Reflexive Ethnographic Science

Reflexive Ethnographic Science

This Page Intentionally Left Blank

Reflexive Ethnographic Science Robert Au nger

ALPTRW IPA E S S A Division of Rowman & Littlejield Publishers, Inc. Toronto 0 Oxford Walnut Creek Lanham New York

ALTAMIRA PRESS

A Division of Rowman & Littlefield Publishers, Inc. 1630 North Main Street, #367 Walnut Creek, CA 94596 www.altamirapress.com Rowman & Littlefield Publishers, Inc. A wholly owned subsidiary of the Rowman & Littlefield Publishing Group 4501 Forbes Boulevard, Suite 200 Lanham, Maryland 20706 PO Box 317 Oxford OX2 9RU, UK Copyright 0 2004 by ALTAMIRAPRESS All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of the publisher.

British Library Cataloguing in Publication Information Available

Library of Congress Cataloging-in-PublicationData Aunger, Robert Reflexive ethnographic science/Robert Aunger. p. cm. Includes bibliographical references and index. ISBN 0-7591-0274-0 (hardcover : alk. paper)-ISBN 0-7591-0275-9 (pbk. : alk. paper) 1. Ethnology-Philosophy. 2. Ethnology-Methodology. I. Title. GN345.A932 2003 305.8'001-dc21 2003011038 Printed in the United States of America

BrnThe paper used in this publication meets the minimum requirements of American National Standard for Information SciencesPermanence of Paper for Printed Library Materials, ANSI/NISO 239.48-1992.

CONTENTS

vii

ix

List of Tables and Figures Acknowledgments CHAPTER I

A Crisis in Confidence

CHAPTER 2 Investigating Existing Ethnographic Methods

i

21

CHAPTER 3

1s Reflexivity Necessary?

39

CHAPTER 4

The Wrong Way 0ut:Typology and Idealism

64

Reflexive Realism:A New Way of Doing Ethnography

94

CHAPTER 5

CHAPTER 6

Measuring the “Strength of Belief”

I I6

CHAPTER 7 Toward a Reflexive Ethnographic Science

130

APPENDIX A The Case Study: Food Taboos in the lturi

145

APPENDIX B Reflexivity Is Necessary

163

APPENDIX C Idealism’s Failure

215

25 I

275

28 I

References Index About the Author

V

This Page Intentionally Left Blank

TABLES AND FIGURES

Tables Table 2.1 Approaches to the Analysis of Ethnographic Data Table A.l Composition of Ituri Sample Table A.2 Coding Schemes for Ituri Food Avoidance Data Table A.3 Modal Response by Ethnic Group Table B.1 Demographic Composition of Informants in Repeat Interviews Table B.2 Paired Interviews by Ethnic Group Table B.3 The Multivariate Model Table B.4 Determinants of ‘‘Forgethlness” Table B.5 Determinants of “Mistakes” Table B.6 Determining Discrepancies Table C.l Cultural Consensus in the Ituri Table C.2 Distribution of Consensus Responses

37 149 154

155 163 164 167 173 175 177 217 224

Figures Figure C.l

Multidimensional Scaling of Ituri Consensus Datasets Figure C.2 Optimal Scaling of Sudanic Informant Characteristics

vii

227 241

This Page Intentionally Left Blank

ACKNOWLEDGMENTS

T

he most gratifying aspect of finishing a lengthy project such as this is finally being presented with the opportunity to formally express my deep and sincere thanks to those who contributed to my wellbeing during the ten years it has taken for this project to reach its present stage of fruition. First, I must express, albeit in this small way, my debt to Robert Bailey. It has largely been under his wing that I was fledged, both as a researcher and field-worker. We spent a number of years in daily contact-including nine months, mostly sharing the same room, during my first fieldwork experience-without getting too much on each other’s nerves (at least he never said anything to me). His fairness and generosity as an adviser and friend are beyond my ability to emulate. Particular thanks also to Richard Grinker for sharing a field season, his field notes, his knowledge of the Lese language, and for writing insightful information on Ituri social life for me to cobble. Thanks to my family for being my family and allowing me time to work on the book rather than on their happiness. My wife has been vital in this regard, by taking on many roles that would have otherwise delayed this publication even further. She has also performed the vital function of reading the entire manuscript in its final stages, making numerous improvements. Asante sana kabisa also to the myriad Lese, Budu, Efe, and Tswa who agreed to participate in the study, and allowed me to live as a socially marginalized but economically significant member of their community. I could not have completed the present study without the help of the eighteen Lese and Budu men who assisted in the collection of data: merci mingi. For almost twenty years now, the Ituri Project has changed the lives of everyone in the area where we work-even more irrevocably for those who

ACKNOWLEDGMENTS

have chosen to align themselves more intimately with the Project. I only hope that now, through the agency of the Ituri Fund, we can bring greater stability of circumstance, as well as increased access to education and medical care, to everyone in the area, despite their currently deteriorating material and political circumstances. In particular, I’d like to thank David Wilkie for his unstinting work on the Ituri Fund. (Any royalties derived from the publication of this book will go to the Fund.) Contributions are accepted from any quarter whatever, and are tax deductible, as well as being genuinely humane. I thank Steven Borgatti and Malcolm Dow for assistance in interpreting ANTHROPAC output; Kim Romney for bibliographic assistance; and Nancy Alvarado, Devon Brewer, Linda Garro, and Kim Romney for providing reprints and preprints. Previous versions of parts of this book have been immeasurably improved by readings from Nancy Alvarado, Robert Bailey, Gillian Bentley, Peter Richerson, and Douglas White. Russ Bernard, Douglas Caulkins, and an anonymous reviewer provided very valuable comments on the entire book. Field research in the Ituri was supported by National Science Foundation Grants to Robert Bailey, a graduate fellowship from the.University of California, a Lambe fellowship from the Institute for Humane Studies, and a grant from Sigma Xi. Henry Harpending provided access to the Pennsylvania State University mainframe computer. I also want to thank L‘Institut des Musees Nationaux du Zaire, under the directorship of Dr. Lema Guete, for assistance in the field. Thanks to Rosalie Robertson and Mitch Allen, my editor and the publisher at AltaMira, respectively, for their enthusiasm for this book. Thanks also to copy editor Wendy Schlosberg for her excellent work and to Natasha Gunawardana for producing the index, with assistance from Jasmine Kershaw. Portions of chapter 3 were modified from previously published material in “Against IdealisdContra consensus” in Current Anthropology 40:S93-S101 (1999). Portions of Appendix B were modified from previously published material in “Sources of variation in ethnographic interview data: Food avoidances in the Ituri Forest, Zaire,” in Ethnology 33:65-99 (1994). Portions of chapters 5 and 7 were modified from previously published material in “On ethnography: Story-telling or science?”CurrentAnthropology 36:97-130 (1995). X

I

A CRISIS IN CONFIDENCE

E

thnography, or the representation of unfamiliar groups’lifeways in the form of written records, has a long and distinguished history ‘in cultural anthropology. Indeed, it remains the epitome of professional work in this field. However, in the last ten years or so, a number of scandals have rocked the ethnographic community. Major debates have been sparked about the nature of ethnographic authority and the validity of ethnographic descriptions, sending the discipline into a tailspin of selfdoubt and internecine recrimination. A recent New York Times article summarizes the popular impression of this academic infighting: Virtually all of the field’s leading figures have been struck by poison arrows. Margaret Mead?Dupe! Franz Boas? Spy! Colin Turnbull? Hoaxer! Marshall Sahlins?Imperialist! Indeed, the excessive ferocity of anthropological warfare has fractured the discipline and tarnished its public image. It’s become the academic equivalent of “The Jerry Springer Show.”

It’s not clear how ethnography can recover its dignity. The problems began when later generations of ethnographers began to return to classic field sites and found different cultures in situ-and it wasn’t just that aborigines were now using mobile phones. Instead, fundamental facts about how people live-their sexual proclivities or altruistic tendencies-were turned upside down. Previously, of course, the authority of an ethnographic report was assumed to be, and effectively was, inviolate due to the lack of alternative viewpoints available for one to consult about a particular cultural group. However, Derek Freeman’s (1983) widely publicized attack on the validity of Margaret Mead’s (1928) interpretation of Samoan adolescent sexuality vividly brought to 1

CHAPTER 1

public attention the possibility that different anthropologists could come to opposing conclusions regarding the nature of the same society. The controversy concerning the validity of Freeman’s criticisms of Mead still continues (for overviews, see Caton 1990; Cote 1992; Foerstel and Gilliam 1992; Holmes 1987; Orans 1996). Perhaps the first of these very public major scandals in ethnography was between Robert Redfield and Oscar Lewis. Redfield began his dissertation fieldwork in a Mexican village during the same year that Mead published the famous results of her research in Samoa. After eight months of research, he produced what was the first real anthropological study of a modern peasant community. Redfield (1930) contended that life in Tepoztlan was characterized by rich family life, communal solidarity, and cultural homogeneity. This picture of rural life as simple, free, and peaceful was apparently not affected by the fact that Redfield had to cut short his research due to a revolution in the village, which made it unsafe for him to continue his work. Lewis returned to the village where Redfield did his research in 1947. Lewis (1951) presented a very different view of what was happening there. Where Redfield saw happy, natural relationships between villagers, Lewis saw factionalism and social inequality. Lewis wondered why so many villagers seemed to be fleeing the rural area if life was so pleasant thereespecially when, as Redfield himself claimed, rural migrants to a large urban center are often left disoriented and marginalized by its individualism and depersonalized nature. The weight of evidence from succeeding generations of anthropologistsdoing research in Mexico has largely mounted against the romantic picture of Tepoztlan presented by Redfield. The “folk-urban continuum”he identified (in which country life was seen as idyllic and urban life stressful, evil, and artificial) has itself been shown to be a Rousseauian idyll. Just as Mead’s experience seemed largely colored by the theoretical preconceptions of her mentor Franz Boas, Redfield’s ethnographic report appeared to strongly reflect biases he took with him to the field. A more recent example of the same kind of problem arose when Curtis Abraham visited the Ika of Uganda, a group studied first by Colin Turnbull. According to Abraham (1998), the Ika are not the inhuman monsters liable to ignore a starving infant and hoard a precious store of food for themselves as portrayed by Turnbull in his book The Mountain People. Instead, Abraham argues this group has simply found a way to survive as a group in very harsh ecological conditions. 2

A CRISIS IN CONFIDENCE

The careers of several of the most famous English-speaking cultural anthropologists of the twentieth century have thus now been sullied by claims from other field-workers who returned to their field sites many years later. In all these cases, accusations of rather shoddy ethnographic practices are coupled with very different pictures of life in the locale in question. Mead’s enthusiasm for Samoan adolescent freedom was probably born out of excessive youthful romanticism rather than a deliberate desire to deceive everyone or push an agenda. Redfield may have simply been misled by the brief time he spent in Tepotzlan, and Turnbull’s ethnography of the hapless Ika was probably an unconscious reaction to the forest idyll he had found earlier in the life of central African pygmies, as portrayed in his famous Forest People. Although the villages of Tepotzlan, Samoans, and Ika are independent cases, reformists of each case claim that the earlier ethnographers went to the field with a definite message already in mind. It seems unlikely that the revisionists would have come up with counterimages were it not for distortions in the nature of the views promulgated by their very famous predecessors. But the seeming inability of even our most talented ethnographers to “get it right” has fostered the opinion that it is impossible for anyone to construct a reliable picture of ethnographic life. Does this signal the end of ethnography as an intellectual endeavor? O r can we dismiss the revisionists as mistaken in their apprehensions or simply bent on creating controversy? In an important sense, these clashes between titans are beside the point. No one can really be declared a victor in these battles for verisimilitude. Did Mead’s informants lie? Was she duped? Did she create out of whole cloth a picture of life she herself fantasized about from some youthful romanticism? Does Freeman exaggerate the opposite aspects of Samoan life from Mead just to make his own account stand out from her earlier one? The real issue isn’t so much who is right and who is wrongbecause the truth w ill always be contested-but whether any ethnography has value. The synergy between an ethnographer and his or her people is always a unique reflection of the combination of character, context, and contact. And anyone going to the field has blinders of various sorts, from the personal to the theoretical. So any ethnography is of necessity going to be idiosyncratic in some respects. A third ethnographic visitor to these field sites would no doubt discover a third “truth” about the 1ocations.l What use can we make of any of these reports then? 3

CHAPTER 1

Despite the impression of chaos and disruption it has inspired among onlookers, this continuing public uproar over ethnographic authority has had a salutary effect within the discipline. It has encouraged ethnographers to reexamine their methods. In effect, they have subjected “anthropological thought itself to ethnographic description and ethnological understanding” (Scholte 1974,437). In particular, a school critical of traditional ethnography began to consider ethnographies as texts to determine how these documents create an objective representation of unfamiliar lifeways in the minds of readers (for examples, see Fabian 1983; Manganaro 1990; Marcus and Fischer 1986; Sanjek 1990; van Maanen 1988). I will call these critics “textualists”because most of them draw their inspiration from the hermeneutic tradition of textual criticism in the humanities and European social theory, primarily deconstruction and hermeneutics, but anthropological structuralism as well. The general upshot of the textualist critique has been increased attention to the way in which the ethnographer’s field experience is translated into an ethnographic report and a new degree of awareness about the way in which ethnographies are constructed. The textualists have argued that classic ethnographies incorporate linguistic devices that tend to obscure the uncertain and personal nature of ethnographic statements regarding particular features of social life or cultural belief in the group under study (e.g., Clifford 1983; Geertz 1988).Traditional realistic ethnographic practice assumes that reading is like an encounter with Alice’s looking glass: merely by opening an ethnography and passing one’s eyes over the pages, one is transported through the book-as-mirror into the reality of life in another culture (what van Maanen [1988, 741 calls the “doctrine of immaculate perception”). There is no recognition in the ethnographic medium that the social and psychological reality of some far-off place and time is transformed into the mental representations of a reader through at least one intervening intelligence (the ethnographer’s) and several instances of physical mutation (e.g., into patterns of ink on paper). In fact, traditional ethnographers make use of this magical elision in order to achieve unwarranted credibility, to convince us that “had we been there we should have seen what they saw, felt what they felt, concluded what they concluded” (Geertz 1988, 16). Because it is difficult to know whether ethnographic statements are based on anything more than personal impressions, many ethnographies are convincing only to the degree that the ethnographer has mastered 4

