and presumed cognitive function of instructional methods continued to be one of the most embarrassing failures of instructional psychology. Instructional experiments typically employ treatments described as lectures, discussion, collaborative groups, graphic organizers, case studies, computer programs, and video and text materials. None of these descriptions (and often their accompanying operational definitions in research reports) are focused on the “active ingredients” in the instruction that may or may not have led to measured differences in learning outcomes (Clark & Estes, 1999; Clark, 2001). CLT’s emphasis on elements of instructional methods that are germane and so contribute to learning and those that are extraneous and so distract and inhibit learning is a huge contribution to instructional psychology. Examples of methods suggested by CLT to support novice learners include formatting instructional content in focused, integrated pictorial and narrative presentations of topics (Chapters 3 and 7, this volume) and providing demonstrations of how to perform tasks or solve problems in “worked examples” (Chapter 5). CLT research provides strong indications that these methods maximize the processing time in working memory for task information that must be elaborated and stored in long-term memory while they minimize the extraneous cognitive effort required to support learning. CLT advocates also suggest that these methods provide effective
208
Richard E. Clark and Vincent P. Clark
support for the limited executive learning functions available to learners with less prior knowledge (Chapter 2). The explanation for the benefits of these CLT instructional methods helps to explain the half-century of research that demonstrates the failure of discovery, problem-based, inquiry, and constructivist learning (Kirschner et al., 2006). Challenges to CLT-Inspired Instructional Design CLT has developed rapidly but like any theory, there are many unanswered questions and a number of areas in which current theoretical explanations and measures are inadequate. In the next section of this chapter, we review two urgent issues and examine the possible contributions we could expect from reconsidering the importance of biological, physiological, and neuroscience research. Two important problems that must be addressed before we can advance much further with CLT are that we have not yet found an unobtrusive and reliable way to measure cognitive load and we need to determine whether any specific source of cognitive load is productive for individual learners during instruction. Measuring cognitive load during learning. Gross measures of mental workload, such as self-report and secondary tasks (Megaw, 2005), have been challenged (Gimino, 2000). Self-report measures appear to be confounded with personal judgments about the difficulty of a task rather than the amount of mental effort invested. Secondary measures capture the time required for individual learners to react to a random interruption during a task. These latency measures divert learners’ attention from tasks and introduce a variety of messy confounds (see a review by Iqbal, Adamczyk, Zheng, & Bailey, 2005). Br¨unken, Seufert, and Paas (Chapter 9) discuss different solutions and conclude, “cognitive load measurement is still in its infancy” (p. 199). Past attempts to provide a definition of cognitive load in an educational context have focused either on the number of steps and/or interactions between steps required to perform a task – most often called “intrinsic” load (Sweller, 2006) – or on the mental workload experienced by individuals who are learning. One often-repeated example of the difference between low and high levels of intrinsic load is the difference between learning vocabulary in a foreign language and the presumably higher load required to learn to speak a foreign language (Sweller, 2006). Yet, there have been arguments that the construct of intrinsic load may be an unnecessary and distracting return to the behaviorist emphasis on the environment and the directly observable (Clark & Elen, 2006). Is load in the environment or is it a function of the amount of mental work necessary for any individual
From Neo-Behaviorism to Neuroscience
209
learner to accomplish a task depending on individual differences in prior expertise – or some combination of the two factors? Most definitions of cognitive load emphasize the non-automated cognitive operations that must be assembled by any given individual to complete the task (Clark & Elen, 2006; Clark, Howard, & Early, 2006; Lohman, 1989; Salomon, 1983; Snow, 1996). We could expect huge individual differences in cognitive load for any task depending on the amount of automated prior knowledge any one individual brings to the task. Br¨unken, Seufert, and Paas (Chapter 9, this volume) suggest that a learner’s prior knowledge influences load and also that we do not have adequate measures of automated prior learning. We propose that more effort be invested in exploring physiological measures of mental workload to identify the amount of automated knowledge learners bring to instruction and to reliably quantify the mental effort they must invest to achieve a unit of learning. Prior knowledge and germane cognitive load. From a cognitive perspective, the working load experienced during any task is determined in part (and perhaps entirely) by an individual’s prior experience with the task (Chapter 2, this volume). The germane cognitive load necessary to succeed at a task is inversely related to the level of automation of necessary prior knowledge (Clark & Elen, 2006). Other things being equal, when we have less of the prior knowledge required when learning a new task, we must use more mental effort to construct new cognitive operations that support task performance. The more automated the prior knowledge, the less cognitive effort required to apply it during learning. For example, the amount of germane cognitive load required for children to learn division is lower if they have more automated addition and subtraction skills. Children who have learned addition and subtraction routines recently and have had less time to practice and automate would need to invest more mental effort at multiplication than those who have practiced longer (Clark & Elen, 2006). Conscious cognitive processing that serves to assemble and/or implement a productive approach to learning a task is the source of relevant cognitive load. Providing a worked example of a successful approach to a new task during instruction for students with highly automated prior knowledge reduces the necessary, relevant load to its lowest possible level (Kirschner, Sweller, & Clark, 2004). Prior task experience fosters the development of implicit (automated, largely unconscious, procedural) task-relevant cognitive processes (e.g., Woltz, 2003) that are presumed to operate without consuming working memory space and so reduce the demand on working memory. Lohman (1989) described the problem of estimating the amount of cognitive load
210
Richard E. Clark and Vincent P. Clark
from the prior experience measures of individuals on any task when he cautioned: “What is novel for one person may not be novel for another person or even for the same person at a different time . . . [thus] inferences about how subjects solve items that require higher level processing must be probabilistic, since the novelty of each [item] varies for each person” (words in brackets added, p. 348). Br¨unken, Seufert, and Paas (Chapter 9, this volume) discuss this problem and acknowledge that we have not yet found precise measures of cognitive load for individual learners. Kalyuga and Sweller (2005) have suggested that one way to measure implicit knowledge might be to provide students with a problem and some of the initial steps necessary to solve the problem, and then ask them to describe what must be done next. The difficulty with this approach is the evidence that people who have highly automated knowledge about a task can perform the task but cannot accurately or completely describe the steps they follow (see a review by Feldon & Clark, 2006). Variable levels of prior knowledge automation may account for some of the error reported in Kalyuga and Sweller’s (2005) experiments. In general, the lack of a reliable, efficient measure of automated germane prior knowledge is a serious problem for CLT. A primary goal of CLT is to describe specific instructional methods that will maximize relevant and minimize irrelevant cognitive load for each learner at all stages of learning. Thus, when we have determined the total amount of cognitive load experienced by any individual in learning or problem-solving tasks, the next challenge is to break that total down into the proportion of germane (relevant) and extraneous (irrelevant) load being experienced. Yet, identifying the type and origin of mental workload is problematic because for any individual, the amount of load experienced during learning is influenced by the amount of prior knowledge he or she possesses and how automated that knowledge has become with use. Distinguishing between germane and extraneous cognitive load. A second urgent problem, related to the measurement of gross load and also addressed by Br¨unken, Seufert, and Paas (Chapter 9, this volume) is that we have also not yet found a way to reliably determine whether mental work is being invested in productive or unproductive mental activity. CLT is based on the distinction between “extraneous” or irrelevant load (mental effort invested in activities that do not support learning goals) and “germane” or relevant load (mental effort that supports learning and problem solving). And yet, these two key constructs are only inferred post hoc from differences between learning scores that result from different treatments that are presumed to provide more of one than the other type of load. In a later section, we suggest that eye movement and gaze direction technology might be used
From Neo-Behaviorism to Neuroscience
211
as an indicator of what is being processed cognitively and therefore an indicator of extraneous load. It is likely that we will solve the measurement of gross mental workload before we are able to deal with the more difficult problem of distinguishing between different types of load being experienced by a single individual. The next section discusses the construct definition and measurement problems that exist with CLT and possible ways to handle those problems with neuroscience research methods.
possible neuroscience contributions to measuring mental effort in clt Recent neuroscience research has made significant advances toward a better understanding of brain function during learning and problem solving (Szucs & Goswami, 2007). During learning, all information is coded in the brain in the form of synaptic activity that underlies the symbolic representations hypothesized by cognitive psychologists. The combination of neuroscience and cognitive science permits the development of a common, integrated framework consisting of connections between higher-level cognitive representations (such as the hypothesized constructs and relationships in CLT) and lower-level data concerning neuronal and biological functions in the brain and sensory systems (Szucs & Goswami, 2007). The ultimate goal of this integration is to add to our ability to predict and explain how our brain function and biology give rise to our mental functioning during learning and problem solving. This integration would bring us full circle and perhaps redress some of the historical mistakes we made at the turn of the last century. Although neuroscience may not yet have much to offer instructional designers or teachers, researchers might benefit from its focus on precise measurements of brain and sensory processes. One exciting possibility can be found in neuroscience research on mental workload and pupil dilation. Pupil Dilation and Vascular Constriction as Measures of Mental Workload Promising neuroscience measures of cognitive load may be available in two established physiological measurement technologies called pupillometrics (Megaw, 2005) and peripheral vasoconstriction (Iani, Gopher, & Lavie, 2004; Marshall, 2007). In the case of pupillometrics, devices have been developed to measure the amount of pupil dilation along with the direction and
212
Richard E. Clark and Vincent P. Clark
duration of a learner’s gaze. Vasoconstriction measurement requires the wearing of a device on one finger that measures variations in blood flow to the finger. Pupil dilation and mental effort. Considerable evidence supports the claim that pupil dilation is highly correlated with mental effort during learning and problem solving (Beatty, 1982; Beatty & Wagoner, 1978; Iqbal et al., 2005; Iqbal, Zheng, & Bailey, 2005; Kahneman & Beatty, 1966; Recarte & Nunes, 2003). Kahneman and Beatty (1966) compared a variety of encoding, processing, and retrieval tasks, and found that pupil diameter increased proportionally with the mental workload required. In a digit storage and recall task, pupil width increased proportionally with the number of digits encoded and decreased as they were reported. In a separate experiment, digit encoding was compared with digit transformation for the same series of digits. Pupil width was larger when the numbers were added before encoding. They also found that pupil width decreased with task repetition over the course of the study, as task difficulty decreased. This work was extended by Beatty and Wagoner (1978), who examined pupil diameter for a series of letter comparison tasks that increased in complexity, from physical comparisons to comparisons by name, then by category. Again, pupil diameter increased with increasing task complexity. Beatty (1982) reviewed all experimental data on pupil dilation and effort, and concluded that the relationship survives alternative explanations. One controversial aspect of these findings is that the neural circuitry thought to control pupil diameter, located in a variety of deep sub-cortical regions and in the brainstem, is not closely associated with the circuitry involved in working memory, located primarily in the dorsolateral prefrontal cortex. A considerable amount of evidence seems to support the claim that increasing cognitive load affects pupil diameter indirectly through changes in affect-based arousal, perhaps caused by the need to perform mental work (Kahneman & Beatty, 1966; Iqbal et al., 2004; Recarte & Nunes, 2003). And yet, it must be noted that because pupil dilation is apparently mediated by affect-based arousal, we need to learn more about the nature of the relationship between arousal, working memory, and mental effort. If, as some neuroscientists have suggested (Iqbal et al., 2004), mental work is always accompanied by arousal, then pupil dilation might serve as a highly reliable measure of workload. If we find significant individual differences in arousal with prior task knowledge held constant, we would be less inclined to settle on pupil dilation as a measure of mental effort. Early studies of this issue seem to indicate that emotionality may not influence dilation as much as mental effort. A dissertation by Simpson (2006)
From Neo-Behaviorism to Neuroscience
213
provided subjects with both abstract and concrete words that were either very pleasant or very unpleasant and found that, as expected, pupil dilation was greater for abstract words. However, dilation was not different for pleasant and unpleasant words. This question requires more research on individual and group differences in pupil dilation, but the uncertainty it raises does not eliminate the utility of pupil dilation as a measure of cognitive load. Individual and group differences in pupil dilation. Studies have examined individual and group differences in pupil dilation with tasks held constant. For example, Van Gerven, Paas, van Merri¨enboer, and Schmidt (2004) found differences between the pupil dilation of younger and older subjects in a study that examined six levels of memory load based on the classic Sternberg memory task. They concluded that dilation might not always be a good measure of mental effort for older (senior) learners. In an age judgment task in which photographs of faces were either gazing directly at the observer or to the side, Gillian, Hood, Troscianko, and Macrae (2006) reported that pupil dilation was greater and more sustained in female than in male participants when analyzing directly gazing faces of both genders. The authors concluded that their female subjects invested more effort in processing socially relevant (direct-gaze) than socially irrelevant (deviated-gaze) faces regardless of the gender of the face. Heitz, Schrock, Payne, and Engle (2003) described two experiments in which groups of subjects with greater or lesser working memory spans engaged in a memory task. They reported that both groups demonstrated equal pupil dilation during tasks requiring similar mental effort, even though those with greater working memory span achieved higher scores. These data suggest that mental effort may not be a good explanation for differences in working memory but that dilation may be a good indicator of mental work. Individual differences in the automaticity of task prior knowledge are an important issue in all studies of cognitive load (Clark & Elen, 2006). The more practice we experience with a task or critical components of a task, the less mental effort we require to learn related tasks or to assemble component tasks into a more complex set of skills. Cognitive load would presumably be less for a more experienced learner with more automated prior knowledge than one who has less prior knowledge. If we ask students to take pretests consisting of a sample of the types of tasks to be learned and/or tasks requiring the necessary prior knowledge for new learning, the amount of automated prior knowledge should be indicated by the correlation between the amount of pupil dilation during different pretest items and outcome measures, such as item solution speed and accuracy.
214
Richard E. Clark and Vincent P. Clark
Individuals who dilate more and are slower and less accurate will most likely have less automated levels of prior knowledge or less access to relevant knowledge and therefore be required to invest more effort to succeed. The less prior knowledge and the less automated that knowledge is, the more it is necessary to provide instruction that eliminates all extraneous load and provide only the essential steps in a worked example of how to perform the task to be learned. Carswell (2005) used pupil dilation to assess mental workload when surgical residents were practicing with novel laparoscopic surgical technology. He was looking for novel ways to not only improve instruction but also to test alternative technologies for surgery. He tested surgeons with different prior experience levels with traditional technology and with laparoscopic technologies in order to reason about the relative contribution of prior knowledge and variations in the technology to mental workload. Recarte and Nunes (2003) described a study using pupillometry in which the responses of different individuals to similar task conditions could be interpreted as different levels of prior automation. Finally, van Gog, Paas, and van Merri¨enboer (2008) studied the eye movements of people with different levels of expertise at electrical troubleshooting tasks. They reported that experts spent more time than novices looking at faultrelated components of devices but did not measure pupil dilation. They also found an expertise reversal effect (Kalyuga, Ayres, Chandler, & Sweller, 2003) in which conceptual knowledge about troubleshooting interfered with the learning of experts, perhaps because it served as extraneous load for experts but was germane for novices. It would be interesting to replicate this study and others by van Gog and colleagues (e.g., Nievelstein, van Gog, Boshuizen, & Prins, 2008) to collect data on the relative amount of mental effort invested by experts and novices during problem solving. In general, the combination of dilation, eye movement, and duration as measures of mental effort should be combined with subjects that differ in expertise and tasks that differ in complexity. Pupil dilation as a method to assess extraneous load during learning. Most important to CLT researchers is developing reliable ways to measure the amount and origin of extraneous (irrelevant) load during learning. The instructional goal is to anticipate and eliminate all sources of extraneous load so that working memory processing is as efficient as possible. Recarte and Nunes (2003) designed a creative way to combine pupil dilation and eye movement technology to test the amount of extraneous (irrelevant) mental load experienced by drivers to attend to a “hands-free” telephone conversation while driving and compared it with the load experienced attending to the same conversation “live” with a person riding with them in a car. They
From Neo-Behaviorism to Neuroscience
215
employed visual cues such as unexpected emergency road signals during the conversations to see if drivers noticed fewer of these important cues while engaging in conversations. It is interesting but not surprising to note that the amount of cognitive load was identical during both hands-free telephone and live conversations as measured by eye movement tracking and pupil dilation. It was also determined by eye movement tracking and behavioral observation that the extraneous load imposed by the conversations resulted in a 30% reduction in the drivers’ noticing of emergency cues during both the hands-free and live conversations. In a very different study of extraneous load, Verney, Granholm, and Marshall (2004) used pupil dilation to examine the differences between college student performances on a backward masking task that required them to overcome distractions in order to solve target detection problems. Their analysis indicates that students with lower SAT scores invested more wasteful effort focusing on the distractions in the task, which were accounted for by socio-economic differences and prior target detection accuracy. Marshall (2007) describes three problemsolving studies in which pupil dilation reliably distinguished between rest and work, between germane or extraneous effort, and between rested and fatigued states. Devices for measuring pupil dilation. A number of devices are currently available that will measure and analyze the pupil dilation for individuals during learning from computer displays or other fixed display technologies (Recarte & Nunes, 2003). Iqbal et al. (2004) concluded that “pupil size is the most promising single measure of mental workload because it does not disrupt a user’s ongoing activities, provides real-time information about the user’s mental workload and is less obtrusive than other physiological measures such as heart rate or EEG [electroencephalogram]” (p. 1477). In order to measure gaze and pupil dilation, it is often necessary to place a research subject’s head in a vice-like frame (similar to those used during eye examinations) to prevent head movement. Recently, however, relatively light and unobtrusive equipment is beginning to be developed, such as a camera mounted on a light headband worn by subjects described recently by Marshall (2007). It is highly likely that pupil dilation measured by the headband technology is much less intrusive than the interruptions caused by head fixation devices or secondary (latency) measures. Pupillometry may improve our measurement of the amount of cognitive load, and combining dilation with the direction and duration of gaze may also help to solve the problem of the relevancy of the load being experienced. Another less studied technology that also seems to offer the possibility of unobtrusive measurement of mental load is vascular constriction.