A CRISIS IN CONFIDENCE

rhetoric (as shown by the fact that the most respected ethnographers tend to be the best writers). The textualists argue that once readers are aware of the use of such linguistic devices, ethnographic authority can no longer be achieved by such methods: To summarize the problem of anthropological knowledge[:]. . . sociocultural reality presents itself to the anthropologist in fragmented bits and pieces. The outcome of fieldwork is very much dependent on the cooperation of the participants, on many uncontrollable practical factors, and on the personal qualities of the anthropologist, whose own sociocultural framework substantially screens the knowledge that he produces. This all implies that the knowledge produced in the field is necessarily incomplete, distorted, tentative, speculative, and thus essentially contestable. When put down in writing, this knowledge cannot be separated from the way it is presented in the text. In light of the absence of “hard” criteria, a lack of independent information, and a body of generally accepted anthropological knowledge, this raises the question of to what extent plausibility equals rhetorical and stylistic persuasion. (Bakker 1992,40)

Textualists have argued that the ethnography, as a written document, constitutes a form of knowledge that is unknowably transmuted by the cross-cultural translation of source materials, as well as by power relations between the anthropologist and his or her informants. The postmodernist revolution in ethnographic writing has made clear the need to recognize that there is no single, hegemonic truth that an ethnographer can adopt. Since situations are unique, situated knowledge must be variable knowledge, hence the great concern in the newer generations of ethnographers with explicitly representing ethnographic variation. The problem is that this variation is presented raw, often in unfiltered form. Since the way in which each piece of information presented in the ethnographic report cannot be identified, there is also no way for a reader to judge what biases might have been incorporated into that bit of ethnographic reality.This tactic cannot, therefore, reinvigorate the ethnographic project in the long run, because there are still few constraints to keep ethnographies from becoming literary fantasies. Many textualists therefore claim that the classic interpretation of ethnographic descriptions as objective representations of other ways of life must be abandoned altogether (e.g., Clifford 1988). How, then, can new 5

CHAPTER 1

generations of ethnographers report on their studies of other cultures?The old style of depicting ethnographic reality in timeless, objective, impersonal terms--all typological ambitions-is effectively dead. And this critique has sent traditionalists into a hasty and disorganized retreat from which they have yet to recover. But the answer to what can legitimately follow this tradition is not yet determined. Textualists, recognizing the deceptive nature of this rhetorical gambit, have argued that ethnographers should create a conscious reading experience, explicitly interposing the book-as-object between the reader and the reality being depicted. Through an act of will and imagination, the reader is supposed to construct an individualized experience of the ethnographic reality from the imperfect building blocks of written statements, Textualist ethnographers thus call for new methods of investigation and presentation-such as multi-vocal texts and other forms of representational experimentation-as a way of reestablishing the legitimacy of their reports. This is, in essence, a cry for increased attention to heterogeneity, and the explicit acknowledgement that all knowledge is constructed locally by participants in a culture. Ethnographic information cannot be presented “as it is” without contextualization,the textualists say. They seek instead to let the reader do the work of constructing a representation, merely presenting them with a welter of unanalyzed and potentially inconsistent information, often in a raw form such as field notes. More traditional practitioners have been left in some disarray by the crisis in ethnographic authority, but have claimed the textualist revolution’s apparent victory will be a Pyrrhic one if the notions of truth and value are lost in the translation to new forms of ethnographic representation. Other factors also force a reconceptualization of the nature of ethnography. Circumstances have changed since the classic period of ethnographic writing (i.e., the first half of the twentieth century). Ethnography is no longer conducted by white men visiting strange, isolated places. First, no place in the world, thanks to globalization,is isolated any longer. Instead, all places are connected. Communities penetrate each other in multifarious ways, so there is no way to set some group apart as a definable population for study. For some, it is this very connectability-the linkages between places-that needs to become the proper focus of ethnographic research. Second, ethnographers are no longer restricted to the ranks of white, First World males. Third Worlders are being trained at First World universities, and since ethnographic research can now be con6

A CRISIS IN CONFIDENCE

ducted anywhere, they can choose the First World-which is exotic to them, after all-as their locale for research. Those who were once the observed are now the observers: the former subjects of ethnography are making us their subjects! This change in circumstance means that ethnographic practice must change: the standard relationship between powerful observer and powerless observed no longer holds, and must be rethought (Marcus 1998). Certainly, at minimum, it is clear that ethnographers can no longer obscure whether the words on the page are a literal, unedited transcription of an informant’s speech, a “free indirect” translation of what an informant said (Sperber 1985), or the ethnographer’s own opinion. The textualist critics have suggested that ethnographic materials should therefore be presented in the form in which they were elicited. For example, George Marcus and Michael Fischer (1986) and Renato Rosaldo (1989) advocate an experimental ethnography that juxtaposes in the document autobiographical recounting of fieldwork experiences, multiple narrative voices, transcriptions of historical texts, and so on (eg., Okely and Callaway 1992). Since the construction of such documents reflects the inherently fragmentary nature of the object it is to represent-some facet of life in a cultural group-this pastiche of primary materials should jar the reader into an unconscious acknowledgment of the variable conditions under which such materials have been elicited (e.g., Marcus 1986, 168). The ethnographer, having published all of this material in relatively undigested form, leaves readers to come to their own conclusions, preferably through a sympathetic emotional response inspired by the experience of reading the ethnography (e.g., Tyler 1986). For some, however, this is not enough. James Clifford (1986, 19), for example, argues that any single viewpoint on culture is inherently contradictory, “thoroughly historicist and self-reflexive.”The following passage summarizes the types of contextualizationhe views as necessary for a fully cognizant ethnography: Ethnographic writing is determined in at least 6 ways: (1)contextually (it draws from and creates meaningful social milieus); (2) rhetorically (it uses and is used by expressive conventions); (3) institutionally (one writes within, and against, specific traditions, disciplines, audiences); (4) generically (an ethnography is usually distinguishable from a novel or a travel account); ( 5 ) politically (the authority to represent cultural 7

CHAPTER 1

realities is unequally shared and at times contested);(6) historically (all the above conventions and constraints are changing).These determinations govern the inscription of coherent ethnographic fictions. (Clifford 1986,6)

Even if complete contextualization were possible, such a text would have to be of infinite size, due to the infinite regress of contexts-withincontexts. After all, “meaning is context-bound, but context is boundless” (Fardon 1990,18; quoting Culler 1983,123). So Clifford (1989,562) suggests that any hope for a claim to authority that is not “tactical, politically and historically contingent” is itself symptomatic of the “nostalgia for a single valid method in ethnography, the one best way to guarantee the truth of cultural representations . . . In the historical context of postcolonialism (a loose but necessary periodization) and within the specified domain of a troubled Western social science.” Clifford (1988,56-63) argues that he cannot bring himself to advocate any particular stance (he is “systematically ambivalent”) because he is enmeshed in a “complex location” that is only “partially understandable,” even to himself. Compare the ethnographic situation to that of a fiction writer. In the literary case, the author, sitting at a desk, attempts to create (often out of whole cloth), at a physiological level, the emotional response of a real (behavioral) experience that neither the reader nor the author has ever had. In the case of ethnography, however, we have a strange reversal from the case of fiction: the author, who begins his job of description with an intensely personal experience, uses objective or scientific literary devices, appealing to the intellect of the reader, by abstracting away from that experience as a way of establishing authority. Yet the ethnographer, like the fiction-writer, must establish in the reader’s mind the same “I was there” feeling in order to be convincing. Clifford, a leading spokesperson for the school of anthropology as texual interpretation, has made much of the paradox of establishing authority by evoking an “I was there” impression on the reader through means that require the author to suppress the often highly charged, personal nature of the experience. The standard objective tone of ethnography denies the nature of the source of the information, seeking to make it appear from a godlike omniscience, a perspective-less authority established through typologization and objectification of the text (see Geertz’s 1988 analysis of Evans-Pritchard’s literary devices; Pratt 1986). 8

A CRISIS IN CONFIDENCE

One reaction to this dilemma has been to go to the opposite extreme. If it is impossible to be objective, then why not make the intrinsically subjective nature of social research the source of its value, and celebrate it? Ruth Behar (1997), for example, suggests that ethnographers simply give in to the subjective nature of fieldwork, even relish and embellish it, by responding emotionally to their surroundings. As the title of her book suggests, ethnography should aim to induce a sympathetic emotional response in the reader, even “break the reader’s heart.” Anything less is not worthwhile. Another advocate of placing the self at the center of ethnographic writing, of making the ethnography the ethnographer’s own story, is Reed-Danahay (1997). As Marcus (1998) notes, it is common for feminists to advocate the ethnographer be “positioned”within some context in the ethnographic picture (e.g., standpoint theory). Others would go even further, arguing that First World writing about impoverished societies should be a political act attempting to subvert the status quo (Scheper-Hughes 1995). The problem with these solutions is that ethnography then becomes nothing more than a personal testimony to lived experience; it lacks a larger interpretive frame. Behar in effect advocates that ethnography become autobiography; she is trying to find out what makes herself tick by exploring her emotional reactions to everyday experiences, by trying on new identities (for Behar, the “prodigal daughter” returned to Cuba).2 The ethnography-as-document serves only as a testimony to ethnographers’abilities to attach themselves to the observed. The only way to be honest, for readers to appreciate where you are coming from, is to own up to your biases as explicitly as possible by becoming aware of them yourself, and writing them down. Indeed, some would abandon the quest for authority as the golden standard by which ethnographies are to be judged-at least as authority is traditionally conceived. This standard goal is often replaced with that of ethnography-as-experience. Some strains of contemporary ethnography are even moving toward becoming forms of art. Norman Denzin (1997)’ for example, argues that to accommodate the new, globalized setting in which ethnography finds itself, ethnographers must explore new ways of importing experience into their texts-such as by engaging in autobiographical narratives and poetic or dramatic fictional ethnographies. Ethnographic representations seeking to be artfd are meant to tap into new kinds of responses from the audience through an explicitly aesthetic experience.The reader is channeled data through physical, emotional, and 9

CHAPTER 1

visceral means that enhance the value of the ethnography, if not necessarily its authority (e.g., Brady 2000). The primary point the textualists make is that the ethnographer is confronted with a dilemma: to personally embody the essence of another culture while maintaining his or her own identity within the same skin. An ethnographer’s professional objective is to produce a written document that portrays the mentality of an alien culture, while remaining within traditions of Western scholarship and using the methods and categories of academic anthropology. In response to this dilemma of having to be both “Here” and “There,” textualists have attempted to implement a number of fures. But Clifford Geertz characterizes the variety of solutions suggested to this conundrum as no more than a number of different pretensions: There are a number of these pretensions, but they all come down in one way or another to an attempt to get around the un-get-roundable fact that all ethnographical descriptions are homemade, that they are the describer’s descriptions, not those of the described. There is ethnographic ventriloquism: the claim to speak not just about another form of life but to speak from within it; to represent a depiction of how things look from “an Ethiopian (woman poet’s) point of view” as itself an Ethiopian (woman poet’s) depiction of how they look from such a view. There is text positivism: the notion that, if only Emawayish can be got to dictate or write down her poems as carefully as possible and they are translated faithfully as possible, then the ethnographer’s role dissolves into that of an honest broker passing on the substance of things with only the most trivial of transaction costs. There is dispersed authorship: the hope that ethnographic discourse can somehow be made “heteroglossial,”so that Emawayish can speak within it, alongside the anthropologist in some direct, equal and independent way; aThere presence in a Here text. There is confessionalism: the taking of the ethnographer’s experience rather than its object as the primary subject matter for analytical attention, portraying Emawayish in terms of the effect she has on those who encounter her; a There shadow of a Here reality. And there is, most popularly of all, the simple assumption that although Emawayish and her poems are, of course, inevitably seen through an authordarkened glass, the darkening can be minimized by authorial self-inspection for “bias” or “subjectivity,”and she and they can then be seen face to face. (1988,143-45). 10

A CRISIS IN CONFIDENCE

The substance of Geertz’s condemnation of both traditional ethnographical and interpretive anthropological revisionist attempts to establish ethnographic authority seems conclusive. Cultural anthropology rests on a fragile epistemological foundation: the assumption that ethnographic reports are culture-free methods for the translation of personal experience and informant-derived information into a scientifically sound representation capable of comparison with similar reports on other cultures. This assumption entails the beliefs that informants tell the truth and that anthropologists have the ability to objectively infer the relevant psychological and sociological structures from this information. Such an assumption cannot be assured. An objective ethnography-one devoid of context, or alternatively, perfectly contextualized-is obviously impossible. But on the other hand, a travelogue, a “what I did over the summer” narrative recounting personal experiences, is worthless as science. The postmodernist revolution in ethnographic writing has made clear the need to recognize that there is no single, hegemonic truth which an ethnographer can lay hold of. In effect, the textualists contend that because of the intrinsically interpersonal nature of all social studies, a scientific approach to ethnography is impossible (Clifford and Marcus 1986). Is this dire conclusion really necessary?

Attempts at a Solution

I acknowledge that Geertz’s criticism cited above is an accurate description of the current state of theoretical debate in cultural anthropology, insofar as he includes in his critique all the strategies adopted in the past century of anthropological writing. However, Geertz’s arguments do not quite hit the larger mark he intends, nor entirely vanquish the possibility of a scientific anthropology of cultural beliefs, as he (and most of the participants in the debate over ethnographic authority) suggests. Geertz’s criticism is fully applicable to traditional ethnographic practice, but the current debate is debilitated by several assumptions concerning the possiThat is, Geertz efble nature of anthropological investigations of cu1tu1-e.~ fectively disables traditional methods, but not all possible methods of ethnographic representation, because he implicitly takes several things as given, which are not in fact given. 11

CHAPTER 1

I argue that three assumptions characterizing the current debate on ethnography limit the possibilities entertained by the participants in the debate. The first is that science is objective, so that the inability of ethnography to be objective signifies that one must give up a science of cultural representations and be satisfied with armchair discursiveness about alien cultures. The entire postmodernist revisionist debate has focused strictly on methods. Meanwhile, the goal of ethnography has gone unexamined, and hence unchanged. In effect, an unstated assumption of those conducting this debate (both interpretive anthropological critics as well as traditional cultural anthropologists) has been that the proper job of cultural anthropology is to authoritatively describe a culture as a logically consistent, invariant (not just homogeneous) whole: the reification of cultures as solid objects. The difficulty of achieving this aim is not a function of the describer being outside the culture so described. The problem is that cultures are neither necessarily logically consistent nor do they lack variability. No change in method will cure the main problem: the ethnographic objective of using a single template to describe cultural groups-even one aspect (encyclopedic ethnographies are no longer attempted). Other textualists have attempted to limit the potential damage to the reputation of their discipline by changing the goalposts, concluding that both Freeman and Mead (for example) came home with a different, but equally valid truth. Thus, there is no single Samoan truth, that is, there is no possible way to determine the correct interpretation of Samoan reality (e.g., Scheper-Hughes 1984). Each ethnographer is simply like one of the blind men approaching an elephant: each describes a part of this highly variegated animal, but their reports do not add up to a recognizable whole. But what if, in describing a culture, the failure results from the fact thatas Gertrude Stein is reported to have put it-“there is no ‘There’ there.” Perhaps, then, Freeman and Mead are merely describing different parts of a single, but internally structured or heterogeneous culture. In spite of this recognition of variability, current ethnographic practice is not that different from traditional modes, except that fieldwork must typically be continued longer before being considered definitive, and ethnographers make sure to actually live with the people whose lifeways they describe. Nevertheless, single anthropologists still go to the field and after querying small numbers of informants, tell their story about how some people are supposed to live-at least with respect to some aspect of 12

A CRISIS IN CONFIDENCE

their lives, such as marriage practices or religious practices. Certainly, major problems lie in such methods. The effects of this peculiar combination of participation in foreign lifestyles on those having specific personality traits will result in highly idiosyncratic impressions, which are then translated to paper for all to read as definitive. But this is not replicable social science. The second assumption that has limited debate is that the set of possible field data collection methods has been restricted to the traditional ones of participant observation and reliance on key informants. The possibility that nontraditional methods of data collection could produce the kinds of data from cultural anthropological fieldwork that might be scientific has not been considered. A concern with replicability will inevitably involve the abandonment of traditional methods of participant observation and reliance on key informants. Advances in establishing scientific access to other mentalities will not occur from sympathetic coexistence or unstructured discourse with other humans, whatever their cultural background. Traditional methods of cultural anthropologists are not substantially different from those of professional adventurers, except that anthropologists tend to stay longer in one place and learn the local language. Thus, we get difficulties in distinguishing between a travelogue and an ethnography (except the latter is addressed to a different kind of audience and therefore adopts a more authoritative tone). Continued attempts to get inside other people’s heads, or to establish intersubjectivity, using emotional and intellectual sympathy (or Entehen) derived from either passive participant observation or active involvement in social behaviors, are doomed to failure. Ward Goodenough‘s (1965) widely espoused ideal for the cultural anthropologist-to behave in a way considered perfectly natural by a native member of the culture under study-must be given up (it is, after all, Geertz’s first pretension). The third assumption is that the theoretical entities studied by traditional cultural anthropology are compatible with scientific methods (i.e., method, unit of analysis, goal, and meta-analytical assumptions). As Dan Sperber (1985) has pointed out, traditional subjects of cultural investigation (e.g., totemism, kinship, and ritual) are highly abstract theoretical constructs that are unlikely to bear a simple relationship to the kinds of mental representations people have in their heads. They are scholastic distinctions that don’t necessarily reflect the way people categorize the world based on their personal experience. 13

CHAPTER 1

Anthropology receives from ethnography inappropriate concepts and irrelevant issues. An important part of its energy is spent on trying to answer such questions as: What is totemism? Sacred kingship?What is the meaning of sacrifice? . . . All these questions are ill posed. They are framed in interpretive terms. There is no apriori reason to assume that these terms correspond to homogeneous and distinct classes of phenomena, i.e., to potential objects of scientific inquiry. (Sperber 1985, 29-30)

If we examine the larger “ethnography space” which has been illuminated by this broadening of the assumptions underlying recent debates, we will find that an empirical protocol for ethnographic research that is scientific can be found within this new space. The objective of this book will be to establish the characteristics and methods of this empirical cultural anthropology, defined as “the systematic description and classification of objects, events, and processes, and the explanation of those events and processes by theories that employ l a d l regularities, all of the descriptive and explanatory statements employed being testable against publicly observable data” (O’Meara 1989,354). I conclude this introductory chapter with some notes about assumptions of such a reorientation of cultural anthropology, as well as changes in both cultural anthropology and cultural anthropologists that will be required to bring such a program into practice. This is not going to be easy; it won’t come without a cost in terms of training and practice of ethnographers.

New Authority Figures The aim of this book is to save the possibility of a scientific cultural anthropology from ignominy. Any attempt to reinvigorate ethnography must first rescue it from these current difficulties by effectively reestablishing the authority of a cultural representation. The first step, however, must be to recognize that the textualist critique has substance; ethnographers cannot simply proceed as before. Any response to this critique that purports to result in a scientific approach to ethnographic description must therefore meet both the textualist challenge and the requirements of scientific analysis. First, we must acknowledge the brunt of the textualist critique: because of the interpersonal nature of data collection in human studies, 14

A CRISIS IN CONFIDENCE

ethnographic research must be reflexive. This somewhat slippery term rose to prominence in the wake of the Freeman/Mead debate over ethnographic practice, when it was recognized that the reader’s ability to interpret the quality of ethnographic statements must be increased by providing clues to the origin and nature of ethnographic statements in the ethnographic document itself. “Reflexive knowledge, then, contains not only messages, but also information as to how it came into being, the process by which it was obtained” (Meyerhoff and Ruby 1982,2). However, because my goal is to develop a scientific approach to ethnographic research, I use the term somewhat differently from the textualists, who have tended to believe that recognition of the variable quality of primary data requires an abandonment of analysis in favor of the honest (i.e., unmanipulated) presentation of basic materials. Reflexivity for them consists in the presentation of raw ethnographic material, which involves the reader more closely in the ethnographic experience. However, a reader lacking firsthand experience with the group under study cannot supply the ethnographer’s insight. Therefore, the ethnographer cannot simply leave the reader to contribute the analysis, but has the obligation to interpret ethnographic material^.^ As a consequence, I emphasize reflexivity as an aspect of analysis rather than of presentation in the ethnographic d o c ~ m e n t As . ~ Charlotte Davies (1999, 21) says, if we remain realists (accepting the existence of a world independent of our perceptions of it), we can still “study something as an object so long as we are sensitive to and take account of our own implication in and effects on that object.” In particular, I argue that analysis can be undertaken in ethnography as long as analytic methods explicitly take into account the means by which the data were collected. Thus, analysis must allow that the data elicitation process itself, as well as circumstantial aspects of the data collection situation, can influence what informants say or do. The fact that social action is intersubjective requires a consideration of many interpersonal influences as well as other aspects of the situation at the time of observation. I therefore maintain that it is necessary to specify fully the context of the data collection event in the analysis through methodological situationalism (KnorrCetina 1981). This second criterion for an acceptable ethnographic approach, a situationalist perspective, acknowledges that human social behavior is contextualized both in time and space, and with respect to the simultaneous action of other individuals. Methodological situationalism can be distinguished from methodological individualism, which holds that no 15

CHAPTER 1

phenomena emerge in the context of a social interaction that cannot be reduced to the characteristics of the individuals who partake in that interaction (Rhoads 1991,118). It is also distinct from methodological collectivism, which argues that reference to individual mental constructs or behaviors is superfluous because humans are merely the passive instruments of socialstructural processes (Ritzer 1992,79). Where methodological individualism allows only for human agency and methodological collectivism only for social institutional determination, methodological situationalism allows for both. It “locates knowledge neither in someone’s head, as solipsists do, nor in an external and observer-independent universe . . . as objectivists or natural scientists . . . insist on, nor in text, as many hermeneuticians and discourse analysts argue, but in an essentially circular social practice involving perceiving, thinking and acting (including languaging) beings” (Krippendorff 1991,115-16). My requirement that the approach be scientific places additional constraints on ethnographic practice. A scientific study must meet two methodological criteria: the researcher must use (and report) an explicit method for making inferences from primary data and a means of assessing the nature and quality of data prior to the inference step.6 Otherwise, inferences are made from a flimsy foundation that may affect the conclusions reached. Unless such steps are taken in the presentation of ethnographic materials, it becomes extremely difficult for readers to compare alternative claims about cultural facts, a problem made obvious by the scandals of recent times. The key is to use what I call “reflexive analysis.” Reflexive action in general occurs when the object of an activity is the same as its subject. The form of reflexivity we require here is that the analytical method needs to take account of the context of data elicitation. In particular, data collection procedures must be introduced that allow spurious influences on primary data to be identified by systematicallyvarying these influences. For example, in the typical case of interviewing, one can change interviewers and the circumstances of the interview between interviews. The biases introduced by these extraneous factors on informant responses are then detected and eliminated through the use of a formal method of analysis, such as multivariate statistics, based on the entire sample of interviews. With these purified data, the project of describing cultural variation can be reliably undertaken. The ethnographer can, for example, present to readers the structure of this variation-such as the frequencies of belief within 16

A CRISIS IN CONFIDENCE

classes of individuals, as characterized by their social roles-suitably framed by standard interpretive glosses. Reflexive analysis is thus capable of controlling for its own idiosyncrasies, of taking into account spurious influences on primary ethnographic data. By operating on itself through the instrument of reflexive analysis, the researcher’s representation of ethnographic reality is transformed to correct for individual idiosyncrasies, thus affording a truer picture of the nature of cultural variation within and between groups. In effect, ethnographic research becomes an example of self-correcting action, so that the outcome of a researcher’s activity-an ethnographic representation-becomes reliable. This in turn requires changes to standard ethnographic practice in terms of data collection, analysis, and presentation. But the use of this method rejuvenates the possibility of legitimately representing cultural groups unfamiliar to readers.’ What if the ethnographer’s concern is with beliefs and values? It is relatively innocuous to classify people according to their occupation, age, or even (for the most part) sex. These variables can be readily observed and tabulated for presentation in a summary report. What is less morally acceptable is to apply epithets regarding beliefs and values to entire groups as a whole, as in statements like “Americans believe in one omnipotent God.” Well-identified domains like religious belief can exhibit rampant variation within a population-even a traditional ethnographic one. So individuals can be misidentified by such a global strategy. Beliefs and values are, of course, the least visible aspects of culture, although they manifest themselves in complex ways through speech and other kinds of behavior. They are also perhaps the most exotic aspects of culture, and therefore constitute a traditional focus of attention in ethnographic work. But reflexive analysis is a method through which one can successfully dredge up beliefs and values from the depths of the mind, and make them presentable to the world at large in just the same way as other kinds of ethnographic information. In the course of this book, I provide an example of just such an ethnography. I use reflexive analysis on my own ethnographic data concerning food avoidances in a multiethnic population, including pygmy foragers and their horticulturalkt Bantu associates living in the Democratic Republic of the Congo. Thus, the book represents an example of how to conduct such research-it is, in effect, itself an example of reflexive ethnographic science. Since reports of belief are particularly likely to suffer from bias and misrepresentation, the book focuses on this type of ethnographic 17

CHAPTER 1

data. The need for reflexivity is conclusively established not only by the empirical investigation of the focal ethnographic database, but also by a critical survey of existing ethnographic methods. The book therefore combines elements of a report on my own empirical research and a review of the research literature on ethnographic representation and authority. Most importantly, it generates a new theoretical point of view on ethnographic research capable of restoring scientific ambitions to the field. The book is intended to offer a solution to perhaps the fundamental debate in contemporary ethnography: the source of ethnographic authority. Simultaneously,it advocates a new way of practicing ethnography that can move it toward a scientific endeavor. As a contribution to a scientific cultural anthropology, it constitutes a foil to those in cultural studies and related fields who deride the possibility of verifiable ethnographic representations. Instead, it points the way toward a unique combination of traditional and postmodern objectives, through the reflexive achievement of authority. This book suggests a way we can have our ethnographic cake and eat it too, to be both scientific in approach yet responsive to the problems associated with the inevitably social nature of such research. Obviously, this is but a brief outline of an ambitious program for reinvigorating ethnographic practice. But since the proposed method for achieving a reflexive science requires considerable changes to standard ways of doing things, it must of necessity be supported by a strong case for the inadequacy of existing methods. In the next chapter, the need to go through the arduous task of finding a new approach to ethnographic research is questioned. Perhaps an existing approach already satisfies our needs, although it may not be widely recognized for doing so. Unfortunately, a survey of such approaches suggests this is not the case-we really do need to develop a novel approach to the reflexivity problem. However, we also need to be sure that reflexivity is at the root of our current troubles. Therefore, chapter 4 shows that the problems arising from a lack of attention to reflexivity are indeed significant. This demonstration is empirical in nature, based on a somewhat unusual kind of ethnographic database. This analysis provides an adequate foundation for my argument that a rigorous reflexivity is truly required. One way of realizing reliable representations of cultural lifeways is advocated by those who favor idealizing away from the messy details of cultural variation to infer the consensual beliefs and values of a group. However, I argue in chapter 4 that this route to retrieving authority is se18

A CRISIS I N CONFIDENCE

verely hampered by both empirical and conceptual problems inherent in any such approach. These metaphysical constructs have the same problems of reliability-even though reached through objective methods-as the prior ethnographic practice of basing reports strictly on the subjective impressions of the ethnographer. Again, this argument is backed up by empirical validation. In subsequent chapters, I advocate another approach toward reliable ethnographic representations,but one that is realistic in its attempt to measure cultural “variation on the ground.” This approach remains reflexive, however, because it deals squarely with biases introduced into primary data by the circumstances of its collection. I conclude the book with observations about how ethnographic practice would be different if the methods I advocate were to form the basis for future ethnographic projects, and discuss the philosophical as well as ethical virtues of such a future.