216
Richard E. Clark and Vincent P. Clark
Vascular constriction and mental effort. Iani et al. (2004) reported that a measure of the constriction of blood vessels in the fingers is a measure of sympathetic nervous system activation and might serve as a reliable measure of mental effort. They conducted two experiments in which they varied task difficulty and the level of engagement of their subjects in the task, and reported that increased vascular constriction (reduced blood flow to the fingers) was highly correlated with performing tasks (constriction was greater when working than when resting) and was greater with more difficult than with less difficult tasks. They also reported a strong correlation between vasoconstriction and pupil dilation. Iani, Gopher, Grunwald, and Lavie (2007) examined the vascular constriction of pilot performance in a computer-based flight simulator in which the difficulty of the task could be manipulated. They found that constriction was greater with difficult than with easier tasks. In general, vasoconstriction seems to provide an alternative way to measure gross cognitive load, yet it does not seem to offer a way to determine the source of the load being measured. At this point, the most promising way to measure both mental load and the source of the load seems to be the use of technology that captures pupil dilation along with gaze direction and duration.
imaging methods for monitoring changes in cognitive load Whereas pupil dilation provides intriguing evidence regarding the changes in neuro-cognitive activity that underlie cognitive load, more direct measures of brain function are available. A large number of brain imaging studies have examined working memory. Working memory provides a temporary store that supports cognitive processing, and the capacity of working memory is commonly thought to be closely associated with cognitive load. The higher the cognitive load required to perform a task, the greater the demand on working memory. The imaging methods discussed in the following sections can be used to examine the neural activity that supports working memory and therefore can indicate how changes in cognitive load affect brain function. Three basic processes of working memory have been identified: a brain network for the maintenance of auditory and verbal information, a separate network for the maintenance of visual and spatial information, and a central executive network for attentional control and manipulation of items in working memory (Baddeley, 1986), although evidence for the central executive is controversial (see the discussion by Sweller, 2004). Working
From Neo-Behaviorism to Neuroscience
217
memory includes three processing activities that occur in sequence: encoding, maintenance, and retrieval. Each of these processes involves a different pattern of brain activity, and each can be affected differently by changes in load. Each can be distinguished by differences in time, for example encoding must occur before retrieval. Isolation of the brain processes supporting each of these stages based on timing can be accomplished using event-related potentials (ERPs), which record the small fluctuations in voltage at the scalp surface generated by neural activity in the brain. This can be used to infer the timing of events, but it offers poor spatial resolution and therefore inadequate information about where the processing is occurring in the brain. By contrast, imaging methods that rely on hemodynamic measures, such as functional magnetic resonance imaging (fMRI) and positron emission tomography typically measure changes in blood flow and/or oxygenation that are related to changes in brain function. These hemodynamic methods offer superior spatial resolution compared with ERPs, which is necessary to unambiguously identify the anatomical location of brain networks supporting the different processes of working memory. However, because changes in blood flow are relatively slow, these methods are usually unable to identify rapid changes in brain activity. Event-related fMRI is a method that can be used to achieve a balance between spatial and temporal resolution (Clark, 2002; Clark, Maisog, & Haxby, 1998) by focusing on the characterization of small changes in signal over short periods of time. These methods can distinguish changes in neural activity occurring on the order of a few hundred milliseconds apart, depending on how the data are acquired and analyzed. ERP Studies of Working Memory As described earlier, working memory has a fundamental role in supporting cognitive load. ERPs can be used to examine the neural activity that supports working memory, and therefore how changes in cognitive load affect brain function. Many neuroscience studies have employed delayed response tasks to study working memory. Delayed response tasks require subjects to maintain information in working memory for a period of time before a response is made. This might be a word, an object, a location in space, or some other sensory feature or groups of features that must be held in memory. Often, such tasks involve one or more items that must be held in memory, to be compared with additional items presented later in time before a response can be made. Delayed response tasks often evoke a characteristic sustained negative electrical potential over the scalp termed the contingent negative
218
Richard E. Clark and Vincent P. Clark
variation (CNV). CNVs are evoked during the maintenance of information stored in working memory (Tecce, 1972). It is likely that the CNV results from increased synaptic activity associated with maintaining information in the working memory store. Working memory tasks have been found to evoke activity in a variety of brain regions. Gevins, Smith, and Le (1996) used high-resolution evoked potentials during verbal and spatial working memory tasks. In this study, verbal or spatial attributes were compared between each test stimulus and a preceding stimulus. All stimuli evoked the CNV and a number of other components, which varied in amplitude, depending on the specific requirements of the task. They concluded that working memory is a function of distributed neural systems with both taskspecific and task-independent components and that these and other ERP components can be used to study working memory processes. However, subsequent studies have shown that the interpretation of ERP components to study working memory can be more complex than is typically assumed. Kok (2001) found that the amplitude of positive components evoked from 300 to 500 msec post-stimulus reflected the activation of elements in an event-categorization brain network that is controlled by the joint operation of attention and working memory. This limits the use of these components as a measure of processing capacity or cognitive load because variations in both attention and working memory can influence their production. Luck, Woodman, and Vogel (2000) supported this view by suggesting that many studies confound attention and working memory. They proposed that attention may operate to adjust brain networks supporting working memory and other cognitive processes when brain systems are overloaded and therefore operates to adjust the brain’s ability to process the extra information under conditions of higher cognitive load and thus optimize performance. Finally, Wager and Smith (2003) suggested that selective attention to features of a stimulus to be stored in working memory leads to separate patterns of activation from working memory storage. Selective attention is the process whereby specific objects or classes of stimuli are selected for further processing based on certain defining stimulus characteristics. Depending on the nature of these characteristics (e.g., color, shape, or spatial location), a different pattern of brain response is seen that is unique to those characteristics. Thus, the dynamic properties of these interrelated neural and cognitive systems make it difficult to use these measures to quantify specific features, such as cognitive load. Even with these limitations, carefully designed studies that take these and other issues into consideration can reveal much about how the brain deals with variations in cognitive load.
From Neo-Behaviorism to Neuroscience
219
fMRI Studies of Cognitive Load Most fMRI studies of cognitive load effects examine the identity of brain regions that support different aspects of working memory. These studies typically use parametric designs. These designs reveal the neural correlates of working memory load by identifying those regions in which activity changes as the level of cognitive load is changed across repeated measurements. This method assumes that additional cognitive load will increase the brain responses in a proportional way, otherwise known as the pure insertion hypothesis (Raichle, 1994). Using these methods, a number of published studies have characterized brain networks that support working memory and how these networks change with changes in cognitive load. N-back tasks are one such design that involves the presentation of stimuli in a series, in which subjects are asked to compare the current stimulus with stimuli presented one or more items earlier in a series. For a delay of one stimulus, the N-back task is similar to the delayed response task. However, for more than one stimulus delay, N-back tasks differ from delayed response tasks in the use of intervening stimuli presented between the two stimuli being compared. With an increasing delay between the first and second item to be compared, the number of intervening items that must be maintained in working memory to perform the task increases, and this increases cognitive load in turn. These tasks also differ in that two comparisons are made for most stimuli in an N-back task, first with the stimulus presented N stimuli before it and then with the stimulus presented N stimuli after. Callicott et al. (1999) used fMRI to identify characteristics of working memory capacity using a parametric N-back working memory task. In this study, as the number of items was increased, task performance decreased. As cognitive load was increased, some brain regions indicated changes in activity that followed an inverted U shape. Large regions of dorsolateral prefrontal cortex, along with smaller regions of premotor cortex, superior parietal cortex, and thalamus, revealed changes in activity. The authors concluded that this pattern was consistent with a capacity-constrained response. At lower levels of load, less activity was required to support the working memory processes. At middle levels, more activity was required to maintain the same level of performance. At very high levels of load, the performance of the network breaks down, resulting in both reduced activity and reduced performance. These results reflect the findings in cognitive instructional psychology (e.g., Clark, 1999; Clark & Elen, 2006; Gimino, 2000; Salomon, 1983) where prior knowledge predicts mental effort under conditions in which tasks become increasingly difficult.
220
Richard E. Clark and Vincent P. Clark
These results demonstrated that a portion of the brain networks supporting working memory is sensitive to variations in cognitive load, whereas other portions do not appear to be as sensitive. Jaeggi et al. (2003) employed an N-back task with four levels of difficulty using auditory and visual material, and did not find the same inverted U-shape relationship. The participants’ tasks were performed separately or simultaneously as dual tasks. When performed separately, activation in the prefrontal cortex increased continuously as a function of memory load. An increase of prefrontal activation was also observed in the dual tasks, even though cognitive load was excessive in the case of the most difficult condition, as indicated by reduced behavioral performance. These results suggest that excessive processing demands in dual tasks are not necessarily accompanied by a reduction in brain activity. More recently, O’Hare, Lu, Houston, Bookheimer, and Sowell (2008) examined the development of these brain networks using a Sternberg working memory task with three load levels. The Sternberg task involves asking subjects to encode a set of stimuli (e.g., “1,” “3,” and “9”) and later presenting a series of stimuli and asking them to indicate which of these stimuli match the encoded set and which are new. The larger the size of the encoded stimulus set, the greater the cognitive load. The activated brain networks were found to depend on the participants’ age, which ranged from 7 to 28 years. Adolescents and adults showed cognitive-load effects in frontal, parietal, and cerebellar regions, whereas younger children showed similar effects only in left ventral prefrontal cortex. These results demonstrate that increasing load produces different brain network responses from childhood through adulthood. As a result, we may find developmental differences between the ways that young children and adults handle cognitive load during learning. Some of the differences observed across studies may result from variations in learning tasks. Using fMRI, working memory is often associated with increased activity in the prefrontal cortex, typically in Brodmann areas 6, 9, 44, and 46 (Cabeza & Nyberg, 2000). In area 6, located in the frontal cortex, activations are commonly found across tasks, including verbal, spatial, and problem-solving tasks, and thus may be related to general working memory operations that are not associated with other sensory or cognitive features of the task. By contrast, the exact pattern of activation in other brain areas is related to the specific nature of the task used. Increased activity in area 44, which lies next to area 6 in the lateral frontal cortex, is found for verbal and numeric tasks compared with visuospatial tasks, which may be related to phonological processing. Activations in areas 9 and 46, located on the frontal pole, are stronger for tasks that require manipulation of working memory contents compared with tasks that require only maintenance of
From Neo-Behaviorism to Neuroscience
221
items in working memory (Owen, 1997; Petrides, 1994, 1995). Ventrolateral frontal regions (including areas 45 and 47) are involved in the selection and comparison of information held in working memory, whereas medial and anterior frontal regions (areas 9 and 46) are involved in the manipulation of multiple pieces of information. Some studies have shown that working memory for object information engages ventral prefrontal regions, whereas working memory for spatial locations engages dorsal prefrontal regions (Courtney, Ungerleider, Keil, & Haxby, 1996, 1997). However, other studies suggest that working memory for objects engages left frontal regions, whereas working memory for spatial information engages right frontal regions (Belger et al., 1998; Smith, Jonides, & Koeppe, 1996; Smith et al., 1995). Taken together, these studies suggest that the organization of frontal brain networks that support working memory still holds a number of secrets in terms of the cognitive basis around which they are organized. Working memory studies also show activations in brain regions outside of the frontal cortex, including the parietal areas. In the case of verbal tasks, these activations tend to be larger on the left, which supports Baddeley’s phonological loop model, which maintains that information is stored and rehearsed in series (Awh et al., 1996; Paulesu, Frith, & Frackowiak, 1993). Working memory tasks are also associated with altered activity in anterior cingulate, occipital, and cerebellar cortices. However, these tend to be more sensitive to stimulus characteristics and task demands, rather than cognitive load, suggesting that they perform operations that support working memory indirectly through their interaction with these other regions. One exception to this is the finding of Druzgal and D’Esposito (2001), who showed that activity in ventral extrastriate visual areas increased directly with load of an N-back working memory task using facial stimuli. They concluded that both prefrontal and extrastriate areas worked together to meet the demands of increased cognitive load. Advanced methods of brain imaging offer many insights into the neural mechanisms that support working memory and the effects of changes in cognitive load on these mechanisms. Some progress has already been made in understanding the brain basis of processes related to cognitive load. Our ultimate goal is to achieve a unified theory that bridges the gap between cognitive psychology and neuroscience. We are beginning to see parallels across these two fields, as described earlier, but there is still much work to do. As brain imaging methods improve, and as cognitive psychologists are more willing to understand brain imaging technologies and to use this sort of information in forming hypotheses, a better understanding of cognitive load than could be achieved by either discipline alone can ultimately be achieved.
222
Richard E. Clark and Vincent P. Clark
summary and conclusion We have come full circle since Judd, an early cognitive psychologist, lost an argument to Thorndike, an early advocate of neurological and biological psychology. That lost argument serves as a cautionary metaphor for the bias that prevented American psychologists from focusing on cognitive questions for fifty years. It may also have produced a reaction whereby cognitive psychology is now experiencing a reverse bias against biological and neurological insights about learning and problem solving. The point of this review is to emphasize that the solution to some of the thorny problems facing CLT requires that we step away from our century-long dispute and become open to the insights offered by past and future advances in both cognitive psychology and neuroscience. It seems reasonable to expect that neuroscience might aid the search for ways to reliably quantify cognitive load and to identify the sources of germane and extraneous load. We might also increase our understanding of how individual and group differences in prior knowledge, culture, and working memory span might influence brain function, resulting in quantifiable differences in activity recorded with brain imaging methods such as ERPs and fMRI, and how this affects our understanding of differences between various learning tasks and instructional methods. We recommend a renewed commitment to exploring the use of pupil dilation accompanied by gaze direction and intensity studies to develop a more reliable and valid estimate of individual cognitive load and to help identify sources of germane and extraneous load. Pupil dilation could also be used to investigate how differences in the amount of prior expertise in a knowledge domain influence the type and amount of cognitive load experienced by learners. We expect that the more specific prior knowledge learners possess about the class of tasks being learned, the less load they will experience compared with learners who have less prior knowledge. It may also be possible that germane load for novices might become irrelevant load for experts and that this might be the source of the expertise reversal effect described by Kalyuga et al. (2003). We also suggest that fMRI methods of brain imaging offer many possible hypotheses based on evidence from neural mechanisms that support working memory and on the effects of changes in cognitive load on these mechanisms for different types of tasks and learners. Neuroscience studies draw most often on Baddeley’s (1986) model of working memory and search for evidence for three separate networks that maintain visual and spatial information, verbal information, and the control of attention and manipulation of items being held. In addition, neuroscience
From Neo-Behaviorism to Neuroscience
223
looks for evidence for three processing activities that occur in sequence in each of the three networks during learning and task performance – encoding, maintenance, and retrieval. Although a number of technologies are used to identify and validate these processes, the most complete and accurate is event-related fMRI. In general, the brain regions associated with most of these processes have been identified, but complex ambiguities and arguments persist. To this point, fMRI studies have provided additional evidence for the processes that occur in working memory and the brain structures that appear to support those processes. It also appears that working memory load consists of both task-specific and task-independent components. In addition, it appears that some experiments may confound working memory and attention processes. Claims have been made, for example, that when cognitive load increases, attention processes may be automatically evoked and serve to reduce load by forcing attention to more germane attributes of tasks (Wager & Smith, 2003). It is also possible that increases in load may evoke processes that focus attention on extraneous events (Clark, 1999). In addition, fMRI studies have provided evidence for the inverted-U hypothesis about the relationship between cognitive load and mental effort similar to the one suggested by Salomon (1983). When load is low, effort is also low, but as cognitive load increases, fMRI indicators of load also increase until it reaches a very high level in which the brain networks supporting working memory seem to fail, with accompanying decreases in mental effort and test performance. It also appears that some tasks may not produce the inverted U. At least one well-designed study (Jaeggi et al., 2003) identified dualcoding memory tasks in which increasing load (judged by both fMRI data and subject performance) did not yield decreasing effort. In general, there appear to be a number of important interactions among variations in task types, working memory processes, and cognitive load. Some areas of the brain seem to be active during all working memory processing, and some areas seem to specialize in different types of processing. For example, separate areas have been associated with verbal and numeric tasks, whereas others seem to be active during visuospatial tasks. In addition, tasks that require manipulation of the contents of working memory (thought to be associated with executive functions) activate different areas than tasks that require maintenance of both visuospatial and verbal-numeric information in working memory. Other studies have found evidence to suggest different regions support spatial location and object information. fMRI studies have also provided strong evidence for age-related developmental differences in the operation of working memory. As load increases
224
Richard E. Clark and Vincent P. Clark
in younger children, working memory activities appear in the left ventral prefrontal cortex, but in adolescents and adults, the same tasks produce cognitive-load activity in the frontal, parietal, and cerebellar regions. The reason for these differences and their consequence for instruction and/or learning are unknown. As neuroscience methods improve in spatial and temporal resolution and as new methods are developed, more precise information will be obtained. However, we know now that the cognitive sub-processes involved in performance of challenging learning and problem-solving tasks and the brain networks that support them interact in complex ways. In a single study, it is easy to confound the effects of changes in cognitive load on working memory with changes in attention as well as in perceptual and response processes, affect, and arousal, which all occur together in related ways. Therefore, it is vital that these methods are used carefully and alternative hypotheses be considered as we progress. Ultimately, though, we expect that these methods will lead to a better understanding of the neural and cognitive mechanisms that underlie cognitive load. references Awh, E., Jonides, J., Smith, E. E., Schumacher, E. H., Koeppe, R. A., & Katz, S. (1996). Dissociation of storage and rehearsal in verbal working memory: Evidence from positron emission tomography. Psychological Science, 7(1), 25–31. Baddeley, A. (1986). Working memory. New York: Oxford University Press. Beatty, J. (1982). Task-evoked papillary responses, processing load and the structure of processing resources. Psychological Bulletin, 91(2), 276–292. Beatty, J., & Wagoner, B. L. (1978). Pupillometric signs of brain activation vary with level of cognitive processing. Science, 199(4334), 1216–1218. Belger, A., Puce, A., Krystal, J. H., Gore, J. C., Goldman-Rakic, P., & McCarthy, G. (1998). Dissociation of mnemonic and perceptual processes during spatial and working memory using fMRI. Human Brain Mapping, 6(1), 14–32. Berline, D. (1960). Conflict, arousal and curiosity. New York: McGraw Hill. Cabeza, R., & Nyberg, L. (2000). Imaging cognition II: An empirical review of 275 PET and fMRI studies. Journal of Cognitive Neuroscience, 12(1), 1–47. Callicott, J. H., Mattay, V. S., Bertolino, A., Finn, K., Coppola, R., Frank, J. A., et al. (1999). Physiological characteristics of capacity constraints in working memory as revealed by functional MRI. Cerebral Cortex, 9, 20–26. Carswell, C. M. (2005). Assessing mental workload during laparoscopic surgery. Surgical Innovation, 12(1), 80–90. Clark, R. C., & Mayer, R. E. (2007). E-learning and the science of instruction: Proven guidelines for consumers and designers of multimedia learning. New York: Wiley. Clark, R. E. (1999). Yin and yang cognitive motivational processes operating in multimedia learning environments. In J. van Merri¨enboer (Ed.), Cognition and multimedia design (pp. 73–107). Herleen, Netherlands: Open University Press.