Notes 1. Indeed, Lowell Holmes was a third ethnographer who went to Samoa and found his own truth there, which lies somewhere between those of the two main protagonists. See Holmes 1987; Holmes and Holmes 1992. 2. In contrast to Behar is the experience of Jayati Lal, a South Asian woman of color who was raised in India, trained in the West, and went home to do fieldwork, but found she was a “‘native’ now returning to a foreign country” (Lal 1996:192; quoted in Davies 1999:181). 3. One assumption I will not debate, because it does not solve the fundamental problem of interpretation, is that suggested by Crane (1991). H e argues that the debate has artificially limited itself to the discussion of written documents, whereas it is now possible to compose a computerized multimedia ethnography: video, still pictures, and sound bites, combined with written information, all presented in a package. Such a presentation would be easy for users to augment in various ways or to reform, giving the user or reader greater power to manipulate the message he or she receives. The hypertextual aspect increases the ease with which one can have a personalized experience or path through the information, plus avoidance of irrelevant materials instead of the linear path typically traveled through a book. Although Crane does not mention this, mental representations are likely to have a variety of sensory bases, so that an information source that mimics the form of some aspect of a mental representation should facilitate the speed and accuracy with which information is transferred to the learner. But such a solution would still fall under the category of “electronic heteroglossia” in Geertz’s thus-augmented classification scheme, since a Western researcher has determined the structure of the representation, regardless of the variety of voices present within. The medium of the message does not affect its epistemological status with respect to Geertz’s “un-get-roundable” fact. 4. Textualists often include firsthand narratives as some portion of their ethnographic account, placed side-by-side with other textual material, in order not to privilege the ethnographer’s viewpoint. I argue, on the contrary, that the ethnographer’s interpretations 19

CHAPTER 1 are privileged, due to firsthand acquaintance with the subject matter of the report. If a reflexive approach is used, readers are nevertheless able to determine independently on what basis interpretations are made and hence to what degree they will concede those interpretations. As a result, the ethnographer’s point of view does not preclude or preempt those of readers. 5 . Another aspect of reflexivity-heightened consciousness about personal experience through self-reflection prior to writing the ethnographic account-is not excluded either by my interpretation nor that of the textualists. What is excluded in my account is the belief that reflexivity begins and ends with an appreciation of the subjective nature of social research (e.g., Gergen 1989). 6 . “Primary data” will be taken as any information directly elicited from informants, whether responses to questions (which may include reports of behavior at other times and places) or direct observation of informants’ behavior. I mean to exclude from this definition any first-order inference that might be contained in field notes, as well as the notion that primary data are themselves “text.” 7. Alex Stewart (1998) has recently argued that ethnographic authority can be retrieved by greater attention to the veracity of ethnographic reports. Veracity is simply the verisimilitude with which cultural lifeways are depicted, or the truth-value that observations in ethnographic descriptions can claim. Stewart argues that prolonged fieldwork, good relationships with informants, attentiveness to context, the use of multiple types of data collection procedures, and an active search for disconfirming observations are the best guarantors of veracity. However, this is not enough of a change in program because it is much less formally reflexive. We cannot know the value of an observation unless it is explicitly reflexively analyzed.

20

2

INVESTlGATlNG EXIST1N G ETHNOGRAPHIC METHODS

an any method already being used by social scientists provide the foundation for a revival of ethnographic authority? In particular, is there something approaching reflexivity already in practice? This chapter attempts to answer these questions by reviewing a variety of popular approaches to ethnographic description and analyzing how they deal with basic data. (Purely solipsistic and idiosyncratic methods are excluded from this review.) To make the comparisons explicit, and thus facilitate the search for the best available method, it focuses on what is crucial to reflexivity: an explicit consideration of the likely impact of data collection methods on the nature of basic data. This perspective will show that all of these approaches infer some normative value for the group from variable, individual-level observations without an explicit methodology for doing so. Nevertheless, purely abstract arguments are often not convincing. Therefore, in the context of each model there is not only an outline of the method itself, but also a description of one or more empirical studies on humoral beliefs that uses that approach. Humoral beliefs are perhaps the best-studied system in traditional anthropological populations for interand intra-informant variation. They derive from the Platonic system of “qualities” that describe the essences, or different intrinsic natures, of all things (for a history, see Foster 1987).Hippocrates, in his medical writings, argued these qualities could be reduced to four, along two different dimensions: hodcold and dry/wet. However, as this belief system is currently manifested in ethnomedical practice among traditional populations on several continents, this second dimension has been largely lost (thus, the humors are commonly reduced to the “hot” and “cold valences). An important extension of these beliefs has been the description of the natures

C

21

CHAPTER 2

of different foods (Mathews 1983, 827). As an aspect of ethnomedical treatment, imbalance in the body’s own valence is managed through the consumption of foods with an opposite valence: for example,“hot”foods when the body is “cold.”A particular virtue of this belief system is that it is closely related to the major study on food taboos presented later, and so provides a nice counterpoint between my own case study and the existing ethnographic literature on food-related beliefs around the world, The analyses in this chapter suggest that traditional ethnographic methods are deficient, each in a specific way. These deficiencies all derive from lack of attention to some aspect of the ethnographic data collection situation. Considering that the typical context of data collection is an interview, it is apparent that these methods ignore either the role the interviewer plays in determining responses, or the situational factors due to the interviewing context itself (such as the presence of others in the vicinity), or purely methodological influences such as competency in the interview language. These deficiencies can significantly bias the kinds of information acquired and hence any subsequent interpretation of those data. These methodological deficiencies have gone largely unrecognized by ethnographic practitioners, but have left behind a vague sense that something is missing. These deficiencies, in effect, probably led to the recent crisis in confidence among anthropologists. The various approaches, grouped into three classes, are now discussed in turn.

The Pure Informant Model The first model attributes all observed variation in informant behaviors or responses to real differences in belief or behavior that is accurately recorded. Since these models attribute observed variability directly to informants themselves, I will call this the “Pure Informant” model. There is but a single approach falling into this class because only one factor is explicitly considered.This is the simplest possible approach conceptually because it takes informant reports or behavior at face value. When compared with the models that follow, this approach neglects several factors that other approaches identify as influencing ethnographic observations. In particular, there is no consideration of the context in which the observation was made. Instead, researchers adopting this approach argue that 22

INVESTIGATING EXISTING ETHNOGRAPHIC METHODS

within-cultural variability is systematically related to individual characteristics. Different cultural values reflect different degrees or types of exposure to cultural influences. There is no assumption that culture is uniform within groups, and no attempt to infer group-level norms from variable informant reports. Several ethnographic studies of humoral systems argue that the observed variability in belief about the valence of particular foods is due to cultural variation among informants. For example, Michael Logan and Warren Morrill(1979), studying humoral beliefs in Guatemalan villagers, argue that the inter-informant variation they recorded is a measure of subcultural variation in belief (i.e., it is not measurement error or situational variation). These subcultures arise due to varying degrees of acculturation, considered as a purely cultural phenomenon. Similarly, Christine Wilson (1970, 286) attributes variability in humoral assignment among the Malays she studied to differing family traditions.

Pure Context Models The next set of models attributes all observed variability to random effects. They argue that there are no biases or errors introduced by the elicitation process itself. For this reason, I call them “Pure Context” models. These schools are characteristic of traditional ethnography (the great majority of those written before about 1970), where the role of the anthropologist-as-observer is masked by the adoption of an authoritative tone (Clifford 1983) and the use of a “free indirect” writing style, which obscures the anthropologist’sabstraction from informant reports to make assertions applicable to the group as a whole (Sperber 1985). I distinguish two varieties of Pure Context model, based on their assumptions concerning the level of organization at which the variability is assumed to be produced. The SituationalVariation School The first approach in this group is what I call the “Situational Variation” model. This school of thought suggests that all observed within-cultural variation in elicited beliefs represents real differences in informant attitudes and knowledge, which can be interpreted as deviations from the culturally uniform or normative value, due to random situational factors. In 23

CHAPTER 2

this approach, there is no systematic variability in observations due to characteristics of the individual (e.g., by sex or age), except insofar as different individuals have had different personal experiences. This approach suggests that without the randomizing influences of individual lifehistorical experiences, everyone in the cultural group would exhibit the same cultural values. Variation is not methodological since the ethnographer’s presence andor own perceptive biases have no effect on what is observed. Further, there is no measurement error (e.g., respondents are implicitly assumed to report their trait values honestly and correctly). The present approach is distinguished from the Pure Informant school by the assumption that individual-level variation is due to personal experience rather than cultural forces. It is cultural norms that are of interest to researchers in this school. How these norms are inferred from the various observations on different informants is unclear. However, a simple rule is to take the modal value of observations (e.g., the most frequent response) as equal to the normative value and simply discard the other variants. Thus, some researchers studying humoral belief systems have argued that valence assignments simply reflect individual experience with foods. For example, Ethelyn Orso (1970) found that when a Puerto Rican’s personal experience after consuming a food did not agree with the humoral classification he or she had learned for that food, the individual’s reported classification changed to conform to this experience. Brenda Beck (1969) found that South Indians reported they would have to consume a food and experience its effects on their body in order to assign it a humoral classification. Such studies can be considered cases of the Situational Variation approach, since only situational variation-how an individual reacted to some previous consumption event-is considered relevant to that individual’s current classification of food valences. Nevertheless, prior to these individuating experiences, people are implicitly assumed to have been socialized into the group-dominant beliefs.

The Informant Error School Devotees of another Pure Context school agree that all variation observed in cultural data is caused by contextual factors, but argue that this variation has no intrinsic meaning at the individual level. This group includes those ethnographic studies that acknowledge within-cultural variability 24

INVESTIGATING EXISTING ETHNOGRAPHIC METHODS

exists, but that nevertheless suggest variability in behavior and belief in a cultural group is insignificant. I will call this the “Informant Error” school. This approach includes the great majority of traditional ethnographies, which describe the cultural group by normative values for each cultural domain. The typically implicit goal of this approach is related to that of the previous school, except that the desideratum of the analysis is not a measure of individual values, but an estimate of the real value for the cultural group as a whole. So, unlike the Situational Variation school, this approach does not trust that different informant responses represent real differences in belief between them. This is a reasonable position to take, given the tradition of work concerned with how accurately informants report on past events. Russell Bernard et al. (1985, 509), in a summary of this literature, conclude, “[TIhere appears to be systematic distortion in how informants recall just about everything.” Variables that can influence retrospective recall include the means through which the memory is stimulated (for example, a written question versus a visual stimulus),the length of time since the event occured, the informant’s own mental abilities, the sensitivity of the question being asked, a variety of cultural variables (such as a strong tendency to deviate toward normative responses), and so on. (This finding will be strongly supported by the case study described in the next chapter as well.) Individual reports of facts or beliefs can be rife with error and are not to be trusted. But the Informant Error school does not build on this knowledge. “Context” in this case represents unspecified distracting factors believed to have confused the informant at the time of making the response (the means of elicitation of responses are not acknowledged at all). The ethnographer’sjudgment is required to infer the culturally normative belief, although again the method for doing so is not detailed. However, in practice a single informant is typically asked about any particular topic, so there is effectively only one response from which to choose (one of the virtues of the traditional ethnographic practice of using key informants). Holly Mathews (1983), studying humoral beliefs among villagers living in Oaxaca, Mexico, argues that the degree of intra-informant variability she documented with respect to a particular food belief is a function of the number of classificatory dimensions along which that food has a consistent value of either “hot” or “cold.”From intensive interviews, as well as observations of the everyday and ritual contexts in which food was used, 25

CHAPTER 2

Mathews identified several different dimensions along which humoral beliefs for particular foods are independently defined. These dimensions include danger (from consumption of the food), food preparation, and use of the food in treatments for minor illnesses. Responses were much less inconsistent than average (both between informants and by the same informants on separate occasions) for those foods having at least two consistent valuations across these different cognitive dimensions. Classificatory discrepancies were much higher than average among foods with conflicting valuations on two or more dimensions. From this evidence, Mathews argues that there is a shared cultural model for the assignment of valence for foods along each of these dimensions. However, variation in the informants’ interpretation as to the dimension being implied in interview questions caused their valence assignments for foods to be different on different occasions. Thus, variation is considered to be subsumed by a shared cultural system, and individual variability is a reaction to variation in the context of implicit reference. Similar explanations are provided by Carol Molony (1975), who also studied humoral beliefs in Oaxaca, and Michael Brown (1976), who worked in Peru. The implicit model these authors share is consistent with the Informant Error approach: there is a shared cultural model for the assignment of valence to foods. These authors suggest that it was only confusion among informants as to the dimension being implied in interview questions that resulted in variable reports of belief. Unlike the previously mentioned Situational Variability studies, variation in valence reports are not assumed to reflect cognitive differences between individuals in that cultural group arising from varying personal experiences with those foods. Instead, the authors imply that such individualized assignments are due to inefficiency in the techniques used to elicit beliefs.