From Neo-Behaviorism to Neuroscience
225
Clark, R. E. (2001). Learning from media: Arguments, analysis and evidence. Greenwich, CT: Information Age Publishers. Clark, R. E., & Elen, J. (2006). When less is more: Research and theory insights about instruction for complex learning. In J. Elen & R. Clark (Eds.), Handling complexity in learning environments: Research and theory (pp. 283–297). Oxford, UK: Elsevier Science. Clark, R. E., & Estes, F. (1999). The development of authentic educational technologies. Educational Technology, 37(2), 5–16. Clark, R. E., Howard, K., & Early, S. (2006). Motivational challenges experienced in highly complex learning environments. In J. Elen & R. E. Clark (Eds.), Handling complexity in learning environments: Research and theory (pp. 27–41). Oxford, UK: Elsevier Science. Clark, V. P. (2002). Orthogonal polynomial regression for the detection of response variability in event-related fMRI. NeuroImage, 17, 344–363. Clark, V. P., Maisog, J. M., & Haxby, J. V. (1998). An fMRI study of face perception and memory using random stimulus sequences. Journal of Neurophysiology, 79, 3257–3265. Courtney, S. M., Ungerleider, L. G., Keil, K., & Haxby, J. V. (1996). Object and spatial visual-working memory activate separate neural systems in human cortex. Cerebral Cortex, 6(1), 39–49. Courtney, S. M., Ungerleider, L. G., Keil, K., & Haxby, J. V. (1997). Transient and sustained activity in a distributed neural system for human working memory. Nature, 386(6625), 608–611. Cronbach, L. J., & Snow, R. E. (1977). Aptitudes and instructional methods. New York: Irvington Press. Druzgal, T. J., & D’Esposito, M. (2001). Activity in fusiform face area modulated as a function of working memory load. Cognition and Brain Research, 10(3), 355–364. Feldon, D. F., & Clark, R. E. (2006). Instructional implications of cognitive task analysis as a method for improving the accuracy of experts’ self-report. In G. Clarebout & J. Elen (Eds.), Avoiding simplicity, confronting complexity: Advances in studying and designing (computer-based) powerful learning environments (pp. 109–116). Rotterdam, Netherlands: Sense Publishers. Gevins, A., Smith, M. E., & Le, J. (1996). High resolution evoked potential imaging of the cortical dynamics of human working memory. Electroencephalography and Clinical Neurophysiology, 98, 327–348. Gimino, A. E. (2000). Factors that influence students’ investment of mental effort in academic tasks: A validation and exploratory study. Unpublished doctoral dissertation, University of Southern California, Los Angeles. Heitz, R. P., Schrock, J. C., Payne, T. W., & Engle, R. W. (2003, May). Working memory and mental effort: Data from pupillometry. Paper presented at the Seventy-Fifth Annual Meeting of the Midwestern Psychological Association, Fort Wayne, Indiana. Iani, C., Gopher, D., Grunwald, A. J., & Lavie, P. (2007). Peripheral arterial tone as an on-line measure of load on a simulated flight task. Ergonomics, 50(7), 1026–1035. Iani, C., Gopher, D., & Lavie, P. (2004). Effects of task difficulty and invested mental effort on peripheral vasoconstriction. Psychophysiology, 41, 789–798. Iqbal, S. T., Adamczyk, P. D., Zheng, X. S., & Bailey, B. P. (2005, April). Task evoked pupillary response to mental workload in human-computer interaction.
226
Richard E. Clark and Vincent P. Clark
Paper presented at the 2005 Computer Human Interaction Conference, Portland Oregon. Iqbal, S. T., Zheng, X. S., & Bailey, B. P. (2004, April). Towards an index of opportunity: Understanding changes in mental workload during task execution. Paper presented at the 2004 Computer Human Interaction Conference, Vienna, Austria. Jaeggi, S. M., Seewer, R., Nirkko, A. C., Eckstein, D., Schroth, G., Groner, R., et al. (2003). Does excessive memory load attenuate activation in the prefrontal cortex? Load-dependent processing in single and dual tasks: functional magnetic resonance imaging study. NeuroImage, 19, 210–225. James, W. (1890). The principles of psychology (Vol. 1). New York: Henry Holt. James, W. (1904). Does ‘consciousness’ exist? Journal of Philosophy, Psychology, and Scientific Methods, 1, 477–491. Judd, C. H. (1908). The relation of special training to intelligence. Educational Review, 36, 28–42. Judd, C. H. (1910). Evolution and consciousness. Psychological Review, 17, 77–97. Kahneman, D., & Beatty, J. (1966). Pupil diameter and load on memory. Science, 154(3756), 1583–1585. Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). The expertise reversal effect. Educational Psychologist, 38(1), 23–31. Kalyuga, S., & Sweller, J. (2005). Rapid dynamic assessment of expertise to improve the efficiency of adaptive e-learning. Educational Technology Research and Development, 53, 83–93. Kirschner, P., Sweller, J., & Clark, R. E. (2006). Why minimally guided learning does not work: An analysis of the failure of discovery learning, problem-based learning, experiential learning and inquiry-based learning. Educational Psychologist, 41(2), 75–86. Kok, A. (2001). On the utility of P3 amplitude as a measure of processing capacity. Psychophysiology, 38(3), 557–577. Lohman, D. F. (1989). Human intelligence: An introduction to advances in theory and research. Review of Educational Research, 59(4), 333–373. Luck, S. J., Woodman, G. F., & Vogel, E. K. (2000). Event-related potential studies of attention. Trends in Cognitive Sciences, 4, 432–440. Marshall, S. P. (2007). Identifying cognitive state from eye metrics. Aviation, Space and Environmental Medicine, 78(1), B165–B175. Mayer, R. (2001). Multimedia learning. Cambridge, MA: Cambridge University Press. Mayer, R. (2004). Should there be a three-strikes rule against pure discovery learning? The case for guided methods of instruction. American Psychologist, 59(1), 14–19. Mayer, R. E. (2005). The Cambridge handbook of multimedia learning. New York: Cambridge University Press. Megaw, T. (2005). The definition and measurement of mental workload. In J. R. Wilson & N. E. Corlett (Eds.), Evaluation of human work: A practical ergonomics methodology (pp. 525–551). London: CRC Press. Miller, G. A. (1956). The magic number seven plus or minus two: Some limits on our capacity to process information. Psychological Review, 63, 81–97.
From Neo-Behaviorism to Neuroscience
227
Neisser, U. (1967). Cognitive psychology. New York: Appleton-Century-Crofts. Nievelstein, F., van Gog, T., Boshuizen, H., & Prins, F. J. (2008). Expertise-related differences in conceptual and ontological knowledge in the legal domain. European Journal of Cognitive Psychology, 20(6), 1043–1064. O’Hare, E. D., Lu, L. H., Houston, S. M., Bookheimer, S. Y., & Sowell, E. R. (2008). Evidence for developmental changes in verbal working memory load-dependency. NeuroImage, 42, 1678–1685. Owen, A. M. (1997). The functional organization of working memory processes within human lateral-frontal cortex: The contribution of functional neuroimaging. European Journal of Neuroscience, 9(7), 1329–1339. Paulesu, E., Frith, C. D., & Frackowiak, R. S. J. (1993). The neural correlates of the verbal component of working memory. Nature, 362, 342–345. Perkins, D. (1995). Outsmarting IQ: Learnable intelligence. New York: The Free Press. Petrides, M. (1994). Frontal lobes and working memory: Evidence from investigations of the effects of cortical excisions in nonhuman primates. In F. Boller & J. Grafman (Eds.), Handbook of neuropsychology (pp. 9, 59–82). Amsterdam: Elsevier. Petrides, M. (1995). Functional organization of the human frontal cortex for mnemonic processing: Evidence from neuroimaging studies. Annals of the New York Academy of Sciences, 769, 85–96. Porter, G., Hood, B., Troscianko, T., & Macrae, C. N. (2006). Females, but not males, show greater papillary response to direct- than to diverted-gaze faces. Perception, 35(8), 1129–1136. Porter, G., Troscianko, T., & Gilchrist, I. D. (2007). Effort during visual search and counting: Insights from pupilometry. Quarterly Journal of Experimental Psychology, 60(2), 211–229. Raichle, M. E. (1994). Images of the mind: Studies with modern imaging techniques. Annual Review of Psychology, 45, 333–356. Recarte, M. A., & Nunes, L. M. (2003). Mental workload while driving: Effects on visual search, discrimination and decision making. Journal of Experimental Psychology: Applied, 9(2), 119–137. Salomon, G. (1983). The differential investment of effort in learning from different sources. Educational Psychologist, 18(1), 42–50. Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic human information processing: 1. Detection, search, and attention. Psychological Review, 84, 1–66. Shulman, L. J. (1970). Reconstruction of educational research. Review of Educational Research, 40, 371–393. Simpson, H. M. (2006). The effects of word abstractness and pleasantness on pupil size during an imagery task. Dissertations Abstract International Section B. The Sciences and Engineering, 67(1-B), 584. Smith, E. E., Jonides, J., & Koeppe, R. A. (1996). Dissociating verbal and spatialworking memory using PET. Cerebral Cortex, 6(1), 11–20. Smith, E. E., Jonides, J., Koeppe, R. A., Awh, E., Schumacher, E. H., & Minoshima, S. (1995). Spatial vs. object-working memory: PET investigations. Journal of Cognitive Neuroscience, 7(3), 337–356.
228
Richard E. Clark and Vincent P. Clark
Snow, R. E. (1996). Aptitude development and education. Psychology Public Policy and Law, 2(3–4), 536–560. Steele, C. M. (2003). Stereotype threat and African-American student achievement. In C. M. Steele (Ed.), Young, gifted, and black (pp. 109–130). Boston: Beacon Press. Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(1), 257–285. Sweller, J. (2004). Instructional design consequences of an analogy between evolution by natural selection and human cognitive architecture. Instructional Science, 3, 9–31. Sweller, J. (2006). How the human cognitive system deals with complexity. In J. Elen & R. E. Clark (Eds.), Handling complexity in learning environments: Research and theory (pp. 13–26). Oxford: Elsevier Science. Szucs, D., & Goswami, U. (2007). Educational neuroscience: Defining a new discipline for the study of mental representations. Mind, Brain and Education, 1(3), 114–127. Tecce, J. J. (1972). Contingent negative variation (CNV) and psychological processes in man. Psychological Bulletin, 77, 73–108. Thorndike, E. L. (1903). Educational psychology. New York: Lemcke & Buechner. Van Gerven, P. W., Paas, F., van Merri¨enboer, J. J. G., & Schmidt, H. G. (2004). Memory load and the cognitive papillary response in aging. Psychophysiology, 412(2), 167–174. van Gog, T., Paas, F., & van Merri¨enboer, J. J. G. (2008). Effects of studying sequences of process-oriented and product-oriented worked examples on troubleshooting transfer efficiency. Learning and Instruction, 18(3), 211–222. van Merri¨enboer, J. J. G., & Kirschner, P. A. (2007). Ten steps to complex learning: A systematic approach to four-component instructional design. New York: Routledge. Verney, S. P., Granholm, E., & Marshall, S. P. (2004). Pupillary responses on the visual backward masking task reflect general cognitive ability. International Journal of Psychophysiology, 52(1), 23–36. Wager, T. D., & Smith, E. E. (2003). Neuroimaging studies of working memory: A meta-analysis. Cognitive and Affective Behavioral Neuroscience, 3(4), 255–274. Woltz, D. J. (2003). Implicit cognitive processes as aptitudes for learning. Educational Psychologist, 38(2), 95–104.
11 Cognitive Load in Learning with Multiple Representations holger horz and wolfgang schnotz
Technological innovations have led to important changes in teaching and learning during the last decade. Advances in computer technology have enabled the development of flexible learning arrangements, Web-based collaboration, and a broad range of multimedia learning environments. Multimedia allows the combination of different presentation formats, such as pictures, animations, text, or music in flexible ways via different sensory modalities. Media designers often assume that multimedia allows for a better adaptation of instruction to the learners’ needs and preferences. Multimedia is also expected to motivate learners, thus increasing their invested cognitive effort, which in turn should result in better learning. A major topic in the field of learning and instruction is how multimedia instruction interacts with the human cognitive architecture. Mayer (2001, 2005) as well as Schnotz (2001, 2005; Schnotz & Bannert, 2003) have developed theoretical models of multimedia learning that address this issue. Both models aim to explain what goes on in the mind of the learner when he or she learns from spoken or written texts with static or animated pictures. Both models share the assumption that the type and amount of presented information has to be adapted to the limitations of the cognitive system, especially those of working memory. The same assumption is at the core of Cognitive Load Theory (CLT) developed by Sweller and colleagues, which has become increasingly influential in instructional psychology during the last decade (Chandler & Sweller, 1991; Paas, Renkl, & Sweller, 2004; Paas & van Gog, 2006; Sweller, 1994, 2005; Sweller, van Merri¨enboer, & Paas, 1998). The theoretical models of multimedia learning developed by Mayer and Schnotz as well as the CLT developed by Sweller have been used to derive principles and guidelines for the design of multimedia learning environments. In this chapter, we investigate how CLT relates to the theoretical models of multimedia learning. First, we briefly describe and compare 229
230
Holger Horz and Wolfgang Schnotz
figure 11.1. Cognitive Theory of Multimedia Learning (Mayer, 2005, p. 37).
both models of multimedia learning. Second, we discuss guidelines and techniques of improving multimedia learning environments from a cognitive load perspective. Third, we draw conclusions with respect to further research regarding the design of multimedia learning environments.
models of multimedia learning Cognitive Theory of Multimedia Learning Mayer’s Cognitive Theory of Multimedia Learning (CTML) has become probably the most influential approach to knowledge acquisition from multimedia during the last decade (Mayer, 1997, 2001, 2005). A graphical representation of the theory is shown in Figure 11.1. Mayer refers to the dual-coding theory of Paivio (1986) and assumes that the human cognitive system includes a verbal and pictorial (image) subsystem. Accordingly, individuals use different representational formats to internally encode and store knowledge. Based on the working memory model of Baddeley (1992), Mayer assumes that two sensory subsystems exist in working memory: an auditory system and a visual system. The first basic assumption of CTML merges these two concepts. Humans are assumed to process information in working memory through two channels: an auditory-verbal channel and a visual-pictorial channel. The second basic assumption of CTML, which reflects the work of Baddeley (1992) as well as Chandler and Sweller (1991), is that these two channels have limited capacity to convey and
Cognitive Load in Learning with Multiple Representations
231
process information. The third basic assumption of CTML is that humans are active sense-makers: they engage in active cognitive processing to construct coherent knowledge structures from the available external information and their prior knowledge. According to Mayer, active learning from multimedia instructional messages includes a set of five coordinated processes: (1) selecting relevant words, (2) selecting relevant images, (3) organising the selected words into a verbal mental model, (4) organising the selected images into a pictorial mental model, and (5) integrating the verbal model and the pictorial model with prior knowledge into a coherent mental representation. Verbal selection processes lead to a propositional text base, and verbal organisation processes result in a text-based mental model. Similarly, pictorial selection processes lead to an image base, and pictorial organisation processes result in a picture-based mental model. The verbal organisation processes take place in the verbal part of working memory, whereas the pictorial organisation processes occur in the pictorial part of working memory (Baddeley, 1992; Chandler & Sweller, 1991). The text-based model and the picture-based model are then integrated in a one-to-one mapping process. During this mapping, elements of the text-based model are mapped onto elements of the picture-based model and vice versa. Similarly, relations within the text-based model are mapped onto relations within the picturebased model and vice versa. Integration requires that elements and relations of the text-based model and corresponding elements and relations of the picture-based model are simultaneously activated in working memory. As mentioned earlier, Mayer’s cognitive theory of multimedia learning has become highly influential in the field, and it has received support by numerous empirical studies. Nevertheless, there are some open questions, which suggest further theoretical reflection and empirical research. It might be beneficial for both sides if the theory would get into closer contact with text processing research, which has had considerable success during the last three decades (e.g., Gernsbacher, 1990; Graesser, Millis, & Zwaan, 1997; Kintsch, 1998; van Dijk & Kintsch, 1983). For example, Mayer’s ‘verbal model’ (i.e., a mental structure created by organising selected words) might be considered by text processing researchers as a propositional macrostructure, which serves as a basis for the construction of a mental model or situation model. The pictorial model (i.e., a mental structure created by organising pictorial information) seems to correspond to an analog mental
232
Holger Horz and Wolfgang Schnotz
representation, which is usually considered as the essential characteristic of a mental model (Johnson-Laird, 1983). The assumption of two channels in CTML – a visual-pictorial channel and an auditory-verbal channel – blends sensory modality and representation format. The association between the visual modality and the pictorial representation format on the one hand and the association between the auditory modality and the verbal representation format on the other hand are possibly not as close as the model suggests. We assume that verbal information is not necessarily associated with the auditory modality (e.g., as demonstrated by the use of sign languages) and that pictorial information is not necessarily associated with the visual modality. For example, the sound of an event (say, the starting up of an airplane or the call of a bird) can be considered as a sound image (Schnotz, 2005). This implies that there are not only visual pictures, but also auditory pictures in multimedia environments. In other words: visual patterns perceived by the eye can convey verbal as well as pictorial information (i.e., as written text and visual pictures, respectively), and the same is true for auditory patterns perceived by the ear, which can convey verbal and pictorial information (i.e., as spoken text and sound images, respectively), too. A further possible point of discussion refers to the assumed parallelism between text processing and picture processing in CTML. According to our point of view, texts and pictures use different sign systems resulting in fundamentally different forms of representations (Baddeley, 1992; Kosslyn, 1994; Paivio, 1986), which we refer to as descriptive and depictive representations, respectively. This difference between representation formats is at the core of the integrated model of text and picture comprehension (Schnotz, 2005; Schnotz & Bannert, 2003), which we describe in the following section.
integrated model of text and picture comprehension Spoken or written texts, mathematical equations, and logical expressions, for example, are descriptive representations. A descriptive representation consists of symbols describing an object. Symbols have an arbitrary structure (e.g., words in natural language), and they are related to the content they represent by means of a convention (cf. Peirce, 1906). Descriptive representations contain signs for relations, such as verbs and propositions in natural language. Pictures, sculptures, or physical models, however, are depictive representations. A depictive representation consists of iconic signs. Although depictive representations allow extracting relational information, they do not contain symbols for these relations. Instead,
Cognitive Load in Learning with Multiple Representations
233
they possess specific inherent structural features that allow reading off relational information, and they are associated with the content they represent through these common structural characteristics. Descriptive and depictive representations have different advantages in learning: whereas descriptions are more powerful in representing different kinds of subject matter, depictions are better suited to draw inferences (cf. Johnson-Laird, 1983; Schnotz, 1993). Based on the assumption that descriptions and depictions are fundamentally different forms of representations, Schnotz and Bannert (2003) have developed an integrated model of text and picture comprehension (ITPC), which was further elaborated by Schnotz (2005). The model integrates the concepts of multiple memory systems (Atkinson & Shiffrin, 1971), working memory (Baddeley, 1986, 2000), and dual coding (Paivio, 1986). It also integrates the idea of multiple mental representations in text comprehension and in picture comprehension (Kosslyn, 1994; van Dijk & Kintsch, 1983), as well as structural components of the CTML of Mayer (1997, 2001). The model refers not only to multimedia learning, but also to single-medium learning, because written text, spoken text, visual pictures, and auditory pictures (i.e., sound images) can be understood in isolation or in combination. Figure 11.2 shows a graphical outline of the model. The ITPC model is based on the following assumptions: text and picture comprehension are active processes of coherence formation. During comprehension of texts or pictures, individuals engage in building coherent knowledge structures from the available external verbal and pictorial information and from their prior knowledge. Text and picture comprehension take place in a cognitive architecture, which includes modalityspecific sensory registers as information input systems with high capacity but extremely short storage time, a working memory with a severely limited capacity and short storage time, and a long-term memory with both very high capacity and high storage time. Verbal information (i.e., information from written texts or spoken texts) and pictorial information (i.e., information from visual pictures and from auditory pictures) is transmitted to working memory through the visual channel or the auditory channel. Both channels have limited capacity to process and store information. Further information processing in working memory takes place in two different representational channels – the descriptive channel and the depictive channel – whereas information from written or spoken text is processed in the descriptive channel. Information from visual pictures or from auditory pictures (sounds) is processed in the depictive channel. Both channels have limited capacity to process and store information.
234
Holger Horz and Wolfgang Schnotz
figure 11.2. Integrative Model of Text and Picture Comprehension (Schnotz, 2005).
Accordingly, a perceptual level and a cognitive level of processing can be distinguished within the model. The perceptual level includes the information transfer between the sensory registers and working memory. This level is characterized by the functioning of the sensory channels. The cognitive level includes the information processing within working memory as well as the exchange of information between long-term and working memory. This level is characterized by the functioning of the descriptive channel and the depictive channel. For a written text to be understood, visually presented verbal information enters the visual register through the eye. The information is then forwarded through the visual channel to visual working memory. The resulting information pattern in visual working memory corresponds to the text surface representation in reading comprehension. The verbal information is forwarded from visual working memory through the verbal channel to propositional working memory, where it triggers the formation of
Cognitive Load in Learning with Multiple Representations
235
propositions, which in turn triggers the construction or elaboration of a mental model. For a spoken text to be understood, auditorily presented verbal information enters the auditory register through the ear. Then the information is forwarded through the auditory channel to auditory working memory. The information pattern in auditory working memory corresponds to the text surface representation in listening comprehension. The information is then forwarded from auditory working memory through the verbal channel to propositional working memory, where it leads to a propositional representation and finally triggers the construction or elaboration of a mental model. It should be noted that information in visual working memory can be verbal or pictorial and that information in auditory working memory can be verbal or pictorial as well. In some cases, it is even possible that the same sign convey both verbal and pictorial information. Imagine, for example, that the word ‘Pisa’ on a tourist poster is written in such a way that the letter ‘I’ is replaced by a picture of the Pisa tower. In this case, the picture of the tower has a double function, because it can be interpreted in two ways: On the one hand, it can be interpreted as a picture, and on the other hand, it can be interpreted as a verbal symbol (in this case, a letter). Because both verbal and pictorial information can enter auditory working memory and because both verbal and pictorial information can also enter visual working memory, a device is needed in both cases that directs verbal information to the verbal channel and pictorial information to the pictorial channel. Accordingly, the ITPC model assumes a verbal filter and a pictorial filter for auditory working memory as well as a verbal filter and a pictorial filter for visual working memory (symbolized in Figure 11.2 by triangles). The verbal filters and pictorial filters, which have insofar the function of routers, select the corresponding information from the visual or the auditory working memory and forward it to the corresponding representational channel (Schnotz, 2005). Thus, in text comprehension, the reader or listener constructs a mental representation of the text surface structure, generates a propositional representation of the semantic content (i.e., a text base), and finally constructs from the text base a mental model of the subject matter described in the text (Schnotz, 1994; van Dijk & Kintsch, 1983; Weaver, Mannes, & Fletcher, 1995). These construction processes are based on an interaction of bottom-up and top-down activation of cognitive schemata, which have both a selective and an organising function. Task-relevant information is selected through topdown activation, and the selected information is organised into a coherent
236
Holger Horz and Wolfgang Schnotz
mental representation of the text surface structure. Processes of conceptual organisation starting from the text surface representation (through bottomup activation) result in a coherent propositional representation, which in turn triggers the construction of a mental model. Mental model construction implies a transition from a descriptive to a depictive representation. Propositional representations and mental models are assumed to interact continuously via processes of model construction and model inspection guided by cognitive schemata. The mental model is constructed through Gestalt-directed construction rules that lead to a mental model of a typical instance of the content or the situation described in the text. After a mental model has been constructed, schema-directed processes of model inspection can be applied to extract new information from the model. This information is encoded in a propositional format and, thus, elaborates the propositional representation. For a visual picture to be understood, visually presented pictorial information enters the visual register through the eye. Then information is forwarded through the visual channel to visual working memory, where it results in a visual perceptual representation of the picture. A pictorial filter selects pictorial information from visual working memory and forwards it through the pictorial channel, where it leads to the construction or elaboration of a mental model. The mental model can be used to extract new information and to encode this information in propositional working memory. In sum, visual picture comprehension requires first creating a visual mental representation of the image through perceptual processing and then constructing both a mental model and a propositional representation of the learning content through semantic processing. In perceptual processing of visual pictures, task-relevant information is selected through top-down activation of cognitive schemata and then visually organised through automated visual routines (Ullman, 1984). Perceptual processing includes identification and discrimination of graphic entities, as well as the visual organisation of these entities according to the Gestalt laws (Wertheimer, 1938; Winn, 1994). The resulting visual perception is an internal depictive representation, created as a surface representation of the external picture. It retains structural characteristics of the external picture, and it is sensory specific because it is linked to the visual modality (cf. Kosslyn, 1994; Shepard, 1984). To understand an external picture rather than only to perceive it, semantic processing is required. The individual has to construct a mental model of the depicted subject matter through a schema-driven mapping process, in which graphical entities (i.e., visual configurations such as the bars in a graph) are mapped onto mental entities
Cognitive Load in Learning with Multiple Representations
237
(such as the oil price at time x) and in which spatial relations (such as ‘higher than’) are mapped onto semantic relations (such as ‘more expensive than’) as encoded in the mental model (cf. Falkenhainer, Forbus, & Gentner, 1989/1990; Schnotz, 1993). In other words, the comprehension of external pictures is considered as a process of analogical structure mapping between a system of visuo-spatial relations and a system of semantic relations (cf. Gentner, 1989). In understanding realistic external pictures, the individual uses schemata of everyday perception. In contrast, for the understanding of logical external pictures such as diagrams or graphs, which contain abstract non-realistic representations (Horz & Schnotz, 2008), the individual requires specific cognitive schemata (so-called graphic schemata) to extract information from the visuo-spatial configuration (e.g., Lowe, 1996; Pinker, 1990). For an auditory external picture (i.e., a sound) to be understood, auditory external pictorial information enters the auditory register through the ear. Then information is forwarded through the auditory channel to auditory working memory, where it results in an internal auditory perceptual representation of the sound. An auditory pictorial filter selects pictorial information from auditory working memory and forwards it through the pictorial channel, where it leads to the construction or elaboration of a mental model. The mental model can be used to extract new information and to encode this information in propositional working memory. Altogether, the ITPC model assumes a continuous interaction between the propositional representation and the mental model, both in text comprehension and picture comprehension. In text comprehension, the starting point of this interaction is a propositional representation, which is used to construct a mental model. This model can then be used to extract new information to further elaborate the propositional representation. In picture comprehension, the starting point of the interaction is a mental model, which is used to extract new information that is also added to the propositional representation. Accordingly, there is no one-to-one relationship between external and internal representations. External descriptive (i.e., text) and depictive (i.e., picture) representations both lead to internal descriptive and depictive mental representations. Meaningful learning from text and pictures requires a coordinated set of cognitive processes, including selection of information, organisation of information, activation of prior knowledge, and active coherence formation by integration of information from different sources. In the comprehension of written or spoken texts, learners select relevant verbal information from words, sentences, and paragraphs as external sources of information. They
238
Holger Horz and Wolfgang Schnotz
organise this information, activate related prior knowledge as an internal source of information, and construct a coherent propositional representation as well as a coherent mental model. In the comprehension of visual pictures, learners select relevant pictorial information from drawings, maps, or graphs as an external source of information. They organise this information, activate related prior knowledge as a further (internal) source of information, and construct a coherent mental model complemented by a propositional representation. In the comprehension of auditory pictures (i.e., sound comprehension), the learner selects relevant acoustic information, organises this information, activates related prior knowledge as an internal source of information, and constructs a coherent mental model complemented by a propositional representation.
commonalities and differences Commonalities The integrated model of text and picture comprehension and the cognitive theory of multimedia learning have various assumptions in common, but they also differ in some respects. One commonality is that both models assume a cognitive architecture with multiple memory stores, including a working memory system of limited capacity. Another commonality is that they also assume different channels for processing verbal and pictorial information, which corresponds to the basic hypothesis of dual coding theory. Moreover, both models assume hierarchically ordered levels of processing in working memory (cf. Cermak & Craik, 1979). Lower and Higher Order Processing In this section, we distinguish between lower order processing and higher order processing. Roughly speaking, lower order processing refers to the processing of verbal and pictorial surface structures, whereas higher order processing refers to the processing of semantic structures, such as propositional representations or mental models. Lower order processing implies all subsemantic and perceptual processes, which are necessary to transfer information from the auditory and visual registers to working memory. Furthermore, lower order processing includes the cognitive processes that create a text-surface representation in auditory working memory or a visual pictorial representation in visual working memory. All lower order processes are normally executed in a
Cognitive Load in Learning with Multiple Representations
239
widely automated way and are only marginally influenced by intentional processes. Higher order processing includes thematically oriented semantic selection and processing of information, which results in a propositional representation or a mental model of the learning content. Higher order processing also includes the interaction between propositional representations and mental models, which we refer to as model construction and model inspection (Schnotz, 2005). Finally, higher order processing includes access to thematically related information in long-term memory and its integration into propositional representations and mental models. Contrary to lower order processing, higher order processing can be influenced to a large extent by intentional processes. Differences The two models differ in the following respects. In the CTML model of Mayer (2005), sensory modality and representational format are merged by the assumption of an auditory-verbal channel and a visual-pictorial channel. In the ITPC model of Schnotz (2005), on the contrary, verbal information is not necessarily associated with the auditory modality, but can also be conveyed by the visual modality. Similarly, pictorial information is not necessarily associated with the visual modality, but can also be conveyed by the auditory modality (e.g., sound images). Accordingly, the ITPC predicts that specific combinations of instructional elements create extraneous load. For example, when learners are expected to learn how the brakes of a car stop the car, instruction designers could present a schematic animation of how the brakes work. Let us assume that the sound of a car while slowing down is added to the animation to provide a more elaborated perceptual input. If, by mistake, the sound of a truck were added, the inappropriate sound would be propagated through the auditory sensory channel and through the depictive representational channel. It would create an inappropriate sound image and activate inappropriate prior knowledge. This would hamper the construction of a correct mental model and, thus, cause extraneous cognitive load. The CTML, on the contrary, does not consider sound images as a part of multimedia learning and does therefore not consider them as a potential source of additional cognitive load. The ITPC model distinguishes between multiple sensory channels (visual, auditory, touch, and others) within lower order processing and distinguishes two representational channels (verbal and pictorial) within higher order processing. A distinction between a lower level of processing and a
240
Holger Horz and Wolfgang Schnotz
higher level of processing is also included in the CTML model, in which sounds and images correspond to the lower (perceptual) level, whereas verbal and pictorial models correspond to the higher (cognitive) level. A further difference between the two theoretical approaches is that the CTML model assumes the construction of a verbal mental model and a pictorial mental model, which then have to be integrated. In contrast, the ITPC model proposes that only one mental model, which integrates information from different sources, is constructed. However, the models are consistent with one another in that both assume that pictorial and verbal materials are integrated in working memory. Because the ITPC assumes that picture comprehension includes processes of structure mapping, the structure of pictures plays a crucial role for the construction of mental models. If a visualization with a taskinappropriate structure is presented, mental model construction will be hampered because the structural conflict between the required mental model and the external visualization imposes an extraneous cognitive load on working memory (Schnotz & Bannert, 2003). On the contrary, because the CTML does not assume structure mapping processes, it does not predict such structural interference (Mayer, 2005). Whereas CTML assumes parallel processes of text and of picture comprehension, the ITPC model considers the relation between text comprehension and picture comprehension as essentially asymmetric. More specifically, the ITPC model assumes that graphics in a learning environment will influence the mental model more directly than text, whereas text will influence the propositional representation more directly than graphics. Depending on the required task performance after learning, germane load would be enhanced by a stronger emphasis on the text if primarily a propositional representation is required, whereas germane load would be enhanced by a stronger emphasis on the graphics, if primarily a mental model is needed.
cognitive load in multimedia comprehension Basic Assumptions of CLT The two models of multimedia learning described previously assume that comprehension and learning are highly dependent on the constraints of the human cognitive system. These constraints refer to the sensory registers (which are highly temporally limited) and to working memory (which is limited in capacity as well as temporally limited; cf. Baddeley, 1986), whereas
Cognitive Load in Learning with Multiple Representations
241
long-term memory is assumed to have a practically unlimited capacity (cf. Atkinson & Shiffrin, 1968). The assumption of a highly limited working memory and its implications for teaching and learning is also at the core of CLT (Sweller et al., 1998). According to CLT, instructional design should be adapted to the constraints of the human cognitive architecture. During learning, new information is processed in working memory, which eventually also changes the content of long-term memory. Because, according to CLT, any kind of conscious cognitive processing puts a cognitive load on working memory, processes of guided cognitive learning are always associated with a cognitive load. Consequently, CLT stresses the importance of optimally adapting instruction to efficient usage of the limited working memory capacity in learning situations (Chandler & Sweller, 1991; Paas et al., 2004; Sweller & Chandler, 1994; Sweller et al., 1998). CLT distinguishes among three kinds of loads: intrinsic, extraneous, and germane (see Chapter 2, this volume). The complexity of a learning task causes intrinsic load. The degree of intrinsic load depends on the learner’s prior knowledge, his or her intellectual abilities, and the complexity of the learning material (Schnotz & K¨urschner, 2007). Cognitive activities that specifically aim at learning cause germane load, which reflects the cognitive effort of schema abstraction and schema automation. Germane load should be increased because schema automation reduces the cognitive load of learning tasks (see Chapters 2 and 8, this volume). Extraneous load is defined as an unnecessary load resulting from an inappropriate instructional format. Hence, extraneous load should be reduced as much as possible, whereas germane load should be increased as much as possible.
guidelines for manipulating cognitive load at different levels of processing in multimedia learning Various design guidelines based on CLT have been published to optimize learning processes with multimedia or to prevent negative cognitive load effects in multimedia learning (e.g., Mayer & Moreno, 2003; Moreno & Mayer, 2007; see Chapters 7 and 8, this volume). In the following sections, we focus on the kinds of cognitive load affected by these guidelines and to what extent the corresponding instructional design principles affect higher order or lower order processes. We investigate three groups of guidelines or principles: (1) guidelines that refer to general characteristics of media-based learning (multimedia learning vs. single-medium learning), (2) guidelines
242
Holger Horz and Wolfgang Schnotz
concerning specific instructional design characteristics of multimedia learning, and (3) guidelines for adapting learning environments to specific learner characteristics. Multimedia Learning versus Single-Medium Learning The general characteristic of multimedia learning is the usage of multiple representational formats, such as text and pictures. This general characteristic is also reflected in the well-known multimedia principle: students learn better from text and pictures than from text alone (Mayer, 1997, 2001). The multimedia principle does not make suggestions for how multimedia learning environments should be designed. It suggests that multimedia learning environments should be designed and used rather than singlemedium environments. The multimedia principle implies the prediction that learning with verbal explanations and corresponding visual representations will result in more successful learning than learning with only verbal explanations or only visual representations. Recent research within the field of CLT has demonstrated, however, that the multimedia principle does not apply under all conditions. If a learner’s prior knowledge is high, learning from a single medium can lead to better learning results than multimedia learning (Kalyuga, Ayres, Chandler, & Sweller, 2003; see Chapter 3, this volume). The multimedia principle also does not apply if learners possess insufficient spatial abilities (Plass, Chun, Mayer, & Leutner, 2003). However, if there are no such obstacles, the design and use of a multimedia learning environment is advisable. The multimedia principle is the first principle that should be considered by instructional designers. It is a precondition for the application of further guidelines. Guidelines Concerning Specific Instructional Design Characteristics of Multimedia Learning The second group of guidelines refers to specific instructional design characteristics of multimedia learning. These guidelines correspond to the principles of spatial and temporal contiguity, two redundancy principles, the coherence principle, and the modality principle. Moreover, the techniques of segmenting and aligning belong to this category. These principles and techniques focus on single aspects of multimedia learning systems. Spatial and temporal contiguity principles. The two contiguity principles claim that spatial or temporal distance between different semantically related media (typically, written text and pictures) should be minimized.