Culture-Centered Models The third group of models is distinguished by its attention to the inference of normative values to characterize cultural groups. Such studies almost exclusively elicit responses through use of questionnaires or pen-and-paper tests, which do not involve an active interviewer. The questionnaire situation is viewed as being “self-diagnostic”because the informant responds to the inert piece of paper. Such methods therefore typ26

INVESTIGATING EXISTING ETHNOGRAPHIC METHODS

ically assume that measurement of informant values is perfect. This method of data collection is dominant in the psychological sciences, from which this school draws its inspiration. In particular, the models in this group are very similar to the sophisticated statistical models developed in psychological testing. There are two approaches in this group, the Cultural Consensus school and Cross-Cultural Psychological models. Both are group-level approaches, and thus distinct from the Pure Informant approach in being interested in cultural rather than individual-level phenomena. Cultural Consensus theory was developed quite recently by a group of anthropologists (Romney, Weller, and Batchelder 1986) as a formal approach to cognitive anthropology in a cross-cultural situation. Cross-Cultural Psychology has been around rather longer, primarily spurred by the question of ethnic differences in intelligence, but more recently including other aspects of cognition (such as culturally important skills). Both schools attempt to determine the cognitive structure representative of some cultural domain in a particular society, based on individual performances on standardized psychological tests (e.g., categorization tasks). The Cultural Consensus School

The objective of this school is to identify group-characteristic responses from small samples of informants when the correct cultural beliefs are not known a priori (the typical situation for an ethnographer studying previously undescribed groups).’ Cultural Consensus models are thus designed to deal with the opposite situation to that addressed by psychometric testing models. Psychometric test theory estimates an examinee’s unknown “cognitive ability,” based on his or her responses to test items (questions) of unknown difficulty (as well as other characteristics of the examinee in some formulations).2 Cultural Consensus models, on the other hand, address the problem of latent dependent variables, or the situation where the correct answer key to a test is unknown (Batchelder and Romney 1989). Thus, Cultural Consensus models estimate informant abilities and, based on agreement between informants, determine the correct answers to a set of questions with assumed degrees of difficulty. This approach assumes that the observed correlations in responses between informants measure the confluence of their knowledge of the true answers (i.e., that there is a single correct “answer key”). Inconsistencies 27

CHAPTER 2

between informants are therefore considered to be errors in knowledge about a uniform cultural system, although due to real cognitive variation between examinees. Individual data is nevertheless primarily a means of determining consensus values. Cultural Consensus models can therefore be considered a class of the Culture-Centered model, but distinguished from other testing approaches by the assumption that real values exist only at the group level. As with the psychometric approach, an estimate of the ability of each informant (termed “competence”by Cultural Consensus theorists) is one result of the model. However, competencies are found as a dependent, not independent, variable, flipping the usual testing perspective. The estimated competence of an informant is measured as the degree to which that informant’s set of responses coincides with the consensual responses of informants in the sample. The estimation method (usually maximum likelihood) uses information from the entire set of questions to estimate the informant-specific parameter of ability. (This is different from the approaches above, which are estimated independently for each question.) The second product of a Cultural Consensus analysis is a determination of which answer to a given question is most characteristic of the group of examinees (the “consensual”response). This is achieved using the degree of inter-examinee correlation in responses, weighted by their estimated competencies. This value, found by pooling information over the estimated competencies, is then assumed to be the correct value of the trait for the cultural group. This move distinguishes the Cultural Consensus approach from the psychometric approach by switching its attention to group-level objectives. The supposition of a group-level description of cultural beliefs links the Cultural Consensus approach to the previous set of Pure Context models, which have the same goal, although they are much less explicit in practice about how to derive norms from variable data. James Boster and Susan Weller (1990),using a Cultural Consensus approach, test Mathews’s (1983) contention that variability in humoral beliefs only reflects the context to which informants assume the question concerning a food’s valence refers. They duplicated Mathew’s study design in Tlaxcala, Mexico (nearby to Mathews’s group at Oaxaca), using questions specific to the dimensions suggested by Mathews (e.g., food preparation, health). Their data show that the degree of variation in the assignment of valence values does not depend on whether a particular context (e.g., food preparation) is specified in the question. Since explicit ref28

INVESTIGATING EXISTING ETHNOGRAPHIC METHODS

erence to a particular context does not reduce the observed variability in humoral beliefs (as measured by the degree of consensus among informants), they argue that context cannot explain the differences in valence assignments between informant^.^ Further, the consistency in valence assignments between informants is high, regardless of context. They therefore conclude that Tlaxcalans exhibit a “reasonably coherent hot-cold cultural system”(Boster and Weller 1990,178). Variation is due to degrees of expertise in the cultural system, not to any methodological problems.

Cross-Cultural Psychology

When psychological anthropologists study other cultural groups, their work is often based on the pen-and-paper” tradition of psychological testing, but applied to the cross-cultural field situation (e.g., Berry et al. 1992,231). Performance on a task is related to its difficulty for a particular informant (as in the psychometric model). Problems in measuring cultural knowledge are considered to be the residual from fitting statistical models to the data-contextual variability is considered to be of unknown size and dimension. Again, no explicit attention is paid to the method of data collection (whether questionnaire or interview). As with the Cultural Consensus approach, the model uses all of the information on an informant to assess individual performance, and then abstracts to a group-level description of cognitive structure. This abstraction is typically achieved through statistical reduction of the individuallevel data (e.g., through multidimensional scaling). Individual variability in performance in the elicitation task is considered to be spurious and random (as in the Pure Context school, but unlike the Cultural Consensus approach). Since cross-cultural psychologists actually use statistical analysis, the model above is not merely illustrative in this case, but rather an accurate (if schematic) portrayal of analytic methods used by practitioners of this approach. There are no empirical examples of humoral beliefs using this approach because of its psychological nature.

Problems with the Existing Approaches Although I have considerably simplified the various positions just outlined, one point should be clear: these approaches make very different assumptions 29

CHAPTER 2

concerning the role of the elicitation process, the kind of variability attributable to informants, and the role of situational factors in determining observations. Nevertheless, it is difficult to assign many humoral studies definitively to any particular school of ethnographic description. For example, the Logan and Morrill (1979) study, identified above as belonging to the Pure Informant class, could as easily have been assigned to the Cultural Consensus school approach if the relevant cultural group was simply redefined from an ethnic group (as in Cultural Consensus) to a subculture (as considered by Logan and Morrill). Although Logan and Morrill's analytical methods are not those characteristic of Cultural Consensus, the attribution of the observed variation would then be identical to Cultural Consensus in terms of the criteria used in this chapter: the cultural system would be uniform (i.e., exhibit consensus) within the defined group, due to group-level cultural processes, with cultural values assumed to be measured without error. The difficulty of determining the proper location of a study in the catalog of approaches I have described arises because methodological positions vis-ii-vis various aspects of the data collection situation have not been made explicit by these researchers. Most studies in the ethnographic literature have typically worked without any explicit model for the treatment of informant variability. In addition, there are often statements hinting tangentially at concern with particular effects that are not included in the overall argument of a paper. An example is provided by Mathews (1983), discussed above. While it is not the main thrust of Mathews's argument and she does not specify the interviewer as a source of variability in observed response per se, Mathews does admit (almost parenthetically) that some portion of the variability in different contexts is likely to be methodological. Mathews (1983,840) suggests her informants are doing their best to assign valences to foods with no particular (i.e., culturally defined) value in the relatively unfamiliar context of a conversation with an anthropologist. Part of the reason for within-informant variability in valence assignments, she suggests, has to do with the abstract nature of the interview discourse as context (i.e., there is no person to be treated for humoral imbalance at hand). An exception to the difficulty of placing an anthropological study in a particular category is Foster (1979) who, almost alone in this literature, makes his methodological concerns explicit. Indeed, such concerns form the substance of his article. Foster may admit to a wider variety of factors 30

INVESTIGATING EXISTING ETHNOGRAPHIC METHODS

impinging on interview responses than other researchers because of his inability to elicit replicable responses during a long-term study of humoral beliefs in Tzintzuntzan, Mexico. Foster allows not only that there is intracultural variation in responses (based on answers from individual informants that are consistent with themselves but which contradict the majority response for a particular food), but also that there is significant withininformant variation that is simply ever-present: “answers are situational” (Foster 1979, 183). Most significantly from our perspective, he suggests that part of the problem is the “eliciting techniques of the anthropologist” (Foster 1979,181). His study is thus the only one in this literature explicitly concerned with problems of elicitation (i.e., that exhibits features of reflexivity). As we have seen, even within the literature on humoral systems, the various studies that admit informants vary in belief fall under different approaches, ranging from the Situational Variation school to the Cultural Consensus model. These studies therefore assign the variability they observe to different factors, even though in almost all cases the only data available are the beliefs reported by different informants. This can be seen especially in the difference between Mathews’s Informant Error and Boster and Weller’s Cultural Consensus interpretations of nearly identical datasets from neighboring communities. The assignment of intra- or inter-informant variation to the different possible sources of such variation probably reflects the guesswork and/or biases of the researcher to a greater or lesser degree, because an appropriate model of the data collection situation was not used. My review of this set of studies indicates that greater attention to reflexivity and the forces that produce within-cultural variation in ethnographic data is central to clarifying interpretation in ethnographic research. Methods of analysis that do not recognize the way in which data was collected can lead to confusing results. Thus, the review just undertaken strongly suggests that each of the different approaches reviewed above are inadequate for ethnographic research because they omit consideration of one or more of the factors required of a reflexive approach (i.e., the method of elicitation, the context of the elicitation situation itself, and informant cognitive variation). These factors necessarily exist, whether or not the ethnographer recognizes them. Exclusion of one or more of these factors leads to conceptual and analytical confusion, as exemplified in the studies from the humoral literature using such approaches (discussed above). The inadequacies of the 33

CHAPTER 2

various approaches to the analysis of ethnographic data can be effectively determined by examining how each approach deals with these factors. I now make a more direct assessment of each of the problems with existing ethnographic methods.

Elicitation Effects

All of the approaches discussed above assume that, as long as data analysis is rigorous, any data collection method would produce the same set of results. Of course, variability in the interpretation of the dataset would remain, reflecting the goals and ideas of the researcher, but the process of data elicitation itself is assumed to have no effect on what is collected. However, there is now a wealth of information in the sociology of science arguing against this positivistic understanding of scientific research (e.g., Latour and Woolgar 1979; Collins 1982). The means by which data are elicited itself biases the nature of the data acquired. Thus, the most conspicuous factor missing from all of the theoretical approaches outlined above is a consideration of elicitation effects. As might be expected, no empirical study in the humoral literature reviewed above does more than suggest that the elicitation procedure might significantly influence informant responses. The first reason for this failure is that the classic approach to conducting ethnographic fieldwork has been for the anthropologist alone to gather all data, either through personal observation or by talking to informants. Obviously, no elicitation effect can be measured when there is no variability in elicitation method. However, this does not mean that the method has no effect on informant responses or behavior; it only means there is no way to compare observations made by different observers. For example, when only a single interviewer is used and questions are only asked once of informants, there is no way to determine the validity of the informants’responses. Only through comparisons between repeated interviews with the same informant or through use of multiple interviewers with a sample of informants can the degree of reliability of informant responses be estimated. The second reason for the lack of attention to elicitation effects, characteristic of the Culture-Centered school, is that it bases its approach on an analogy to the psychometric approach (most explicitly by the Cultural Consensus theorists-see Batchelder and Romney 1989). This analogy 32

INVESTIGATING EXISTING ETHNOGRAPHIC METHODS

can be inappropriate to the ethnographic case. Although Cultural Consensus is designed for use in anthropology (e.g., Romney, Weller, and Batchelder 1986, 313) where interviewing is the standard data collection method, it is based on the psychometric approach, which is designed for written tests. To base theories of ethnographic research on a model designed for questionnaires is to ignore significant differences in the elicitation situation between cognitive testing and most ethnographic fieldwork. Since studies in traditional anthropological populations typically involve illiterate subjects, interviewers often must be used for response elicitation in anthropology. Nevertheless, like the psychometric school, the anthropological adaptations of that approach give the elicitation procedure no role in their models whatsoever. The exclusion of elicitation effects, when they exist, has statistical consequences. For example, in the Boster and Weller (1990) study on humoral beliefs cited earlier, two groups were compared. The American sample was a group of undergraduate students, assessed using questionnaires during a class. The Mexican sample was, however, interviewed by an unspecified number of interviewers in quite a different context. When the goal is to calculate quantitative measures (e.g., degrees of informant variability), comparison of these two samples is statistically illegitimate since the methods of data collection are incompatible. In such a case, any model of attitudes or beliefs that ignores interviewer effects seriously overestimates the degree of independence between responses (which are in fact a function of the interviewer), and hence the amount of information in a given number of observations (Anderson and Aitkin 1985,181).Since the Culture-Centered and Pure Context approaches assume no elicitation effects, use of these approaches when such effects are in fact significant can lead to statistically inappropriate conclusions. If these models are used for interview-based data, the models should be modified to incorporate interviewer effects or, alternatively, the significance tests should be adjusted. For example, the assertion by Cultural Consensus workers that fewer than thirty informants typically suffice to characterize the consensus beliefs in a cultural group (Romney,Weller, and Batchelder 1986,326) is an underestimate of the number required for such a task when interviewer effects are significant. There is a large literature in sociology on the empirical significance of interviewer effects because sociologists often conduct large surveys employing many interviewers. Sudman and Norman Bradburn (1974) determined 33