Cognitive Load in Learning with Multiple Representations
243
For the following reasons, effects of discontiguity are especially strong when text is combined with animation. To be integrated into a mental model, verbal and pictorial information has to be simultaneously held available in working memory. Because of the fleeting nature of animations and the necessity to split one’s visual attention when reading text and observing an animation, the mental surface structure representations of text and animation (resulting from lower level processing) will be relatively incomplete. It is well known that viewing movies and reading subtitles simultaneously is relatively difficult, particularly when a lot of text is included in subtitles (Koolstra, 2002). The same problem exists in multimedia learning. Learners cannot build up a coherent text surface representation or an image representation under the condition of high discontiguity combined with the fleeting nature of animations, because this results in high costs of information search processes. On the level of higher order processing, low contiguity may be harmful too, because the construction of a mental model depends on the availability of appropriate surface structure representations. As a result, higher order cognitive processing is restricted by the effects of discontiguity on lower order (surface level) processing. Although these higher order processes may be hampered as a result of low contiguity, the cause of these problems does not lie within the higher order cognitive processing. In fact, the problems result from an insufficient surface representation, which derive from impeded lower order processing. Problems caused by discontiguity can be attributed primarily to lower processing level in multimedia learning. Hence, guidelines derived from spatial and temporal contiguity principles aim at the lower level of information processing in working memory. Redundancy principles. In research on multimedia learning, the term ‘redundancy’ has been used differently in different context, which has finally resulted in the formulation of two different redundancy principles. Moreno and Mayer (2002) use the term ‘redundancy’ in a relatively specific way, namely, in the context of redundant narrated and written texts combined with pictures. We will call the corresponding principle the specific redundancy principle. Sweller (2005) uses the term ‘redundancy’ in a more general way, as he refers to any verbal or pictorial source of information as redundant if this source presents nothing new for learners because they already possess the corresponding information. We will call the corresponding principle the general redundancy principle (cf. Schnotz, 2005). If any of these kinds of redundancy exist among different sources of information in multimedia learning, cognitive load will increase without improvement of learning. Hence, the additional load caused by redundancy should be classified as extraneous load (Sweller, 2005). The question arises, ‘At which
244
Holger Horz and Wolfgang Schnotz
level of information processing does this additional extraneous load occur in multimedia learning?’ Regarding the specific redundancy principle, we assume that when oral and written text is presented concurrently, learners try to build up simultaneously a text surface representation of the auditory as well as a text surface representation of the written text. After a visual representation of a written text segment is constructed, the information of the written text also has to be processed in auditory working memory. A cognitive overload of the auditory working memory arises due to the parallel building of a visual and auditory text representation, because learners tend to process both sources of information (rather than ignoring one of them), even if the two texts are identical. Working memory is impaired because of an overload at the level of lower order processing of text-surface representations from two concurrent sources, whereas higher order processing is affected only indirectly. There might also be a problem of asynchrony between listening and reading because the speed of the narrative text is fixed, whereas reading speed can vary according to the individual requirements of information processing (Bornkessel, Fiebach, Friederici, & Schlesewsky, 2004). It should be noted that learners are obviously unable to ignore a redundant written text in order to avoid cognitive overload. Whether the single learning sequences in the experiment of Moreno and Mayer (2002) were relatively short so that learners had no chance to find out that the redundant written text could be ignored seems to deserve futher investigation. If text length was important for the specific redundancy effect, one could speculate that in case of longer (redundant) spoken and written texts combined with pictures, learners would be more likely to ignore the written text, which subsequently would result in a considerably reduced cognitive load. The general redundancy principle assumes that if a single medium is self-explanatory and sufficient to understand the subject matter, redundancy between different information sources in a learning environment will hamper learning performance (Sweller, 1994, 2005). In other words, if one learning medium is sufficiently intelligible for the learner and a second medium provides the same (i.e., redundant) information, the latter will impose an extraneous load on the learner’s working memory. In the following sections, we focus on redundant combinations of written text and pictures. Such combinations require always some split of attention, because the learner’s eyes cannot read the text and observe the picture simultaneously. Besides this cognitive load, which can be reduced as much as possible through the principle of spatial contiguity, reading the text or observing the picture per se does not cause a cognitive overload within lower order
Cognitive Load in Learning with Multiple Representations
245
cognitive processing, when neither the text nor the picture is too complex at the surface level. In fact, the negative effect of two properly designed but redundant media is a result of superfluous processing in terms of integrating information, which is only redundant and therefore does not further contribute to mental model construction. It is still an open question as to whether only total redundancy leads to negative consequences or whether partial redundancy between different media will influence the learning process in a negative way as well. It seems plausible to assume that a partial redundancy between different media in a learning environment is required for an integration of both information sources (Moreno & Mayer, 2002). In the case of partially redundant media, some redundancy (i.e., semantic overlap) is required to create cross-referential connections between the different sources of information. Moreover, additional information, only presented in one medium, supports the further elaboration of the mental model. Coherence principle. The coherence principle implies that learners perform better when extraneous material (such as interesting but irrelevant words, pictures, sounds, or music) is excluded rather than included (Moreno & Mayer, 2000; Harp & Mayer, 1997; Chandler & Sweller, 1991). A general conceptual problem of the coherence principle is that it compares learning from different sources, when one source entails important and unimportant information, whereas the other source entails only important information. Following the multimedia learning models, the different learning outcomes are no surprise because different informational content was learned, and one can therefore expect that different mental models were created. On the one hand, the coherence principle aims at optimizing information processing at a higher level in working memory when it suggests that no confusing information should be integrated into learners’ mental models. On the other hand, an advantage of learning materials without irrelevant content should be expected also with lower order processing, simply because less information has to be processed. Overall, the coherence principle leads to an optimization primarily at the level of higher order processing in working memory, but it has positive effects at the level of lower order processing in working memory as well. Modality principle. The modality principle postulates that students learn better from animation and narrated text than from animation and written text. Following Mayer (2001) and Sweller (1999), using two channels expands the effective working memory capacity and reduces the probability of an overload that would be unavoidable if a single sensory channel rather than two channels were used for two sources of information (Sweller, 1999). In
246
Holger Horz and Wolfgang Schnotz
the light of multimedia learning models, the effectiveness of the modality principle is due to the lower level of information processing because the visual channel has a lower load when it is processing only one source. If text is presented in an auditory format, mental model construction will be facilitated because the visual channel does not have to provide all the information for constructing an image representation and a text-surface representation simultaneously. The assumption of a higher overall working memory capacity is based on the idea that capacities in working memory are not flexibly allocated to the auditory or the visual channel, but are rather partially dedicated to a specific channel. Therefore, learners are able to process more information within a specific time interval by using both processing channels. Besides the fact that a lower total cognitive load leads to more free cognitive capacity of working memory, higher level processes are not directly influenced by the modality principle. Segmenting. The recommendation to segment complex materials in smaller parts leads, on the one hand, to lower spatial or temporal contiguity, which contradicts the contiguity principle. On the other hand, segmenting has proved to be an important design option when complex transient learning materials contain too many relevant elements so that an overload would otherwise occur in working memory. Such an overload might be caused by two factors: (1) Learners may be unable to generate an adequate mental model because a continuous transient presentation of information, such as an animation or an audio stream, is too fast or too manifold to be adequately processed with limited capacity within a limited time. Learners can be cognitively overstrained by the need to integrate new elements in working memory while the mental model construction based on the previously seen information is still in progress. In this case, higher temporal distance between different information elements (including some time for ‘cognitive wrap-up’; cf. Just & Carpenter, 1980) may facilitate information processing. In addition, static pictures (such as detailed anatomic pictures, for example) may also lead to cognitive overload if learning time is limited. Segmenting the whole picture would facilitate the visual processing of relevant content. In terms of multimedia learning models, segmenting is a technique that supports higher order cognitive processing in working memory. First, presenting learning content in smaller segments may better enable learners to construct an adequate mental model. Second, learners internalize the new information in long-term memory by further cognitive processing. After processing one segment, information from the following segment can be processed more efficiently compared with complex (transient) learning materials that are
Cognitive Load in Learning with Multiple Representations
247
not segmented. The costs of higher spatial and temporal incontiguity are relatively unimportant if the segmentation does not separate content that is essential for initial understanding. Therefore, a video and a corresponding audio file, for example, should not be presented in different temporally separated units so that they cannot be processed together. (2) Additionally, segmenting may facilitate processes on a lower level of information processing in working memory, but this effect is a relatively indirect one. Mayer and Moreno (2003) posited segmenting as a possible method to reduce cognitive overload if both channels are overloaded. This cognitive overload arises from model building activities when an ongoing stream of new dynamic information is presented to the learner. Hence, the learner’s ability to build text surfaces and image representations in these situations may be hampered because of the dynamic nature of instructional materials when the mental model building processes at the higher level of processing are still in progress. As a result, learners simply overlook critical information and therefore do not incorporate it into the text surface and pictorial surface representations. Signaling. The recommendation to signal materials, that is, by inking relevant sentences of a text, is different from all optimization guidelines. On lower levels of information processing in working memory, the intention of signaling is to solve problems of extraneous material. Learners’ perception can be guided to the most essential elements on the learning environment by marking them. In doing so, the attention of learners is drawn away from extraneous material so that learners are more likely to allocate their cognitive resources to the essential elements (see Chapter 7, this volume). On a higher level of processing, signaling may support the processes of understanding and integrating information. Altogether, instructional designers are able to enhance all levels of information processing by signaling. Guidelines for Adapting Learning Environments to Specific Learner Characteristics The third category of design guidelines is represented by principles of individualized learning (see Chapter 4, this volume), with a focus primarily on specific learner characteristics. The design guidelines of this category specify optimized instructional conditions for specific groups of learners according to the research tradition regarding aptitude-treatment interaction (Snow, 1989). Principle of individual differences. The principle of individual differences aims at adapting elements of instructional design to different ability levels
248
Holger Horz and Wolfgang Schnotz
of learners. This principle postulates that instructional design may affect learners differently, depending on their prior knowledge and their cognitive spatial abilities. In other words, this principle moderates the effects of other general design principles (e.g., the redundancy principle and contiguity principle), as we demonstrate in the following paragraphs. Further, we point out that prior knowledge and spatial abilities are effective at different levels in multimedia learning. (1) The effects of prior knowledge are related to the higher level of processing. As Mayer (2001) concludes from his research (e.g., Mayer & Sims, 1994), learners with higher prior knowledge are able to compensate for weaknesses of instructional design because they do not need as many cognitive resources for mental model construction at the level of higher order cognitive processing as learners with lower prior knowledge. The mental model is easier to construct for learners with higher prior knowledge because they are able to use more elaborated schemas and because they need to integrate less new information than learners with lower prior knowledge do. Additionally, learners with higher prior knowledge have more automated schemas available, which can be applied in the process of learning and lead to lower cognitive load compared with learners with lower prior knowledge. (2) Contrary to prior knowledge, spatial ability seems to affect information processing at all levels of multimedia learning. Mayer and Sims (1994) found that learners with high spatial ability performed better when words and pictures were presented simultaneously rather than successively, whereas learners with low spatial abilities showed no differences between both kinds of presentation. This finding suggests that learners with high spatial abilities are better able to integrate verbal and pictorial information compared with learners with low spatial abilities.
conclusions According to our analysis, the majority of principles to optimize instructional design and techniques to prevent cognitive overload focus on the lower level of information processing by reducing extraneous load. Researchers such as Mayer, Sweller, Chandler, Moreno, van Merri¨enboer, Plass, and many others have contributed highly influential research on this topic. Although mindful learning is associated with higher level processes of understanding and integrating information, most optimizing principles and techniques that intend to prevent cognitive overload do not directly influence this higher order cognitive processing. There are some techniques that address the optimization at the higher level of information processing
Cognitive Load in Learning with Multiple Representations
249
during learning (cf. part 3 of Mayer, 2005), but they address mostly specific settings and contain fewer general instructional principles that would serve as guidelines for multimedia design to enhance lower order information processing. Perhaps for practical reasons, most research in this domain has been conducted to investigate cognitive load at lower levels of cognitive processing: Surface level design characteristics, such as the distances between words and pictures, continuous versus segmented video streams, or narrated versus written texts can be operationalized more easily than design characteristics that refer to higher order semantic processing. We suppose that there exists a still undetected potential for optimizing multimedia learning at a higher level of cognitive processing by fostering metacognitive processes, as suggested by Roy and Chi (2005) or the use of graphical and other instructional aids (Seufert & Br¨unken, 2006; Seufert, J¨anen, & Br¨unken, 2007). Related aspects of integrating information from different sources have also been described by Pl¨otzner, Bodemer, and Feuerlein (2001). Concerning the relation between ITPC and CTML, instructional designers should decide which framework they will use to analyze instructional materials or which framework should be the theoretical starting point when constructing a learning environment. One aspect could be the granularity of planning. CTML is adequate to explain major effects of multimedia information processing, but it is less detailed with regard to graphics comprehension and does not consider auditory pictures. Future research in the field of CLT could also be a source of evidence for the assumptions of the CTML or ITPC models. Altogether, it seems to us that the following conclusions can be drawn with respect to optimization principles and techniques to prevent cognitive overload in multimedia learning. First, research about multimedia instructional design has focused so far primarily on the importance of reducing extraneous load to ease lower level information processing. Second, future research in that field should focus more on techniques to optimize learning processes at higher levels of information processing. Third, further research should also take interactions between multimedia instructional design and learner characteristics more systematically into account. references Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motivation: advances in research and theory (Vol. 2, pp. 742–775). New York: Academic Press.
250
Holger Horz and Wolfgang Schnotz
Atkinson, R. C., & Shiffrin, R. M. (1971). The control of short-term memory. Scientific American, 225, 82–90. Baddeley, A. D. (1986). Working memory. Oxford, UK: Clarendon Press. Baddeley, A. D. (1992). Working memory. Science, 255, 556–559. Baddeley, A. D. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Science, 4, 417–423. Bornkessel, I. D., Fiebach, C. J., Friederici, A. D., & Schlesewsky, M. (2004). ‘Capacity’ reconsidered: Interindividual differences in language comprehension and individual alpha frequency. Journal of Experimental Psychology, 51, 278–279. Cermak, L., & Craik, F. (1979). Levels of processing in human memory. Hillsdale, NJ: Erlbaum. Chandler, P., & Sweller, J. (1991). Cognitive load theory and the format of instruction. Cognition and Instruction, 8, 293–332. Falkenhainer, B., Forbus, K. D., & Gentner, D. (1989/1990). The structure-mapping engine: Algorithm and examples. Artificial Intelligence, 41, 1–63. Gentner, D. (1989). The mechanisms of analogical learning. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 197–241). Cambridge, UK: Cambridge University Press. Gernsbacher, M. A. (1990). Language comprehension as structure building. Hillsdale, NJ: Erlbaum. Graesser, A. C., Millis, K. K., & Zwaan, R. A. (1997). Discourse comprehension. Annual Review of Psychology, 48, 163–189. Harp, S., & Mayer, R. (1997). The role of interest in learning from scientific text and illustrations. On the distinction between emotional interest and cognitive interest. Journal of Educational Psychology, 89, 92–101. Horz, H., & Schnotz, W. (2008). Multimedia: How to combine language and visuals. Language at Work, 4, 43–50. Johnson-Laird, P. N. (1983). Mental models: Towards a cognitive science of language, interference, and consciousness. Cambridge, UK: Cambridge University Press. Just, M. A., & Carpenter, P. A. (1980). A theory of reading: From eye fixations to comprehension. Psychological Review, 87, 329–354. Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). Expertise reversal effect. Educational Psychologist, 38, 23–31. Kintsch, W. (1998). Comprehension: A paradigm for cognition. Cambridge, UK: University Press. Koolstra, C. M. (2002). The pros and cons of dubbing and subtitling. European Journal of Communication, 17, 325–354. Kosslyn, S. M. (1994). Image and brain. Cambridge, MA: MIT Press. Lowe, R. K. (1996). Background knowledge and the construction of a situational representation from a diagram. European Journal of Psychology of Education, 11, 377–397. Mayer, R. E. (1997). Multimedia learning: Are we asking the right questions? Educational Psychologist, 32, 1–19. Mayer, R. E. (2001). Multimedia learning. New York: Cambridge University Press. Mayer, R. E. (Ed.). (2005). The Cambridge handbook of multimedia learning. New York: Cambridge University Press.
Cognitive Load in Learning with Multiple Representations
251
Mayer, R. E., & Moreno, R. (2003). Nine ways to reduce cognitive load in multimedia learning. Educational Psychologist, 38, 43–52. Mayer, R. E., & Sims, V. K. (1994). For whom is a picture worth a thousand words? Extensions of a dual coding theory of multimedia learning. Journal of Educational Psychology, 84, 444–452. Moreno, R., & Mayer, R. E. (2000). A coherence effect in multimedia learning: The case for minimizing irrelevant sounds in the design of multimedia instructional messages. Journal of Educational Psychology, 92, 117–125. Moreno, R., & Mayer, R. E. (2002). Verbal redundancy in multimedia learning: When reading helps listening. Journal of Educational Psychology, 94, 156–163. Moreno, R., & Mayer, R. E. (2007). Interactive multimodal learning environments. Educational Psychology Review, 19, 309–326. Paas, F., Renkl, A., & Sweller, J. (2004). Cognitive load theory: Instructional implications of the interaction between information structures and cognitive architecture. Instructional Science, 32, 1–8. Paas, F., & van Gog, T. (2006). Optimising worked example instruction: Different ways to increase germane cognitive load. Learning and Instruction, 16, 87–91. Paivio, A. (1986). Mental representations: A dual coding approach. Oxford, UK: Oxford University Press. Peirce, C. S. (1906). Prolegomena to an apology for pragmaticism. The Monist, 16, 492–546. Pinker, S. (1990). A theory of graph comprehension. In R. Freedle (Ed.), Artificial intelligence and the future of testing (pp. 73–126). Hillsdale, NJ: Erlbaum. Plass, J. L., Chun, D. M., Mayer, R. E., & Leutner, D. (2003). Cognitive load in reading a foreign language text with multimedia aids and the influence of verbal and spatial abilities. Computers in Human Behavior, 19, 221–243. Pl¨otzner, R., Bodemer, D., & Feuerlein, I. (2001). Facilitating the mental integration of multiple sources of information in multimedia learning environments. In C. Montgomerie & I. Viteli (Eds.), Proceedings of the World Conference on Educational Multimedia, Hypermedia & Telecommunications (pp. 1501–1506). Norfolk, VA: Association for the Advancement of Computing in Education. Roy, M., & Chi, M. T. H. (2005). The self-explanation principle in multimedia learning. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (pp. 271–286). New York: Cambridge University Press. Schnotz, W. (1993). On the relation between dual coding and mental models in graphics comprehension. Learning and Instruction, 3, 247–249. Schnotz, W. (1994). Aufbau von Wissensstrukturen. Weinheim, Germany: Beltz. Schnotz, W. (2001). Sign systems, technologies, and the acquisition of knowledge. In J. F. Rouet, J. Levonen, & A. Biardeau (Eds.), Multimedia learning? Cognitive and instructional issues (pp. 9–29). Amsterdam: Elsevier. Schnotz, W. (2005). An integrated model of multimedia learning. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (pp. 49–69). New York: Cambridge University Press. Schnotz, W., & Bannert, M. (2003). Construction and interference in learning from multiple representations. Learning and Instruction, 13, 141–156. Schnotz, W., & K¨urschner, C. (2007). A reconsideration of cognitive load theory. Educational Psychology Review, 19, 469–508.
252
Holger Horz and Wolfgang Schnotz
Seufert, T., & Br¨unken, R. (2006). Cognitive load and the format of instructional aids for coherence formation. Applied Cognitive Psychology, 20, 321–331. Seufert, T., J¨anen, I., & Br¨unken R. (2007). The impact of intrinsic cognitive load on the effectiveness of graphical help for coherence formation. Computers in Human Behavior, 23, 1055–1071. Shepard, R. N. (1984). Ecological constraints on internal representations: Resonant kinematics of perceiving, thinking, and dreaming. Psychological Review, 91, 417– 447. Snow, R. E. (1989). Aptitude-treatment interaction as a framework for research on individual differences in learning. In P. L. Ackerman, R. J. Sternberg, & R. Glaser (Eds.), Learning and individual differences: Advances in theory and research (pp. 13–59). New York: Freeman. Sweller, J. (1994). Cognitive load theory, learning difficulty and instructional design. Learning and Instruction, 4, 295–312. Sweller, J. (1999). Instructional design in technical areas. Camberwell, Australia: ACER Press. Sweller, J. (2002). Visualisation and instructional design. In R. Ploetzner (Ed.), Proceedings of the International Workshop on Dynamic Visualizations and Learning (pp. 1501–1510). T¨ubingen, Germany: Knowledge Media Research Center. Sweller, J. (2005). Implications of cognitive load theory for multimedia learning. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (pp. 19–30). New York: Cambridge University Press. Sweller, J., & Chandler, P. (1994). Why some material is difficult to learn. Cognition and Instruction, 12(3), 185–233. Sweller, J., van Merri¨enboer, J. J. G., & Paas, F. G. W. C. (1998). Cognitive architecture and instructional design. Educational Psychology Review, 10, 251–296. Ullman, S. (1984). Visual routines. Cognition, 18, 97–159. van Dijk, T. A., & Kintsch, W. (1983). Strategies of discourse comprehension. New York: Academic Press. Weaver III, C. A., Mannes, S., & Fletcher, C. R. (Eds.). (1995). Discourse comprehension. Hillsdale, NJ: Erlbaum. Wertheimer, M. (1938). Laws of organization in perceptual forms in a source book for Gestalt psychology. London: Routledge & Kegan Paul. Winn, W. D. (1994). Contributions of perceptual and cognitive processes to the comprehension of graphics. In W. Schnotz & R. Kulhavy (Eds.), Comprehension of graphics (pp. 3–27). Amsterdam: Elsevier.
12 Current Issues and Open Questions in Cognitive Load Research ¨ roland brunken, jan l. plass, and roxana moreno
The previous chapters have outlined the theoretical background, basic assumptions, and some key applications of Cognitive Load Theory (CLT) in its current state of development. The fundamental idea underlying CLT is that instructional design decisions should be informed by the architecture of the human cognitive system. CLT can therefore be described as a cognitive theory of instructional design. CLT has been very influential in educational research since the 1980s. It has inspired a growing number of research studies aimed at deriving empirically based guidelines for instructional design. Moreover, at its present stage of development, CLT is arguably one of the most influential instructional design theories. However, the extant research on cognitive load raises questions about the assumptions underlying CLT, some of which have not been consistently supported by the empirical data, suggesting the need to update the theory by incorporating recent empirical findings on cognition and learning (Schnotz & Kirschner, 2007). The first goal of this chapter is to summarize the theoretical developments of CLT and highlight some of its strengths and limitations. An additional contribution of CLT research includes efforts to develop practical, valid, and reliable measures of its main construct: cognitive load. However, as suggested in Chapter 9, the existing body of cognitive load research fails to exhibit methodological consistency regarding cognitive load measurement and lacks appropriate methods to measure other relevant constructs, such as the different load types proposed by the theory (DeLeeuw & Mayer, 2008). The second goal of this chapter is therefore to summarize the methodological developments of CLT research and highlight some of their strengths and limitations. Finally, most of CLT research has focused on identifying instructional design principles for problem-solving tasks in well-defined domains, using 253
254
Roland Br¨unken, Jan L. Plass, and Roxana Moreno
laboratory studies that consist of very brief interventions. An open question is whether CLT can be used to guide the design of more authentic learning environments in ill-defined domains and over longer periods of time. The third goal of this chapter is therefore to suggest new venues for research on CLT that would broaden its empirical base. In sum, the goal of this chapter is to synthesize the strengths and weaknesses of current theoretical and methodological developments of CLT and to suggest new directions for future research. At present, the lack of theoretical clarity on key cognitive load concepts and the absence of valid measures that can reliably distinguish among the three types of cognitive load threaten the explanatory and predictive power of the theory. Consequently, several empirical studies aimed at testing CLT show inconclusive findings or findings that contradict the very assumptions underlying CLT. In this chapter, we present examples of this issue and suggest some productive directions for the advancement of our understanding of the relation between cognitive load and learning.
conceptual development of cognitive load CLT conceptualizes learning as information processing that is actively carried out in the human cognitive system and that results in lasting mental representations. A direct implication of CLT is that instructional methods need to be based on the structures and affordances of the human cognitive system to support the learning process. Most of the original CLT assumptions about the human cognitive system were supported by basic research in cognitive science, neuropsychology, and educational psychology. The two basic concepts on which CLT is built are the limited capacity of working memory and the long-lasting structure of long-term memory (Cowan, 2001; Ericsson & Kintsch, 1995; Miller, 1956). According to CLT, learning is an active, resource-consuming process, resulting in schema formation (Bartlett, 1932; Johnson-Laird, 1983). Perhaps the most attractive aspect of CLT is its simplicity and intuitiveness. It seems logical that effort will only result in learning gains when instruction is designed to avoid the overload of cognitive demands on the learner. However, several issues arise when delving deeper into the theory. In the next sections, we provide our evaluation about some of the assumptions underlying CLT, including the relation between cognitive load and learning, the definition of the sources of cognitive load, and the additivity hypothesis. In addition, we point out three essential areas that require significant development before CLT can become a viable theoretical framework
Current Issues and Open Questions in Cognitive Load Research
255
to understand learning phenomena: the relation between cognitive load and mental representation theories, the relation between cognitive load and working memory theories, and the role of motivation, affect, and metacognition in cognitive load. Current Issues regarding the Relation between Cognitive Load and Learning The most basic assumption, made in the original concept of CLT, is that there is a negative relation between cognitive load and learning (Sweller, 1988): the lower the load experienced during learning, the higher the learning outcome. The negative relation between cognitive load and learning has been empirically supported by many cognitive load effect studies (see Chapter 7, this volume). For example, the research on split attention and coherence effects shows that when instruction is designed to force the learner to integrate visual and verbal materials or process extraneous information, learning is hindered. The negative relation between cognitive load and learning is also supported by the research on worked-out examples, which shows that asking students to study worked-out problems increases learning compared with independently solving the same problems (see Chapter 5, this volume). In sum, this line of research suggests that instruction should be designed to make learning as easy or as effortless as possible. However, several studies have challenged this conclusion (Schnotz & Kirschner, 2007). For example, as described in Chapter 4 (this volume), the inverse relation between load and learning depends on the prior knowledge of the individual. Unlike novices, experts or students with higher levels of knowledge in a domain may benefit from learning with non-integrated rather than integrated information or from independent problem-solving practice compared with worked-out example instruction. In addition, making instruction too easy may hinder learning by failing to provide the necessary challenge to motivate the learner (Pintrich & Schunk, 2002). More than 50 years ago, Atkinson (1958) showed that the highest level of effort invested in a task occurs when the task is of moderate difficulty. Tasks that are either very easy or very hard elicit the lowest level of effort, which in turn produces the lowest performance on a task. The core instructional design issue, therefore, is not to reduce the amount of cognitive load per se, but rather to find the appropriate level of cognitive load for each learner. Therefore, a better description of the relationship between cognitive load and learning is an inverse U-shape: low cognitive load hampers learning, learning increases with increased load until an
256
Roland Br¨unken, Jan L. Plass, and Roxana Moreno
optimal level of load has been reached, and then decreases as cognitive load exceeds the learner’s cognitive capacity. The negative relationship between cognitive load and learning only seems to apply to the particular case in which the learner is operating at the limits of his or her capacity. Current Issues regarding Definitions of the Three Cognitive Load Types The lack of satisfaction with the oversimplistic assumption that all reductions in cognitive load result in increased learning motivated later revisions of CLT. Specifically, CLT researchers distinguished intrinsic load from extraneous load (Sweller, 1988) as a way to capture the difference between the cognitive demands that are intrinsic to the difficulty of the materials to be learned and those that result from the unnecessary processing imposed by poor design. A second theoretical development was the concept of germane load (Paas & van Merri¨enboer, 1994). It was suggested that this third load type accounted for the finding that instruction that increases cognitive load with methods that are aimed at fostering schema acquisition increases learning. According to current CLT, being able to distinguish among these three sources of cognitive load is key for understanding the relationship between learning and instruction. The definition of intrinsic load as load related to the complexity of the information, extraneous load as the processing of information that is not relevant to learning, and germane load as the mental effort invested in relevant cognitive activity is both intuitive and clear at first glance. Yet, a closer look at these assumptions reveals several inconsistencies. For example, intrinsic and germane types of cognitive load may be interrelated: the higher the intrinsic load inherent in the material, the more the germane load required to process the material. The intrinsic load experienced by a given learner would be a function of this learner’s actual engagement with the material, not only the material’s complexity. Engagement, however, is a concept that is not included in the current model of CLT. Recent research has raised questions concerning the definition of intrinsic load. In the past, intrinsic load had been defined as an attribute of the learning material based on the amount and interrelation of the concepts that had to be learned (element interactivity) (Paas, Renkl, & Sweller, 2003). According to this definition, intrinsic load cannot be manipulated by means of instructional design because it is an attribute of the material to be learned itself. However, recent research suggests there is a need to reconsider this idea. For example, the segmenting principle (Mayer & Chandler, 2001) can be interpreted as an instructional method that reduces intrinsic load by
Current Issues and Open Questions in Cognitive Load Research
257
chunking the material to be learned in smaller units (see Chapter 7, this volume). Still, this interpretation needs to be directly tested in future research by including measures for the two types of load of interest. Another conceptual problem with CLT’s definition of intrinsic load is that it can be argued that the difficulty of the material to be learned also depends on the learner’s prior knowledge. For a novice learner, the size of an element to be learned might be much smaller than for a domain expert who can chunk information more effectively and therefore can subsume more information within one element. To date, there is no model integrating the role of learner characteristics, such as individual differences in priorknowledge, working memory capacity, and domain-specific abilities that can help predict the relative intrinsic difficulty of the materials for a specific learner in a specific situation. A further problem is that CLT’s definitions of extraneous and germane loads are circular and therefore fail to explain and predict learning. For example, the distinctive characteristic of extraneous load is that it arises from the unnecessary cognitive processing that is imposed on the learner because of suboptimal instructional design. But when is instructional design suboptimal? It is only after observing the results of the studies that the researchers offer conclusions about the type of load imposed by the design: if it increases learning, the method is interpreted to be optimal and to have decreased extraneous load, increased germane load, or produced a combination of the two effects. In contrast, if the instructional method is found to hinder learning, it is interpreted as having increased extraneous load. For example, most design principles of multimedia learning, such as those presented in Chapter 7 (this volume), are based on empirical studies comparing different design variants of the same information. Those designs in which learners outperform others with respect to knowledge acquisition are seen as imposing less extraneous cognitive load because of their better learning outcomes, even though cognitive load is often not measured (e.g., Mayer, 2001). More research is needed that directly measures the types of cognitive load imposed by the learning environment to shed light on the distinctive characteristics of extraneous and germane load types (see Chapter 9, this volume). Similarly, CLT argues that germane load is the result of methods that foster cognitive activities that are necessary for schema acquisition. Yet, the theory is silent on the description of such activities. What types of processing are likely to promote schema acquisition and how? The lack of an answer to this question in our view highlights a more basic problem of CLT.