CHAPTER 2

in their early review of these studies that the demographic characteristics of interviewers (e.g., ethnic group, sex, age, or education level) typically do not significantly bias informant responses. J. A. Hagenaars and T. G. Heinen (1982,105-6) even assert, also based on a review of sociologicalsurveys, that there is only a small difference in the consistency of responses when different interviewers were used in replicated interviews on a single informant compared to the use of the same interviewers. However, there has typically been little variability in the demographic characteristics of interviewers within any given sociologicalstudy (Hagenaars and Heinen 1982,95). Such studies have nevertheless found that the opinions of the interviewer concerning the substantive topic of the question and/or the interviewer’s expectation of the informant’s response do have significant effects on informant responses (for a review, see Brenner 1982). More recently, more careful investigation of interviewer effects has shown that just about every aspect of an interviewer can lead to biased responses, including their age, gender, social class, personality, and own opinions (e.g., Kane and Macaulay 1993). A textbook has even been written to deal with the problem (see Fowler and Mangione 1990.) It is more than likely, then, that characteristics of interviewers as well as their expectationsconcerning the interview situation will be significant when the interviewer and the informant do not share cultural backgrounds. Indeed, those survey studies conducted in a cross-cultural setting have typically shown significant interviewer-based biases in responses (e.g., Zehner 1970; Blanc and Croft 1992). For example, William Axinn (1991) found that female interviewers were more likely to get reports of contraceptive use in a rural population. Thus, it can be expected that ethnographic research will typically exhibit such effects,just like its sociological counterpart. It is also debatable whether the influence of the researcher on responses is reduced through the use of pen-and-paper tests, because methodological problems analogous to those of interviewing are encountered in the use of questionnaires. Many aspects of questionnaire design direct the attention of the respondent, just as in the interview situation, and thus have an influence on what is written down. An implicit recognition of this role for the questionnaire designer is evident in the massive literature in sociology and psychology concerned with how to write a valid questionnaire (for a summary, see Converse and Presser 1986; Foddy 1993; Fowler 1995). Considerable attention has been directed, for example, at the priming effects of question order on the nature of responses to 34

INVESTIGATING EXISTING ETHNOGRAPHIC METHODS

particular questions (e.g., Schwarz and Sudman 1992).Thus, even if questionnaires are used in an ethnographic context, as by some Cultural Consensus workers, this does not exempt such studies from the need to consider elicitation effects. Further, since decision making in small groups is a function of the membership of that group (e.g., Sheeran 1983;Janis 1982; Burnstein and Vinokur 1973), the ethnographer’s presence as participant observer probably has some effect on the behaviors of those being observed. Even naturalistic or noninvasive data collection techniques, which do not involve the presence of an observer (e.g., through the use of cameras or tape recorders), are likely to have some effect on what subjects say and do, unless the observation method is covert (in which case it runs into ethical difficulties) (Webb et al. 1981). Thus, all legitimate data collection methods require reflexive analyses.

Contextual Effects

While several of the ethnographic models seem to be aware of the different contexts in which data are collected (e.g., as suggested by the name Pure Context school), this is not in fact the case, since none of the approaches reviewed fully specifies a data collection situation. Context has a different meaning when data analysis is not based on reflexivity and statistical analysis. For example, the Pure Context approaches attribute all of the observed differences between informant observations to “context.” How that contextual variability is interpreted varies between (and indeed defines) the two different approaches in this school. Nevertheless, the concept of context used by traditional ethnography (i.e., the Situational Variation school), as well as the Informant Error school, is quite abstract. When using participant observation, ethnographic conclusions are derived from impressions written down after participating in the social life of the group being studied. The context of statements about a particular ritual is likely to be those aspects of a single event that caught the ethnographer’s attention, whereas the context of more general conclusions (e.g., about social institutions like kinship) is probably more vague (even subconscious) and based on a variety of situations. As a result, the context out of which an ethnographic statement is constructed may be unknown, even to the ethnographer. There is typically little explicit recognition in these approaches of the 35

CHAPTER 2

fact that data are collected at particular points in time and space, nor that there may be specific factors occurring during those circumstances that influence how informants behave. The notion of context that can be derived from the presentation of reports using these frameworks is therefore quite amorphous. O n the other hand, most of the Culture-Centered models do not allow that there is any random variability due to the context of data elicitation; all variation is explained as unknown but empirically estimated as differences in informant cognition. However, as I argued above, estimated parameter values are biased when the model is improperly specified, due to the exclusion of measurement effects. The exception in this case is the Cross-Cultural Psychological approach, which allows that variation not explained by task performance measures after the statistical reduction of the primary data is attributable to random, unknown methodological factors (e.g., measurement errors and/or situational variability). However, lumping these different factors together in this fashion is not very illuminating, and certainly not reflexive. This has not gone unnoticed, even among cross-cultural psychologists. In a recent textbook on the subject, Berry et al. (1992, 231) remark that “cross-cultural psychology has typically failed to work systematically at all levels to achieve a specification of context variables that are responsible for task performance and behavioral variation across natural habitats.” From this large literature concerned with intra-cultural variability in humoral beliefs, I have thus shown that there is confusion about what factors influence informant responses. This confusion arises because insufficient attention is paid to the nature of the data collection process (i.e., to the interview situation). Therefore, each of these theoretical approaches ignores one or more of the factors necessary to discriminate between the sources of variability in cultural beliefs (see table 2.1 for a summary of the analysis in this chapter). In conclusion, the analysis in this chapter indicates that the more closely an approach approximates the ideal of reflexivity, the more clearly the empirical research based on that approach is able to disentangle the sources of variation in primary data. This suggests that ethnographic studies based on a truly reflexive approach would be even more satisfactory. Adopting a reflexive approach to ethnography obviously requires considerable about-turns in both data collection and analytical procedures for any ethnographer used to existing approaches. These changes are embod36

INVESTIGATING EXISTING ETHNOGRAPHIC METHODS Table 2. I.

Approaches to the Analysis of Ethnographic Data

Sources of Variability

School of Thought “Pure Informant” “Situational Variation” “Informant Error” Cultural Consensus Cross-Cultural Psychology “ReflexiveAnalytical”

Measurement Error/ Elicitation Effects No No

No No* No*

Yes

Contextual Variability

Cultural Variation

No

Individual Level

Individual Level Group Level Group Level Individual Level

No No

Group Level Group Level

Individual or Group Level

Individual or Group Level

*dam typically elicited using impersonal questionnairesrather than interview

ied in the Reflexive Analytical approach described later. First, however, we need to determine that the kinds of factors a reflexive approach would highlight are actually important in determining what informants say and do. That is the job of the next chapter.

Notes 1. Historically, Cultural Consensus can be seen as an outgrowth of the earlier anthropological movement called ethnoscience, whose methods have largely been seen as a failure (Gardner 1985,249-53). The objective in both cases has been to uncover the shared aspects of cognitive structure related to cultural belief systems. The affinity of these two approaches is exemplified by the use of pile sort, free listing, and similar data collection methods in both cases (see Weller and Romney 1987). However, Cultural Consensus is different from ethnoscience in its analytical procedure, which has changed from componential analysis (a rather vague determination to infer structural rules of relation among the semantic terms belonging to some cultural domain based on participant observation) to a more psychometric approach of reliance on a group of informants and explicit statistical procedures. Although both approaches can be considered either cultural or psychological in orientation, I have considered them primarily cultural because of their interest in ethnically specific features of belief systems and comparison between cultural groups. The similarity between these two approaches is sufficient to consider Cultural Consensus as a more recent incarnation, and thus able to stand for both of them here. 2. Recent theoretical developments in the literature on psychological testing have centered on item response theory (e.g., Hambleton, Swaminathan, and Rogers 1991). This theory uses logistic regression models as outgrowths of the psychological testing’s historical foundation in the Rasch model (a one-parameter logistic model), which is distinguished from other logistic regression models primarily by having latent (unobservable) independent variables. 3. Boster and Weller’s argument is undermined, however, by the fact that they do not duplicate the analysis on which Mathews based her conclusions. Mathews showed that 37

CHAPTER 2 those foods assigned consistently by an informant to one or the other valences across two or more contexts exhibit lower rates of discrepancy. Boster and Weller, although they show inter-contextual consistency to be high overall among their Tlaxcalan informants, present no data on variation in the consistency of valence assignments by dzjkntfoods. Mathews’s point was that there is less variability in reported valences among those foods exhibiting higher consistency across contexts. Such foods should therefore exhibit a reduced likelihood that confusion among informants about the implicit context of questions will result in higher observed variability for the reported humoral qualities of those foods. Whether or not informant confusion about the context of questions is a significant determinant of observed variability, therefore, remains open to debate, even for humoral beliefs in rural Mexico.

38

3

IS REFLEXIVITY NECESSARY?

T

he humoral food avoidance systems discussed in the previous chapter are probably the best-studied systems in the anthropological literature for inter-informant and intra-informant reliability. As we saw, these studies reflect confusion as to the causal sources of observed variation and its ontological status. Whether this variability is due to methodological problems or represents real differences in belief has been the subject of some controversy. Logan and Morrill(1979), working on humoral beliefs in Latin America, argue that the inter-informant variation they recorded is a measure of subcultural variation in belief (i.e., real, not measurement or situational, variation). Researchers examining humoral systems in Malaysia also seem to assume that they are observing real cognitive variability, although their statements are vague. Carol Laderman (1983, 38) reports variation in the assignment of humoral traits among Malay, although she provides no quantitative description of the variability. Laderman’s only suggestion is that “variability within a system is a hallmark of Malay belief about the essential nature of life” (Laderman 1983,37). Brown (1976), on the other hand, investigating a humoral system in Peru, seems to favor a methodological source for the variation he observes. He suggests that increased variability in inter-informant response appears in his sample with respect to less typical foods, or foods that had conflicting valuations on the two dimensions he specified: culinary and medicinal use. Mathews (1983,841) argues further that the degree of intra-informant variability with respect to a particular food is a monotonic function of the number of classificatory dimensions along which that food has a consistent value of either “hot” or “cold.” The dimensions that Mathews identified were “danger,”“neutralization,”and “healthfdness.” There was only a 4.3% 39

CHAPTER 3

inconsistency between responses involving foods having at least two reinforcing valuations across these three cognitive dimensions; 10.9% for foods with meaningful import for only one dimension; 39.3 % for foods with conflicting valuations on two or more dimensions; and 39.1% for foods with no particular assignment on any of the hypothesized dimensions. Mathews therefore argues that the variation in her data is due to individuals having variable interpretations of the dimension to which the interview question was referring. In effect, she believes there is a shared cultural model for valence of foods along each of these dimensions; it was only informants’ variation in the interpretation of the dimension being asked about that led to different responses. Both Mathews and Brown, then, would predict variability in informant assignment of “hot” or “cold” status to a food based on context, along a number of dimensions. Boster and Weller (1990), who use the Cultural Consensus approach, tested Mathews’s contention that context matters. They questioned forty Tlaxcala informants using questions concerning individual foods designed to restrict the contextual reference for the assignment of the humoral quality, along the dimensions suggested by Mathews: first, no reference to any specific context; second, references to specific dimensions of food preparation, health, and “temperature.” Their list of eighty foods included seventy-five foods that were also on Mathews’s list. They argue that, contra Mathews (1983), consistency in the valence of a food across specific contexts does not reduce the degree of intra-cultural variability in the assignment of humoral quality to foods among their informants. This is quite a range of interpretations of essentially the same data. The basic question remains open: What exactly is responsible for intraand inter-informant variation in responses? Boster and Weller (1990,172) conclude that this controversy has not been resolved because “we believe there are still no clear criteria for identifying variation as either cognitive, contextual, or simply noise . . . theoretical, methodological, and substantive problems are tightly interwoven. Resolution of the question whether variation is contextual or cognitive can come only through careful comparison of the patterns of agreement among informants.” It should be clear by now that this is not enough. Information about respondents alone w i l l not solve the interpretive problem. Any procedure seeking to legitimize an ethnographic representation must account for how the idiosyncrasies of the database from which it is constructed arose in the first place. Chapter 2 suggested that current ethnographic methods 40

IS REFLEXIVITY NECESSARY?

generally fail to produce reliable inferences about cultural life because they don’t conceive of data collection as a social enterprise. The present chapter seeks to prove that these failures have real consequences; that data collection methods significantly influence basic data variability. It examines whether conceiving of data collection as a social interaction picks up salient aspects of the factors that generate variation in data. A case study will show that just those factors considered important by the data collection situation approach do in fact bias basic data. The existence of significant effects associated with specific interactions between informants and ethnographers prove that data collection is a truly social situation to which each party contributes some influence. The evidence presented in this chapter shows that any analysis that ignores data collection effects must be considered flawed. If an ethnographer is going to present variation to the reader, he or she must know which variation is real-that is, based in the informants themselves-and which is artifactual, or due to outside influences on the informants. Otherwise, the reliability and hence the credibility of the study is reduced. I begin, however, with a discussion of how we can operationalize a conception of data collection as a social enterprise.