258
Roland Br¨unken, Jan L. Plass, and Roxana Moreno
Usually, CLT argues that learning results in schema acquisition; however, this is a largely unproven statement. CLT lacks a theoretical conceptualization of how information is processed in the cognitive system and how knowledge is represented. This is the core difference between CLT and the theoretical models of Mayer (2005) or Schnotz (see Chapter 11, this volume), which make explicit assumptions about mental representational processes. Nevertheless, the research on the relationships among cognitive load, cognitive processing, and mental representations is still in its infancy. The few empirical studies available show that the relationships among the cognitive process of organization, (germane) cognitive load, and learners’ learning prerequisites are usually complex and not clearly explained in terms of simple main effects (Seufert & Br¨unken, 2006; Seufert, J¨anen, & Br¨unken, 2007; Seufert, Sch¨utze, & Br¨unken, 2009). A final problem is that whether a particular load is extraneous or germane will also depend on the specific educational goals of the instruction. For example, manipulating a science simulation or using a chart to represent simulation results may generate germane load if the goal is to promote the deeper understanding of science principles as measured by a problemsolving transfer test (Plass et al., 2009). However, the same methods may induce extraneous load if the objective is to promote the recall of the principles to be learned (Plass, Homer, & Hayward, 2009; Wallen, Plass, & Br¨unken, 2005). In the latter case, it may be optimal to teach the principles explicitly rather than to ask students to infer the principles from their interactions with the simulations. In other words, the type of load induced by any instructional method depends on the type of processing that is targeted by the instruction. However, in many cases, learning goals are fluid and change over time, and the goals of the learner may not be the same as the goals of the instructional designer. In addition, although it may be possible to define clear learning goals for brief interventions, complex learning environments typically facilitate learning with outcomes that go beyond the comprehension of the content, such as the development of learning strategies, metacognition, self-regulation, self-efficacy, and others. The current issues surrounding CLT’s definitions of intrinsic, extraneous, and germane cognitive load described earlier also raise concerns about the definition of total cognitive load experienced by a learner, which we discuss in the next section. Current Issues regarding the Additivity Hypothesis An additional assumption of CLT is the hypothesis that the total amount of cognitive load experienced during a learning situation can be estimated
Current Issues and Open Questions in Cognitive Load Research
259
by adding the three types of cognitive load experienced by the learner (Paas et al., 2003). According to CLT, efficient learning can only take place when the total amount of cognitive load does not exceed the capacity of the learner’s working memory resources. Otherwise, learning is impaired due to cognitive overload. However, the specific distribution of resources among the three types of cognitive load will affect how much is learned. For instance, the lower the extraneous load imposed by a certain instructional design is and the higher (up to a certain point) the germane load invested by a learner is, the more efficient the learning will be. The simplicity of the additivity hypothesis is attractive, yet, it is problematic on a variety of levels. One of the most important issues is that this definition does not take into account the fact that recent models of working memory postulate different subsystems that process verbal and non-verbal information separately and independently from one another (Baddeley, 1986, 2000). Effects such the modality effect and the resulting ideas of offloading of cognitive load from one subsystem of working memory to another (Chapter 7, this volume) are not currently considered in the additivity hypothesis (Rummer, Schweppe, F¨urstenberg, Seufert, & Br¨unken, 2010). Therefore, the idea of simple linear additivity of the three load types, as described in early cognitive load research (Paas et al., 2003), needs to be revised to describe how the different sources of load contribute to the overall cognitive load. Such a model would need to consider the working memory model used as a foundation for an update of CLT and to define additivity based on the different subsystems of this model. It would also need to be based on the updated definitions of the different load types. Schnotz (Schnotz & Kirschner, 2007; Chapter 11, this volume) has made several related arguments about why intrinsic load may not be necessarily fixed and how extraneous load could be further broken down in subcategories. However, the empirical base for such a revision of the additivity hypothesis needs to be strengthened, for example by experimentally combining different extraneous and germane load effects in one experimental setting and measuring the overall resulting cognitive load. Mental Representations and Cognitive Load With respect to the representational format of the knowledge stored in long-term memory, CLT was originally based on schema theory. A schema is a complex representational knowledge structure that includes declarative and procedural knowledge related to a specific topic or process, such as the functionality of a technical system or the steps of placing an order in a
260
Roland Br¨unken, Jan L. Plass, and Roxana Moreno
restaurant (Bartlett, 1932). However, in past years, CLT has also been discussed in the frame of other mental representation theories, such as Paivio’s dual coding theory (Paivio, 1986), the cognitive theory of multimedia learning (Mayer, 2005), and the integrated model of text and picture comprehension (Schnotz & Kirschner, 2007; Chapter 11, this volume). Although CLT seems compatible with several knowledge representation models, it does not explicitly specify the relations between cognitive load and specific forms of knowledge representation. However, the quality of knowledge construction depends on the mental representations that have been generated during the learning process. For example, a learner may have constructed a propositional representation of the explanation for how a car braking system works and would therefore be able to successfully describe the elements involved in such a system on a later test. However, this does not necessarily mean that the learner will be able to successfully solve a problem, such as troubleshooting a malfunction in the car’s braking system. Several studies have shown that the type of knowledge and skills acquired depend on the type of mental representations fostered by instruction (Schnotz, Boeckheler, & Grzondziel, 1999; Wallen et al., 2005). If different knowledge representations foster different skills or levels of understanding, one should also ask whether they produce different levels of cognitive load during learning. For instance, presenting a visual model of a scientific system in addition to a verbal explanation of how the system works has been found to promote students’ problem-solving transfer skills (see the multimedia effect, Chapter 7, this volume). However, CLT is currently not able to explain whether the added pictures help learning by causing a decrease in students’ overall cognitive load, an increase in students’ germane load, a decrease in students’ extraneous load, or by facilitating the construction of a mental representation that is better suited for the problemsolving task. Future CLT developments should therefore specify the relation between the construction of different mental representations and cognitive load constructs. The current long-term memory assumptions underlying CLT are unable to provide an answer to this issue. Working Memory and Cognitive Load According to CLT, learners experience cognitive load within their limited working memories as the result of engaging in several cognitive activities, such as extracting information from an external representation, integrating different external representations into one model, or integrating new information into existing schemas. However, there is evidence that this
Current Issues and Open Questions in Cognitive Load Research
261
assumption is overly simplistic. A good example is the split-attention effect (see Chapter 7, this volume). According to CLT, this effect is the result of the advantage of presenting multiple sources of information in an integrated format rather than in a split-source format. Is this indeed a working memory effect, or can it be explained as the result of perceptual and attentional processing mechanisms? Likewise, is the color-coding effect (Kalyuga, Chandler, & Sweller, 1998) the result of reducing the amount of resources in working memory, or can it be explained by the process of selective attention? Currently, CLT does not clearly distinguish among the basic perceptual, attentional, and working memory processes that occur during learning. Here, again, is a need to further develop the theoretical foundation of CLT and its underlying assumptions of how the cognitive system works. A related area in which CLT requires theoretical refinement concerns the working memory model used by the theory. There is a large amount of ongoing research in the field of working memory and a number of different models describing how this cognitive system works (Miyake & Shah, 1999). CLT seems to endorse Baddeley’s (1986, 2000) model of working memory, which postulates that there are two specialized subsystems for the processing of visual and verbal information (the visuo-spatial sketchpad and the phonological loop, respectively) and a central executive that controls and coordinates the processing of information. However, similar to the case of mental representations, CLT has no explicit assumptions about the architecture of working memory and its relation to cognitive load. Most of the processes involved in learning activities, such as integration and storing, are not exclusively carried out in the subsystems of the working memory but also include the central executive. The connections of CLT to this or another working memory model need to be developed in more detail, clarifying, for example, the relation of the working memory subsystems to CLT’s visual and verbal processing channels. The issues that we raised about the additivity hypothesis may be partially the result of the lack of such a connection. In Baddeley’s (1986, 2000) model, the two subsystems are not specific to the modality of information (i.e., through which of our senses the information is perceived), but rather to its mode of representation (i.e., whether it uses visual or verbal representations). For example, verbal information is always processed in the phonological loop, independent from its modality (text or narration). This model, therefore, cannot be used as the basis of the cognitive load explanation for the modality effect, which assumes that written and spoken language are processed in separate channels (Rummer, F¨urstenberg, & Schweppe, 2008; Rummer, Schweppe, Scheiter, & Gerjets, 2008; Rummer et al., 2010).
262
Roland Br¨unken, Jan L. Plass, and Roxana Moreno
In summary, the relation between CLT and current working memory theories needs clarification. Because the construct of working memory is fundamental to the idea of cognitive load, future theoretical developments of CLT should take a closer look at the basic foundations of working memory as informed by experimental psychology and neuropsychology research. The lack of theoretical clarity in this area shows a conflict between CLT predictions and those that would result from taking into consideration the empirical evidence of basic research on working memory (Rummer et al., 2010; Seufert et al., 2009). Motivation, Affect, Metacognition, and Cognitive Load A final issue related to the conceptual development of CLT that we would like to address is its restriction to the cognitive aspect of the learning process. The same limitation inspired the revision of Mayer’s (2005) original cognitive theory of multimedia learning (CTML), which has recently been reframed as a cognitive-affective theory of learning with media (CATLM; Moreno, 2005, 2009, in press; Moreno & Mayer, 2007). In this model, affective, cognitive, and metacognitive factors are integrated to explain learning from different instructional methods and media. Although it is well known that metacognitive, affective, and motivational constructs are central to learning, they have not been the focus of cognitive load research (Bannert, 2006; Paas, Tuovinen, van Merri¨enboer, & Darabi, 2005). Therefore, there is great potential to test specific hypotheses about the relation among motivation, cognition, cognitive load, and learning to advance CLT. Are cognitive load effects affected by students’ motivation? Can students’ use of learning strategies and metacognitive control be described in terms of germane load, bridging two major research areas of educational psychology? Although empirical evidence exists that could inform this line of research, it has not yet been systematically investigated within a CLT context. In summary, although the strength of CLT is to have proposed the relations among the human cognitive architecture, learning, and instruction, there are several questions regarding the conceptual development of CLT that suggest that the theoretical base of CLT is in need of further development. CLT is a theory in its own right, with its own questions and empirical base, yet its applied nature requires a more careful evaluation of the basic research assumptions on which it relies. Future advances in the theoretical development of CLT should incorporate recent findings from neuroscience, neuropsychology, and brain research. CLT has the potential to become a
Current Issues and Open Questions in Cognitive Load Research
263
powerful framework for bridging basic and applied research on learning and instruction. To reach its potential, however, CLT’s basic concepts need to be made more precise, ill-defined concepts revised, and implicit assumptions made explicit.
cognitive load methodological developments The relations among cognitive load, mental resources, and learning proposed by CLT raise important methodological questions about how to validate any prescriptive design effects in a clear and reliable manner. This section summarizes the methodological developments in the cognitive load field with their corresponding strengths and limitations. We focus on four issues of the cognitive load methodology: the experimental research paradigm, the question of load measurement, the assessment of learning outcomes, and the relation to methodological approaches from other comparable research fields. The Experimental Research Paradigm As a psychological theory of learning and instruction, CLT is subject to verification by a strictly empirical, experimental approach. Almost all research in CLT is conducted in highly controlled experimental laboratory studies with highly structured materials in well-defined domains, such as mathematics, science, or technical areas (Sweller, 1999). Although this type of research is in line with the common paradigm in basic psychological research, it is not without alternatives within applied psychological or educational inquiry. For example, competing theoretical approaches to learning and instruction coming from a constructivist point of view are using a wide variety of alternative research methods, which are often heavily criticized by cognitive load researchers (Kirschner, Sweller, & Clark, 2006). However, it is necessary to ask what the specific gains and losses of the chosen methodological approach in cognitive load research are. Moreover, it is necessary to consider whether a broader methodological approach could contribute to the development of CLT. The main concern regarding experimental research in applied psychology is associated with the question of generalization of empirical findings to authentic learning scenarios. With respect to CLT, there are two primary aspects that have to be taken into account: the limited number of subject domains in which CLT has been tested, and the limited range of learning scenarios that have been under experimental investigation. Although
264
Roland Br¨unken, Jan L. Plass, and Roxana Moreno
many experiments have been carried out in well-defined highly structured domains, such as mathematics, statistics, or technical areas, little is known about the applicability of CLT to ill-structured, open domains, such as social science, history, or philosophy. For example, in the domain of history, learning from visual external representations is of high importance; however, the function of a visual representation, such as a historic picture, is likely to vary from that of the visualization of a technical system, such as a car brake system. Little is known about the processing of multiple external representations in history, and more research is needed to determine whether the cognitive load effects found with technical visualizations can be generalized to domains such as history. The second question arises from the differences in treatment durations between experimental and authentic learning settings. Within most of cognitive load research, laboratory learning scenarios include brief sequences of instruction and immediate post-test assessments of knowledge acquisition. However, this type of treatment is not representative of the way learning takes place in real-life scenarios, such as in classrooms or apprenticeships, which are characterized by longer periods of learning and delayed testing, sometimes weeks after the learning phase. It is unclear whether cognitive load effects are likely to persist in longer and more complex learning scenarios. Issues of Cognitive Load Measurement Chapter 9 (this volume) describes the current methods of cognitive load measurement based on direct or indirect observations of learner behavior related to the cognitive demands of a learning situation. The chapter argues that most of these methods produce scores that are of acceptable reliability and are correlated with the learning outcomes in the expected way. However, the question of validity needs further investigation. For example, with respect to the dual task methodology, it is not clear if performance on the secondary task is determined by lower-level processing, such as visual search or processes of attention, or by higher-level processing, such as selecting or organizing information in working memory. Likewise, it is not clear if selfreports of perceived mental effort are valid measures of cognitive load, as they are likely to be affected by students’ motivation and affect. Because the cognitive load construct is underspecified, as discussed earlier, it is not possible to assess the construct validity of these measures. Unless the definition of cognitive load is clarified, the measurement of cognitive load cannot be improved.
Current Issues and Open Questions in Cognitive Load Research
265
A second problem of cognitive load measurement, which in our view has placed current cognitive load research at an impasse, is the lack of methods that allow the differentiated measurement of intrinsic, extraneous, and germane load types (Br¨unken, Plass, & Leutner, 2003, 2004; Moreno, 2006). To date, there are no widely accepted measures available that reliably distinguish among these three dimensions. Most attempts to do so with selfreport measures show high correlations between the items that presumably tap into separate load types, suggesting that the cognitive load construct is one-dimensional (Paas, Tuovinen, Tabbers, & Van Gerven, 2003). Moreover, as Paas demonstrated in earlier cognitive load research, using multiple-item scales has no advantage compared with a simple one-item scale for cognitive load measurement (Paas & van Merri¨enboer, 1993). Since then, no substantial progress has been made on self-reported measures of cognitive load constructs. Unfortunately, our understanding of cognitive load is not likely to advance unless we start overcoming these measurement limitations. Furthermore, is CLT’s impasse due to measurement limitations, or do the measurement limitations signal a fundamental problem in the theory’s assumptions (Moreno, 2006)? In our view, answering this question is one of the main challenges in cognitive load research in the coming years, and there is only a very small number of studies that attempt to answer this question empirically (DeLeeuw & Mayer, 2008). The results from DeLeeuw and Mayer seem to suggest that different measurement approaches are sensitive to different aspects of cognitive load, but it remains unclear whether these differences are caused by real conceptual differences of the load constructs or whether they are artifacts of the shortcomings of the measurement techniques. Issues of Learning Measurement CLT is aimed at helping us understand how instruction should be designed to effectively support learning. But what exactly does “effective” instruction refer to? Does it mean making learning easier or effortless? Does it mean making learning more interesting? Is it about learning faster or promoting deeper levels of understanding? Or does effective learning consist of a combination of these factors? Many other factors could be added to this list. However, the major questions behind this issue are of a methodological nature: What are appropriate variables to be included in cognitive load research, and what can the learning outcome measures included in a research study tell us about the cognitive load experienced by the learners?
266
Roland Br¨unken, Jan L. Plass, and Roxana Moreno
Most research studies measure learning on different levels of processing, such as retention, comprehension, and transfer (Mayer, 2005). Several studies show interesting differences in cognitive load effects with respect to the types of learning outcome tests used: retention effects are usually small or zero in cognitive load studies that investigate methods to reduce extraneous CL. Far transfer effects are usually small or zero in cognitive load worked-example studies using methods to increase germane cognitive load (Seufert et al., 2007). This indicates that reducing extraneous load is important for higher-level learning, whereas the procedures used to foster germane processes are highly content-specific and seem to allow only for limited transfer. However, systematic research with respect to the relation of cognitive load effects and types of learning outcome tests is still in the beginning stages. Some experiments show the dependency of cognitive load effects on test difficulty as well as on the presentation format of test items (see, e.g., the meta-analysis on dynamic visualizations by H¨ofler & Leutner, 2007). In our own research on the multimedia effect, we observed an interaction between the presentation modality of information (visual vs. verbal) and the presentation modality of the test items (visual vs. verbal; Br¨unken, Steinbacher, Schnotz, & Leutner, 2001). In more recent studies on dynamic visualizations, we found a similar interaction between the presentation format of information (static vs. dynamic) and the type of test items (process oriented vs. structure oriented; M¨unzer, Seufert, & Br¨unken, 2009). Such results indicate that more research is needed in this area to understand which type of learning can be effectively supported by which type of presentation. However, several questions remain. Are the learning outcome measures aligned with the instructional methods used in the studies? Most research includes verbal measures of learning. If the learning environments typically consist of multimedia presentations, with static and animated pictures and video instruction components, is the modality of the learning assessments psychologically neutral to cognitive load? Or, regarding the type of learning (intentional vs. incidental), has the student focused on aspects of the presentation that were not deemed essential by the instructional designer? Learners may use their cognitive resources to process irrelevant or less relevant materials. To know exactly whether and how the cognitive resources were spent would require measuring not only learning of the target concepts and skills but also any other incidental learning that might have occurred during the lesson. Such questions have to be taken into account when CLT makes predictions about instructional effectiveness.
Current Issues and Open Questions in Cognitive Load Research
267
Moreover, which additional variables should be considered and when should they be assessed? An interesting approach has been introduced to CLT by Paas and colleagues by measures of learning efficiency, using the ratio of time on task and cognitive load as performance indicator (Paas & van Merri¨enboer, 1993; van Gog & Paas, 2008). This approach considers that instructional efficiency is not only determined by the amount of knowledge acquired but also by factors describing the consumption of resources, such as learning time or cognitive effort. However, in their present conceptualization, efficiency measures are based on the assumption that learning is most efficient when high learning outcomes are combined with low resource consumption, and it remains to be shown whether this approach will indeed result in the design of learning environments with the highest impact. Many of the recent cognitive load research studies used valid and reliable measures of cognitive load and learning outcomes. However, as we argued earlier, to infer the cognitive load induced in any learning situation would also require accounting for variables that are likely to affect the experienced cognitive load for any one learner, such as the level of prior knowledge, dispositional and situational interest, anxiety, perceived support, and metacognitive and cognitive skills and styles. The need to measure these relevant constructs raises some methodological as well as practical questions. For example, should pre-tests be considered valid measures of students’ “expertise,” or do the tests only reveal their readiness to build new knowledge? Moreover, if individual differences are of such importance, how can we handle this in concrete learning scenarios? Again, some modern approaches, such as rapid online assessment of prior knowledge (Kalyuga & Sweller, 2005), show promising new ways, but they also represent new challenges in cognitive load measurement. For example, as argued earlier in this chapter, the concept of intrinsic load is highly correlated with learner expertise. Therefore, using dynamic online assessments of expertise, an adaptive alignment of intrinsic load to the learner’s level of expertise, could facilitate new ways of designing adaptive learning systems that overcome the problems of existing approaches that use sophisticated user modeling, but which have never worked satisfactorily (Leutner, 1992). In summary, cognitive load research should address questions of the methodological approaches used in conducting cognitive-load–related studies. It also needs to focus on the validation of existing and the development of new cognitive load measures that are capable of discriminating the three types of load, based on a clarification of the cognitive load construct itself. Finally, cognitive load research needs to more systematically relate
268
Roland Br¨unken, Jan L. Plass, and Roxana Moreno
measures of learning outcomes to specific instructional strategies, types of knowledge representations, and learner characteristics.