The Data Collection Situation In ethnography, there is a fundamental division between the ethnographer and the “ethnographed.”For example, formal interviewing is a situation in which there is an interviewer (typically the ethnographer) and the informant, in the role of an interviewee. Even in the case of participant observation, we can speak of the observer and the observed. So any form of data elicitation will of necessity involve some kind of social interaction, even if it is observation from afar. Any scientific method useful to ethnography must begin by dealing with the social nature of social science. And if we need to be reflexive, our new mode of practice must begin with a reconceptualization of the nature of primary data. In particular, data must be seen as the result of a collection operation that constitutes a social situation with certain peculiarities.The ethnographer’s mere presence on the scene will often change the behavior of everyone in the vicinity who is aware of that fact. Since most ethical observations of people involve subjects being aware of probes for information, observations on such subjects 41

CHAPTER 3

are likely to reflect the fact that they themselves have at least partial control over what is observed. Furthermore, the ethnographer is not typically a member of the local community; the interaction may be ritualized (e.g., as a formal interview) and tap into certain cognitive routines specific to that situation, or there may be no such cultural resource available to the informant (e.g., when working in Third World populations without the concept of a social survey). All of these factors will affect the content of primary data, the basic materials an ethnographer has to work with. Traditionally, ethnographic researchers have relied heavily on the interview as the primary method of collecting data (Bernard 1988, 203). This prevalence of interviewing is not just characteristic of anthropology. “Certainly, the survey-interview is the best-known method of datacollection in the social sciences’’ (Dijkstra and van der Zouwen 1982,2). As a result, there is a vast literature on how to conduct interviews effectively (e.g., Foddy 1993; Fowler 1995; Sudman and Bradburn 1982). To focus our concerns, I will concentrate hereafter on this standard method of collecting information from ethnographic informants. Many of the conclusions derived from this focus will also be applicable to less-formal data collection methods, such as recurrent participant observation in certain kinds of social situations (for example, ritual contexts like funerals, dances or trials), since they too are also social and can be seen from a dyadic point of view as an at least implicit interaction between the ethnographer-as-observer and the observed. In the case of interviewing, the data collection situation involves recognition that there are two active participants, the interviewer and the informant (each of whom has a number of relevant characteristics); nonrandom contextual factors such as emotional holdovers from previous social encounters between the interviewer and informant; and more random factors such as the weather (Briggs 1986; Mishler 1986). Responses to questions can then be modeled as the result of a wide variety of factors, ranging from informants’private beliefs to their strategic decision making about whether to reveal their true opinions, miscommunication of meanings, and more spurious influences such as distracting events in the vicinity of the interview. Recently, however, interviewing as a data collection method has come under serious theoretical attack, resulting in a reconceptualization of the interview process itself. Recent critics of ethnographic practice (e.g., Briggs 1986; Mishler 1986; Suchman and Jordan 1992) have argued that 42

IS REFLEXIVITY NECESSARY?

the ethnographic interview must be considered a strategic, contextualized form of discourse involving two independent agents, the informant and the interviewer, each of whom has his or her own personal agenda. Furthermore, Briggs believes that anthropologists must recognize that the interview is a highly artificial context for communication in most societies. H e also recommends that the role played by the data collector should be explicitly recognized in the interpretation of interview data. Such critiques have led to a new view that can be summarized in four points: (1)interviews [should be seen] as speech events; (2) the discourse of interviews is constructed jointly by interviewers and respondents; (3) analysis and interpretation are based on a theory of discourse and meaning; (4) the meanings of questions and answers are contextually grounded. (Mishler 1986, ix)

According to this new view, each interview’s context (i.e., situation in space and time) and metacontext (i.e., the participants’perception of the situation of the interview) are necessarily unique. Since the scientific process can be (somewhat crudely) described as the search for regularities, or the presentation of competing explanations for patterns in recurring phenomena (Kuhn 1970; Lakatos 1970), this new view suggests there can be no scientific analysis of interview-based data because one interview, as a unique event, cannot be compared to any other. I will argue here, however, that the new view, while a valid critique of the traditional view, does not necessarily lead to the conclusion that a radical reformulation of the interpretation of interview-based data is required, or that interviewing should be rejected altogether as an irremediably flawed methodology. Instead, a change in perspective is required: if the participants, as well as the situational context itself, are viewed as forces acting to bias informant responses during interviews, then an examination of these causes of variation can form the basis of a rigorous approach to interview data. In order to determine the nature and relative significance of different biases controlling interview responses, a common currency of influence is required. I argue that through the use of formal methods for the disambiguation of causal influences (such as statistical models), the degree and types of biases that different sources of variation have on informant 43

CHAPTER 3

responses can be isolated (and in the case of statistical models, measured by their statistical significance). This formal approach requires specific features of field data collection protocol, such as the use of multiple interviewers and structured interviews repeated on informants. To isolate the sources of variability in ethnographic interview-based data, a theoretical model of the interview situation that identifies all the possible factors that can bias what informants say must be developed.Traditionally, ethnographers have implicitly viewed the interview process as involving an adept interviewer, the professional anthropologist, posing well-understood questions to a pliant informant who responds freely and honestly. Such a view implies no role for ulterior motives, miscommunication, or chance during the process of gathering interview-based data. While ethnographers have been careful to attribute such strategic behaviors to individuals when recording social relationships among the people they study, these factors have not been considered to intrude into the behavior of those same people when they serve as informants. The irony is, therefore, that while there has been great sophistication in ethnographic descriptions of social behavior, there has been little concern for the intricacies of human interaction during data collection in either ethnographic theory or practice. Before variability in informant responses can be partitioned to the effects of various factors, the set of factors that might influence responses must be determined. Recent critics of traditional ethnographic practice have suggested that the interview should be considered a strategic interaction between the informant and interviewer in which each of these participants, as well as the situational context itseK act as forces that bias informant responses (e.g., Briggs 1986; Mishler 1986). This perspective suggests that, in general, there are four categories of influence on what informants do or say: contextual or situational factors, reflecting circumstances surrounding a particular interaction; the ethnographer’s behavior; interactions between characteristics of the ethnographer and informant; and finally the informant’s own unbiased beliefs or behavior. 44

IS REFLEXIVITY NECESSARY?

Each of these factors should be seen as deriving from the influence of one or more of the factors identified by the new view of the interview (see Mishler’s list above). In effect, the interview is seen as a speech event; the interview-as-discourse is considered to be a result of interactions between the participants, and analysis isolates the meaning of speech as grounded in that social context. Statistical methods can isolate the effects due to these four sources of variability. But to analyze the relative contributions of these four categories of influence on responses, there must be systematic variation in the factors within each category. For example, interviewer effects can only be examined when more than one interviewer is used. Further, the biases interviewers introduce into what is recorded as the informant’s response are best estimated when interview methods are comparable between interviewers. Preferably, interviewers should be of different degrees of cultural relatedness to the group being studied in order to introduce variability into the types of interaction expected between interviewers and informants. In addition, interviews have to be repeated with the same informant to isolate situational factors affecting what informants say. Thus, in order to isolate the four types of variability, several features of the primary data collection protocol are crucial to the implementation of this approach to the data collection situation: a number of interviews (sufficient for statistical analysis) repeated with the same informants; use of randomly selected informants exhibiting a wide range of cognitive, life-history, and other characteristics (not just “key” or “expert” informants); use of interviewers whose own cultural background differs from that of their informants to varying degrees; and use of an interview protocol that maximizes the degree of comparability between interviews.’

In this chapter, I outline a method for discriminating between the variability in ethnographic data due to differences among informants’ own unbiased beliefs or behavior and variability due to other factors that also influence their socially situated responses. The method is particularly directed toward determining the relative significance of factors that tend to bias informant responses during ethnographic interviews concerning 4s

CHAPTER 3

individual beliefs, which constitute the bulk of primary ethnographic data (Bernard 1988, 203). I also provide an example illustrating this method’s use. By partitioning variability in interview responses to different sources, I can determine whether all the observed variability is due to spurious or purely methodological causes, or represents, at least in part, meaningful differences in belief between individuals in a cultural group. This perspective emphasizes that interview responses are strategic behaviors. It is therefore important to distinguish between what a person says they believe and what they represent t o themselves as their belief (i.e., their unspoken belief). An interview response or vocalization is a behavioral reaction to a perceived situation, and therefore involves a complex decision-making process. Following suggestions from cognitive science concerning the architecture of human cognition (e.g., Haugeland 1985),I argue that three aspects of informant cognition affect interview responses: knowledge, reasoning, and memory or recall2 Knowledge can be considered as the informational content of long-term memory; recall, the transfer of information from long-term to working memory; and reasoning, the manipulation of information in working memory prior to behavior (Baddeley 1986). Even if two individuals were to have the same knowledge base, they nevertheless could make different responses to the same question. For example, some informants may respond, “I believe X even though they would respond “I believe Y”under other circumstances, because belief Y did not come to mind at the moment of questioning. Meanwhile, other informants successfdly recall belief Y and report it. Alternatively, some informants, afraid of the uses the interviewer might make of their responses, may reason that a lie is the strategically appropriate response (i.e., report belief X when belief Y is what they think to themselves). Again, other informants may be less suspicious, and simply report belief Y. These examples illustrate that variability in responses does not necessarily imply differences in belief; since differences in recall ability or reasoning between individuals may also cause variation in reports. This distinction is equivalent to that made in sociolinguisticsbetween “speech acts” and “belief”: the public behavior of speaking is to be conceptually distinguished from internal mental representations. There is no way to know what another individual is thinking unless mental processes leave traces in behaviors ofwhich the individual is unaware (e.g., facial expressions outside conscious control), thus betraying their thought. The interpretability of these unconscious behaviors also depends on the 46

IS REFLEXIVITY NECESSARY?

interpretability of that behavior as a signal of a particular mental state. These epistemological and social issues are beyond the scope of the present chapter. I will make the simplifying assumption that the only information available for interpreting others’ beliefs is their reports of those beliefs. Thought will be considered intrinsically private and unapproachable. Only public behaviors such as speech can be used to infer belief. However, this is not to imply that verbal statements of belief need therefore to be taken at face value. The ability to interpret remains. However, we still need to model how people talk about their beliefs. The concept of context that arises from a reflexive approach to the notion of data collection as a situated event is more specific. In this case, contextual variability is due to influences surrounding the elicitation event itself. Examples include the kind of weather at the time of the elicitation procedure (e.g., rainy days may make informants more morose and uncooperative) or the presence of bystanders to an elicitation event (if other individuals are within earshot, an informant’s behavior can be influenced by this public situation). Other influences on what informants do or say are unknowable even though they are specific to the elicitation process itself. Nevertheless, using a statistical model, the influence of such factors can be estimated as the proportion of variability in informant responses that remains unexplained by dl of the known factors considered together. This second type of contextual variability, residual variation (measured by the error term in the statistical model), can be considered due to the set of factors that rundomb influence how informants behave. This influence, although random with respect to the phenomenon of interest (ideally), may nevertheless shed light on the phenomenon, for example by being correlated with specified contextual factors. (Although residual variables are technically supposed to be randomly associated with the elicitation situation, in empirical cases this is seldom strictly true.) Even if there is no bias or directionality of effect due to these unspecified factors, the relative importance of this aspect of context can be measured by the overall influence of these unknown contextual influences relative to factors known to influence how informants behave. At least the size of their effect provides an idea of how well the causal model for the phenomenon explains informants’ observed behavior. Let us now see how this approach can be implemented with real ethnographic data. 47

CHAPTER 3

Background on the Case Study The data from this case study concern beliefs about the edibility of foods in a traditional population living in the Ituri Forest of northeastern Democratic Republic of the Congo (formerly Zaire). The Ituri population consists of a variety of ethnic groups, subsisting either by slash-and-burn horticulture or by foraging in the forest, although neither strategy is used exclusively by any particular group. Each horticulturalist group speaks a language from either the Sudanic or Bantu family, as befits their origins in present-day Sudan and central African states, respectively. Due to the multiplicity of self-identified ethnicities in this population, I aggregate them according to primary subsistence technology and language group. This results in four ethnic categories: Sudanic and Bantu horticulturalists (primarily Lese and Budu, respectively), along with Efe (Sudanic-speaking) and Tswa (Bantu-speaking) foragers. The foraging groups have complex social, economic, and cultural relationships with particular horticulturalist clans of the same linguistic category. Only a few examples of Ituri food taboos can be provided here (see appendix A for more discussion). First, family-based taboos are quite dangerous to violate since the consequences include the death of the consumer, his or her child or other relatives, or the violation of a relationship between the clan and supernatural entities. For example, some believe that “Bioka moja [a type of mushroom] spring up from our family’s graves,” implying that ancestors have some relationship with this plant, which is therefore not eaten. Particular clans also refuse to eat some animals because that animal helped an ancestor return to the village when lost in the forest. Some individuals report rejecting certain kinds of carnivores (e.g., raptors, genets) because they eat other animals, thus mimicking the role of humans as predators. Many individuals refuse leopard meat because leopards are viewed as agents of witches seeking to attack and kill people. Other animals are refused because they function as harbingers of the death of someone in the village, either when they enter the village (e.g., insects) or give particular calls (in the case of some birds). While some types of avoidance are absolute, lifelong restrictions, others are specific to certain periods of life. Perhaps the most complicated rule is that against animals that cannot be eaten after puberty until the individual gives birth to a combination of at least one male and one female 48

IS REFLEXIVITY NECESSARY?