conclusion Without any doubt, CLT has become one of the most influential theoretical frameworks of educational psychology in recent years, inspiring researchers all over the world to conduct an enormous amount of high-quality experimental research on how to design instruction in an efficient learner-oriented way. CLT has been widely used as the theoretical framework for several instructional design areas, such as complex problem-solving environments, worked-example instruction, and multimedia learning. Which direction will CLT take in the next decade? In our view, the ongoing discussions in the cognitive load community suggest that new theoretical developments are on their way. More specifically, the following three lines of thought are currently being discussed by cognitive load researchers as possible areas of future extension of the theory: (1) the evolutionary foundation of CLT, (2) the integration of motivational aspects of learning, and (3) the integration of new basic research paradigms into cognitive load measurement. In past years, John Sweller, the founder of CLT, has discussed the relations between CLT and human cognitive architecture from an evolutionary perspective (see Chapter 2, this volume), linking questions of instructional design to restrictions of human information processing that are caused by the demands of evolution. However, as fascinating as these ideas are, they remain highly speculative with respect to their empirical base. Moreover, they may be useful for an understanding of the basic constraints of CLT, but their added value to answering the pressing questions of CLT is not clear. CLT is based on cognitive principles of learning. However, learning is not a purely cognitive phenomenon. It also depends on metacognitive, motivational, and emotional factors, such as task engagement, interest, and the learners’ beliefs and emotions. Although the contribution of these factors to learning has been well established by the learning sciences for quite a long time, they are mostly ignored in cognitive load research. Recent developments of CLT appear to remain focused on cognitive aspects, yet attempt to incorporate more advanced methods of investigation. Actual research presented at international conferences increasingly uses advanced methods such as eye-tracking techniques for the observation of
Current Issues and Open Questions in Cognitive Load Research
269
learner behavior during information processing of multiple representations or worked examples. For example, van Gog and Paas (2008) presented an interesting approach that used experts’ eye movements as a model for novice learners. Within that approach, eye tracking is not only used as a tool for observation, but also as a means for instruction. Although the first results are promising, more research is needed to test the sustainability of this approach. Nevertheless, eye tracking seems to be in vogue in educational research, regardless of its methodological problems, such as whether the duration of a fixation on a specific area of interest indicates information complexity, interest, or simply low readability. This and other methodological questions have to be answered before eye tracking will become a standard procedure in cognitive load research. However, beyond actual research, a trend seems to have become apparent: CLT is trying to link back to its roots in cognitive psychology, which has made substantial progress in the last decade. Whereas the last decade of CLT research was marked by a focus on working memory research, future developments should deal with questions of germane load, information organization, and knowledge representation. This could lead to a better understanding of germane processes and their support by instructional means. references Atkinson, J. (1958). Towards experimental analysis of human motivation in terms of motives, expectancies and incentives. In J. Atkinson (Ed.), Motives in fantasy, action and society (pp. 288–305). Princeton, NJ: Van Nostrand. Baddeley, A. D. (1986). Working memory. Oxford, UK: Oxford University Press. Baddeley, A. D. (2000). The episodic buffer: A new component of working memory. Trends in Cognitive Sciences, 4, 417–423. Bannert, M. (2006). Effects of reflection prompts when learning with hypermedia. Journal of Educational Computing Research, 4, 359–375. Bartlett, F. (1932). Remembering: A study in experimental and social psychology. London: Cambridge University Press. Br¨unken, R., Plass, J. L., & Leutner, D. (2003). Direct measurement of cognitive load in multimedia learning. Educational Psychologist, 38, 53–61. Br¨unken, R., Plass, J. L., & Leutner, D. (2004). Assessment of cognitive load in multimedia learning with dual-task methodology: Auditory load and modality effects. Instructional Science, 32, 115–132. Br¨unken, R., Steinbacher, S., Schnotz, W., & Leutner, D. (2001). Mentale Modelle und Effekte der Pr¨asentations- und Abrufkodalit¨at beim Lernen mit Multimedia [Mental models and the effect of presentation and retrieval code while learning with multimedia]. Zeitschrift f¨ur P¨adagogische Psychologie [German Journal of Educational Psychology], 15, 16–27.
270
Roland Br¨unken, Jan L. Plass, and Roxana Moreno
Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24, 87–114. DeLeeuw, K. E., & Mayer, R. E. (2008). A comparison on three measures of cognitive load: Evidence for separable measures of intrinsic, extraneous, and germane load. Journal of Educational Psychology, 100, 223–234. Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102, 211–245. H¨offler, T.N., & Leutner, D. (2007). Instructional animations versus static pictures: A metaanalysis. Learning & Instruction, 17, 722–738. Johnson-Laird, P. N. (1983). Mental models: Towards a cognitive science of language, interference, and consciousness. Cambridge, UK: Cambridge University Press. Kalyuga, S., Chandler, P., & Sweller, J. (1998). Levels of expertise and instructional design. Human Factors, 40, 1–17. Kalyuga, S., & Sweller, J. (2005). Rapid dynamic assessment of expertise to improve the efficiency of adaptive e-learning. Educational Technology, Research and Development, 53 (3), 83–93. Kirshner, P., Sweller, J., & Clark, R. (2006). Why minimal guidance during instruction does not work: An analysis of the failure of constructivist, discovery, problembased, experiential, and inquiry-based teaching. Educational Psychologist, 41, 75–86. Leutner, D. (1992). Adaptive Lehrsysteme. Instruktionspsychologische Grundlagen und experimentelle Analysen [Adaptive learning systems]. Weinheim, Germany: Beltz. Mayer, R. E. (2001). Multimedia learning. New York: Cambridge University Press. Mayer, R. E. (2005). Introduction to multimedia learning. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (pp. 1–18). Cambridge, UK: Cambridge University Press. Mayer, R. E., & Chandler, P. (2001). When learning is just a click away: Does simple user interaction foster deeper understanding of multimedia messages? Journal of Educational Psychology, 93, 390–397. Mayer, R. E., & Moreno, R. (1998). A split-attention effect in multimedia learning: Evidence for dual processing systems in working memory. Journal of Educational Psychology, 90, 312–320. Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81–97. Miyake, A., & Shah, P. (Eds.) (1999). Models of working memory: Mechanisms of active maintenance and executive control. New York: Cambridge University Press. Moreno, R. (2005). Instructional technology: Promise and pitfalls. In L. PytlikZillig, M. Bodvarsson, & R. Bruning (Eds.), Technology-based education: Bringing researchers and practitioners together (pp. 1–19). Greenwich, CT: Information Age Publishing. Moreno, R. (2006). When worked examples don’t work: Is cognitive load theory at an impasse? Learning and Instruction, 16, 170–181. Moreno, R. (2009). Learning from animated classroom exemplars: The case for guiding student teachers’ observations with metacognitive prompts. Journal of Educational Research and Evaluation, 15(5), 487–501.
Current Issues and Open Questions in Cognitive Load Research
271
Moreno, R. (in press). Cognitive load theory: More food for thought. Instructional Science. Advance online publication. doi: 10.1007/s11251-009-9122-9. Moreno, R., & Mayer, R. E. (2007). Interactive multimodal learning environments. Educational Psychology Review, 19, 309–326. M¨unzer, S., Seufert, T., & Br¨unken, R. (2009). Learning from multimedia presentations: Facilitation function of animations and spatial abilities. Learning and Individual Differences, 19(4), 481–485. Paas, F., & van Merri¨enboer, J. J. G. (1993). The efficiency of instructional conditions: An approach to combine mental-effort and performance measures. Human Factors, 35, 737–743. Paas, F. G. W. C., & van Merri¨enboer, J. J. G. (1994). Variability of worked examples and transfer of geometrical problem solving skills: A cognitive load approach. Journal of Educational Psychology. 86, 122–133. Paas, F. G. W. C., Renkl,. A., & Sweller, J. (2003). Cognitive load theory and instructional design: Recent developments. Educational Psychologist, 38, 1–4. Paas, F., Tuovinen, J., Tabbers, H., & Van Gerven, P. W. M. (2003). Cognitive load measurement as a means to advance cognitive load theory. Educational Psychologist, 38, 63–71. Paas, F., Tuovinen, J., van Merri¨enboer, J. J. G., & Darabi, A. (2005). A motivational perspective on the relation between mental effort and performance: Optimizing learner involvement in instruction. Educational Technology, Research & Development, 53, 25–33. Paivio, A. (1986). Mental representations: A dual coding approach. New York: Oxford University Press. Pintrich, P., & Schunk, D. (2002). Motivation in education: Theory, research, and applications (2nd ed.). Upper Saddle River, NJ: Prentice-Hall. Plass, J. L., Homer, B. D., & Hayward, E. (2009). Design factors for educationally effective animations and simulations. Journal of Computing in Higher Education, 21(1), 31–61. Plass, J. L., Homer, B. D., Milne, C., Jordan, T., Kalyuga, S., Kim, M., et al. (2009). Design factors for effective science simulations: Representation of information. International Journal of Gaming and Computer-Mediated Simulations, 1(1), 16–35. Rummer, R., F¨urstenberg, A., & Schweppe, J. (2008). Lernen mit Texten und Bildern: Der Anteil akustisch-sensorischer Information am Zustandekommen des Modalit¨atseffekts [The modality effect in learning with texts and pictures: On the impact of the auditory recency effect]. Zeitschrift f¨ur P¨adagogische Psychologie, 22, 37–45. Rummer, R., Schweppe, J., F¨urstenberg, A., Seufert, T., & Br¨unken, R. (2010). Working memory interference during processing texts and pictures: Implications for the explanation of the modality effect. Applied Cognitive Psychology, 24(2), 164–176. Rummer, R., Schweppe, J., Scheiter, K., & Gerjets, P. (2008). Lernen mit Multimedia: Die kognitiven Grundlagen des Modalit¨atseffekts [Learning with multimedia: The cognitive foundations of the modality effect]. Psychologische Rundschau, 59, 98–108.
272
Roland Br¨unken, Jan L. Plass, and Roxana Moreno
Schnotz, W., Boeckheler, J., & Grzondziel, H. (1999). Individual and co-operative learning with interactive animated pictures. European Journal of Psychology of Education, 14(2), 245–265. Schnotz, W., & K¨urschner, C. (2007). A reconsideration of cognitive load theory. Educational Psychology Review, 19, 469–508. Seufert, T., & Br¨unken, R. (2006). Cognitive load and the format of instructional aids for coherence formation. Applied Cognitive Psychology, 20, 321–331. Seufert, T., J¨anen, I., & Br¨unken, R. (2007). The impact of intrinsic cognitive load on the effectiveness of graphical help for coherence formation. Computers in Human Behavior, 23, 1055–1071. Seufert, T., Sch¨utze, M., & Br¨unken, R. (2009). Memory characteristics and modality in multimedia learning: An aptitude-treatment-interaction study. Learning and Instruction, 19, 28–42. Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12, 257–285. Sweller, J. (1999). Instructional design in technical areas. Camberwell, Australia: ACER. Sweller, J. (2004). Instructional design consequences of an analogy between evolution by natural selection and human cognitive architecture. Instructional Science, 32, 9–31. Sweller, J., & Sweller, S. (2006). Natural information processing systems. Evolutionary Psychology, 4, 434–458. van Gog, T., Jarodzka, H., Scheiter, K., Gerjets, P., & Paas, F. (2008). Guiding students attention during example study by showing the model’s eye movement. In J. Zumbach, N. Schwarz, T. Seufert, & L. Kester (Eds.), Beyond knowledge: The legacy of competence (pp. 189–196). Berlin: Springer. van Gog, T., & Paas, F. (2008). Instructional efficiency: Revisiting the original construct in educational research. Educational Psychologist, 43, 16–26. Wallen, E., Plass, J. L., & Br¨unken, R. (2005). The function of annotations in the comprehension of scientific texts – Cognitive load effects and the impact of verbal ability. Educational Technology, Research & Development, 53, 3, 59–72. Wittrock, M. C. (1989). Generative processes of comprehension. Educational Psychologist, 24, 345–376.
index
active processing principle, 132, 153, 169, 231 assumptions, 171 activity, behavioral versus cognitive, 154 affect, 11, 262 borrowing and reorganising principle, 33–35 chunking, 135 cognitive-affective theory of learning with media, 157, 160, 262 cognitive architecture, 14, 29, 111 cognitive load construct, 9 interactions between sources, 44–45 measurement, 74, 80, 182, 187, 188, 190, 196, 208, 210, 212, 215, 219, 264 offloading, 259 relation to goals of instruction, 191, 207, 258 relation to learning, 55, 78, 79, 185, 189, 194, 213, 255 sources, 52, 197 cognitive load theory additivity hypothesis, 16, 18, 198, 258–259 assumptions, 12, 14, 16, 18, 111, 254 complex learning scenarios, 41, 114, 118, 126, 186, 264 evolutionary interpretation. See evolutionary framework experimental research paradigm, 263 framework, 11 instructional methods, 207 limitations, 10, 14–15, 183, 210, 218, 264 relation to cognitive theory of multimedia learning, 133–134
relation to learning theories, 20–21, 205, 206 cognitive overload, 136 cognitive skill acquisition, stages, 94–96 cognitive theory of multimedia learning, 188, 230–232, 260, 262 assumptions, 230–231 coherence principle, 136–137, 245, 255 color coding principle, 261 completion principle, 13, 120 constructivism, relation to CLT, 21–22, 42–43, 263 direct initial instruction principle, 56–57 direct instruction, 162 discovery learning, 162 distributed cognition, relation to CLT, 21 dual channel assumption, limitations of, 232 dual channel principle, 132, 230 dual coding theory, 260 dual task methodology, 264 element interactivity, 16, 30, 40–42, 52–53, 118, 123, 135, 184, 195, 256 engagement, 73, 196, 256 environment organising and linking principle, 38–39 essential overload, 135 essential processing, 133, 135 evolutionary framework, 20 centrality of the five principles, 39 human cognition, 29 expertise reversal principle, 30, 44, 57–58, 68, 70, 71, 94, 120, 214 knowledge-gain reversal principle, 96–100
273
274
Index
extraneous cognitive load, 12, 18, 42–43, 53–54, 68, 70, 133, 197, 210, 241, 256 in multimedia learning, 134 measurement, 214, 265 methods of minimization, 55 questions concerning definition, 257 relation to self-regulation, 76 relation to spatial abilities, 72 sources, 54 extraneous processing, 133 fading, adaptive procedure, 104, 105 feedback principle, 164, 165 four-component instructional design model, 60–61, 94, 124 generative processing, 133, 154, 155, 167, 169 germane cognitive load, 17, 18, 43–44, 53, 126, 133, 154, 197, 210, 241, 256 measurement, 265 methods of induction, 114, 115, 117, 120 questions concerning definition, 257 sources, 53 goal-free principle, 13, 30 guidance fading principle, 30 guided activity principle, 161, 162–163 dialoguing, 161, 162 manipulating, 161, 162 iconic representations, 232 imagination principle, 30 individual differences, 11, 157, 165, 195, 209, 212, 213, 257, 267 individual differences principle, 247–248 information store principle, 33 instructional systems, 79 integrated model of text and picture comprehension, 233–238, 260 assumptions, 233 auditory picture (sound) comprehension, 237 cognitive architecture, 233 cognitive theory of multimedia learning, compared to, 238, 239–240, 249 descriptive/depictive channels, 233 descriptive/depictive representations, 232–233 lower and higher order processing, 238–239 mental models, 235, 239, 240
perceptual and cognitive processing levels, 234 propositional representations, 235, 239 spoken text comprehension, 235 structure mapping, 240 verbal and pictorial filters, 235 visual picture comprehension, 236–237 written text comprehension, 234–235 interactive learning environments, 154 intrinsic cognitive load, 15, 18, 40–42, 52–53, 75, 116, 126, 133, 183, 186, 197, 241, 256 alteration due to learning, 42 in multimedia learning, 143–148 irreducible nature, 18 management of, 118 measurement, 265 questions concerning definition, 256 relation to spatial abilities, 74 isolated/interacting elements principle, 30, 59, 119 knowledge biologically primary, 29 biologically secondary, 31 learner characteristics. See individual differences learning cognitive approaches, 12, 210 comprehension, 266 measurement, limitations of, 265–268 retention, 266 transfer, 116, 266 learning efficiency, 267 learning element, 67 limited capacity principle, 132, 230 long-term memory, 14, 32, 33, 36, 38, 48 long-term working memory, 38 meaningful learning, 132 means–ends analysis, 11 mental load, 10, 183, 191, 212, 214 mental representation, 258, 259–260 metacognition, 262 modality principle, 30, 146, 148, 188, 245–246, 259 motivation, 11, 165, 192, 255, 262 multimedia learning, 132–133 multimedia learning environments goals for design of, 134
Index multimedia principle, 156, 157, 242, 266 exceptions to, 242 narrow limits of change principle, 35, 37–38 pedagogical agents, 160, 163 personalization principle, 159–160 pretraining principle, 145 prior knowledge, 11, 36, 44, 60, 75, 135, 170, 195, 209, 213, 248, 257 rapid online assessment, 267 problem solving, 11, 210, 212 pupil dilation, 211, 212, 213, 215 measurement, 215 randomness as genesis principle, 35–36 redundancy principle, 13, 30, 138–139, 243 general, 243, 244–245 specific, 243, 244 reflection principle, 166, 167, 168 scaffolding pictorial, 74 verbal, 74 schema acquisition, 11, 12, 14, 34, 48, 51, 56, 67, 110, 258 measurement of, 49 schema automation, 34, 51, 111 schema theory, 33, 48, 50, 259 segmenting principle, 144, 246–247, 256 self explanation, 93, 95, 102, 105, 123, 166 self regulation, 11, 75, 76, 77 relation to cognitive load, 75, 78 scaffolds, 77 sensory registers, 233
275
short-term memory, 14, 49 signaling principle, 139–140, 247 small step-size of knowledge change principle, 59–61 socio-cognitive theories, relation to CLT, 21 spatial abilities, 72, 73, 248 spatial contiguity principle, 141–142, 242 split-attention principle, 13, 30, 69, 255, 261 symbolic representations, 232 temporal contiguity principle, 140–141, 242 transfer paradox, 116 triarchic theory of cognitive load, 133–134 variable examples principle, 30 visual/verbal channels. See dual channel principle worked-example principle, 13, 93–94 worked-out example, 12, 57, 91, 92–93, 95, 255 worked-out example fading, 99, 100, 105 different fading procedure, 101 evaluation, 103–104 working memory, 133 auditory, 235 capacity, 67, 68, 209, 257 capacity, relation to learning, 241 central executive, 36, 220, 261 limitations, 37 models, 72, 230, 259, 261 phonological loop, 261 propositional, 234 relation to cognitive load, 216, 217, 220, 260 visual, 234 visuo-spatial sketchpad, 261