offspring. Yet other restrictions are positive in nature, or require the consumption of the animal under specific conditions. For example, after giving birth, a woman must be fed a particular animal by her husband in order to restore her strength. Another major category of taboo surrounds pregnancy and tends to involve the homeopathic principle of contagious similarity between an item consumed by the mother and anomalous characteristics acquired by the fetus, discovered when the mother delivers. For example, some women believe that “If I eat yama [tree dassie or hyrax] when pregnant, my child will be born with only three fingers [like that animal].” Spotted catfish are also believed by some to produce similar markings on people’s skin if that food, which is forbidden to them, is nonetheless eaten. In similar fashion, pregnant women refuse to eat a variety of animals because they believe the developing fetus will be born with some anomalous feature characterizing each animal. Personally dangerous avoidances arise from consumption experiences and are reflected in attitudes of disgust. Consequences are typically limited to feelings of nausea or stomach problems. For example: “Matudu [snail] ‘walks’ in people’s urine beside the village.” These beliefs can be quite idiosyncratic: for example, “I fear to eat rnutufi [piping hornbill] because it has the same name as a man I know.” I assume that prior to enculturation or personal experience with foods, the rudimentary idea is that animal foods are consumable. In the Ituri, a particular ethnic group might specialize in a specific type of taboo, or tend to exhibit certain kinds of taboos with respect to different classes of animal. However, they all share this same basic vocabulary of belief-that is, one classification of taboos can be used to describe all four of the ethnic belief systems.This means these systems can be compared. Since these different types of avoidance have different etiologies (some being invented as a personal reaction to experience with the food, while others are transmitted to the individual from a person with authority), and different kinds of consequences (e.g., illnesses of various sorts or attack by witches), it is likely that they have different kinds and degrees of cognitive salience. In addition, the various ethnic groups tend to report these different types of avoidance with different frequencies. For example, foragers typically rely heavily on homeopathic avoidances (Aunger 1992), as might be expected by those who emphasize the homeopathic nature of “primitive mentality” (Levy-Bruhl 1923). AU of these 49

CHAPTER 3

factors should influence an individual’s ability to remember, and thus repeat, reports of their food avoidances. Informants were questioned with respect to their food avoidances concerning 140 different animals. Interviews were repeated with sixty-five informants using different combinations of interviewers (interviewers besides me were members of the local population), in all cases employing the same structured interview format. The length of time between interviews was distributed in roughly normal (i.e., Gaussian) fashion, between a few days to almost a year. The relative importance of various factors affecting the repeatability of interview responses was then determined by means of a multivariate statistical model that was estimated using logistic regression. This statistical model specifies the various effects that the cross-cultural interview, as an event, is likely to exhibit. The model includes informant as well as interviewer characteristics, influences due to the interaction of these two participants, and influences due to the situation of the interview itself. This allowed me to partition the likelihood that an informant would make different responses to the same question (which happened in about 20% of cases) to these various factors. The results from this procedure are described next. Differences in paired responses can arise through either forgetfulness or mistakes, which appear to reflect the two different types of cognitive operation described by these commonplace words: forgetfulness is when an informant mentions an avoidance on one occasion but not another, while mistakes occur when an informant gives two different avoidances as the answer to the same question. The results from the multivariate statistical models show that the probability of an informant responding in either of these ways is a function of a wide variety of factors. The relative significance of these factors is consistent with a commonsense notion of what should determine responses to questions about personal beliefs. In brief (and averaging over the two models for forgetfulness and mistaken responses), 14%of the explained probability of a discrepancy in response was due to interviewer effects, 13% to interactions between informant and interviewer characteristics, and 73% to aspects of informant cognitive variation (53%to knowledge-based cognitive variability, 19%to reasoning or ability-based differences between informants, and l%to memory or recall effects). In other words, after excluding variation due to methodological control factors, about a third of the variability in re50

IS REFLEXIVITY NECESSARY?

sponses is due to each of the three major categories of non-situational effects: interviewer effects, informant characteristics, and interactions between these two actors. The overall significance of the statistical models also indicates that a considerable proportion of the variability observed in responses was not explained by the model: the overall lack of fit of the multivariate models suggests that unspecified situational factors probably also had an impact on informant responses. These other systematic influences on the interview situation simply couldn’t be methodically identified and so were not included in the model. Yet other contextual factors had only randomized effects on responses. Each interview thus appears to be a unique event; informant responses reflect to a considerable degree the circumstances of the interview situation itself. Let me now explain these results in somewhat greater detail.

InterviewepBased Effects Two types of interviewer effects can be distinguished: personality and cultural differences between the interviewers themselves (inter-interviewer effects) and the interviewer’s degree of experience at interviewing. Ten to 15%of explained variability can be attributed overall to these two factors. While the influence of the interviewer is not as great as that of the informant in determining the response made to a particular question, the relative size of the interviewer effects is still alarming methodologically. There was a pronounced temporal trend in the data revealed by this analysis: discrepancy rates decreased significantly as interviewers became more experienced. A complex analysis of the statistical results (see appendix B for details) suggests this trend toward better interviews, as interviewers became more experienced, can be attributed to the interviewer learning to probe more effectively, especially with recalcitrant informants (which explains why interviewer experience was more effective at reducing forgetfulness than mistake-making). Interviewers learned to be more pointedly assertive in prodding informants for answers as it became clear that some were simply more recalcitrant as a matter of personality or predilection when dealing with a situation that was to them both unfamiliar and odd. It was this basic learned skill that had the most pronounced effect on interview outcomes, according to the statistical model. 57

CHAPTER 3

An intriguing result in this regard was that-despite language fluency and a year of previous field experience-as the ethnographer, I enjoyed less advantage from greater experience with the task than did the native interviewers, consistently performing less reliably than the native interviewers. This could be put down to differences in the way particular interviewers perceived their mission. I was explicitly interested in uncovering variation in belief, and innocent of what the individual I was interviewing should say, while the local interviewers were more concerned to get the right answers to their questions (because, as members of the same cultural group, they knew what their informants should say). However, this difference in mission appeared to have little influence on interview responses over time (this mission did not itself change with time or experience). It might also be the case that my mastery of the language improved with time, and thus my ability to elicit reliable responses from informants in this language. But in fact, I was quite fluent in Swahili before beginning to collect interview data on food avoidances. Based on the analysis, it seems clear that my differences in performance cannot be attributed to either language or cultural differences (differences that distinguished other interviewers from their informants) or to my probing ability (which improved in all interviewers). Instead, they may be due to the special qualities derived from being an outsider.

Informant-Based Effects Linda Garro (1986), in a study of ethnomedical knowledge in Pichataro, Mexico, found that traditional curers had somewhat greater knowledge of ethnomedical treatments than nonspecialists. Boster’s (1985) detailed study of Aguaruna manioc cultivators showed that women, who typically spend more time gardening, have greater abilities to discriminate manioc varieties than men. Further, adult Aguaruna are more consistent in their classifications of manioc than children. Schooling also appeared to decrease an individual’s familiarity with the traditional practice of manioc cultivation. What individuals know of their culture is therefore at least partly a function of their social roles. However, such studies are not able to determine whether social role-based variation is due strictly to differences in access to specialized knowledge or to differences in reasoning, since social roles may be linked to cognitive abilities. This deficiency is due to a methodological problem. To answer a question typically involves an 52

IS REFLEXIVITY NECESSARY?

ability to recall what is stored in memory, while specific bits of information must also be present in one’s memory in the first place. However, as recognized in psychological testing, different kinds of questions stimulate different aspects of informant cognition, and cannot necessarily be treated equally, within or between informants. This confounding of knowledge and reasoning has profound implications for studies of cognition, and the lengthy traditions of educational and 1.Q testing have established its ubiquity as a methodological problem (called “responsebias effects”in the psychological literature; for a summary, see Goldstein and Wood 1989). It is necessary to discuss several factors related to informants that are nevertheless methodological in import before finally considering the variables that indicate meaningful variability in the Ituri population’s beliefs. Since differences in behavior and reported beliefs under given circumstances are due to variation in what different informants believe it is proper to do or say, I will consider cognitive variation to be the wellspring from which intra-cultural variation in belief and behavior arises. Further, I suggest three different aspects of informant cognition can influence their decision of how to respond to interview questions: knowledge (an informant’s store of information), reasoning (how that information is manipulated), and memory (or recall ability). Knowledge is the basis on which reported beliefs are formed. However, at any moment, an informant may experience a failure in their ability to recall the necessary information. In addition, the public qualities of the interview situation may bias the reasoning that an informant performed on the information that is successfully recalled. Only knowledge differences are surmised to reflect differences between individuals in privateIy held, as opposed to merely reported, food avoidance beliefs. However, an informant’s own cognitive abilities can also affect their responses in ethnographic interviews. One aspect of informant cognition that influences responses is strategic reasoning. The fact that informants are reticent about expressing their personal beliefs to some interviewers introduces methodological problems in the inference of privately held beliefs. Besides reasoning, informants’ recall ability influences how they respond. Recall ability, like strategic reasoning, represents another confounding effect in the expression of belief, since recall failures can, like the presence of a particular interviewer, bias the reasoning that presages a response. For these reasons, only variables representing knowledge variation (discussed later) will be taken here to reflect meaninghl variation in belief. 53

CHAPTER 3

Before I address the substantive issue of knowledge differences or discuss effects related to memory and reasoning, I must also deal with the fact that the statistical models were estimated using a set of questions, not individual ones, and that these questions probed for different domains of knowledge. I found that the reliability of different questions was the single most important determinant of variability in ethnographic interview responses (36% of explained variation in the probability of a discrepancy). Of course, informants should have different opinions about the edibility of different animals, because some they can eat for one reason, others for a different kind of reason. So it is quite natural for much of the probability of a mistake or forgetful event to be associated with the breadth ofpotential responses to each question (certain kinds of avoidances will only be relevant to a given class of animal foods). Variation in informants’ abilities to recall information pertinent to the expression of beliefs accounted for about 20% of the variability explained by the statistical models of discrepant responses. These differences in recall ability were indirectly measured by an informant’s ability to make consistent responses, despite (1)having a higher overall number of avoidances to remember; (2) having a longer period of time elapse between interviews; or (3) having to answer questions with lower degrees of cognitive salience. This last type, the question-specific salience effects, shows there is also significant variability in responses between questions, as expected when each question addresses different personal beliefs (in this case, food avoidances with respect to a particular animal). There was no influence of time between interviews, and little impact of the overall number of avoidances an individual had to remember. So all recall-based variation simply reflected a question’s salience, or the likelihood that a broader range of responses was possible for some questions compared to others. That no significant effect of what could be called variation in cognitive ability on the reporting of food taboos could be found does not mean that significant variability in cognitive abilities does not exist in this population. No doubt it does! The results only suggest that variability in cognitive ability does not impinge on an individual’s ability to successfdly master their food taboos. Behaving in accordance with food taboos is a prerequisite to functioning as a competent member of Ituri society. Individuals do not share an identical life history, and therefore gain access to, or come into contact with, different aspects of social knowledge. But, from these data, it does not appear that the Ituri social system erects barriers to 54

IS REFLEXIVITY NECESSARY? membership in the social group based on cognitive abilities. That individual variability in access to information does not impede social hnctioning in the Ituri, despite the lack of centralized authorities, is impressive. Last, the amount of time elapsed between an informant’s two interviews is included to account for the possibility that the informant learned a substantially different belief during this interim. I assume that as the time between interviews increases, so does the likelihood that an informant experienced some learning related to a particular belief. However, unlike the other control factors, the Elapsed Time effect is relatively insignificant (only two percent of variability for mistakes). Thus the time between interviews has little impact on the decision processes underlying the probability of repeating responses. This suggests that, for most informants, these cultural beliefs are not changing significantly, at least over the several months between most interview^.^ Since most avoidances are culturally sanctioned taboos believed to be part of one’s social identity, this lack of temporal variability is understandable.

Interaction Effects Of roughly equal importance to the interviewer-specific effects is a variety of interactions between the interviewer and informant. Significant interviewer-informant effects suggest that an interaction between characteristics of the interviewer and those of the informant has an important effect on what informants actually say. A significant effect implies that informants’ perceptions of the interviewer’s intentions change their responses about their personal beliefs: their responses are strategic. Cultural differences between the two participants in the interview include major and minor language and personality differences. Ten percent to 15% of the variability in interview responses explained by the statistical models was due to interactions between characteristics of informants and interviewers: informants biased their responses in specific ways when faced with interviewers of particular types. For example, nearly all of the interviewers showed positive effects of dealing with better-schooled informants: informants with more schooling reported more avoidances, and were more consistent. Educated informants perhaps feel the need to impress fellow graduates (their interviewers) with the range of their knowledge. Different interviewers also had variable success in eliciting different types of avoidances from 55

CHAPTER 3

the youngest informants. In particular, the interviewer who specialized on foragers had much better success than did the other interviewers with this younger age group. This is probably attributable to the fact that, as shown in another analysis (Aunger unpublished data), younger foragers have acquired considerably greater knowledge of avoidances by a given age than have their village-bound horticulturalist peers. Gender-based effects were also in evidence. For example, women became more forgetful as the difference increased in sociopolitical power between their own cultural group and that of the male interviewer. These interaction effects suggest that the ethnographic interview can perhaps best be seen as a dialogue in which two agents seek to negotiate with or manipulate each other, often using the clumsy instrument of a foreign language. The results here confirm the intrinsically social nature of data collection in ethnographic situations, and justify our emphasis on making explicit a consideration of data collection methods in any analyses.

Meaningful Variation in Belief The data situation approach also provides insight into substantive issues. When intra-cultural variation is observed, the truly important question is whether that variation is real or simply due to methodological and situational factors. The primary virtue of an approach that attempts to isolate all possible sources of variation in primary ethnographic data is its ability to determine whether variation between informants in their reported beliefs is due entirely to confounding methodological and situational effects, or at least in part reflects variation in what informants believe. If the variation is not real, it can be safely ignored; the cultural group can be considered uniform; and there is no constraint on making inferences about the group as a whole. However, ethnographic studies typically cannot determine whether observed variation is real or not. For example, as Boster and Weller (1990,172) point out, it is very difficult to tell from the studies of humoral beliefs (reviewed above) whether there are any real differences in belief among informants. Most of the variables included in the statistical models were meant to control for some aspect of the interview situation that might interfere with the expression of beliefs by informants. The task here is to isolate know(edge differences from the effects of other aspects of informant cognition 56

IS REFLEXIVITY NECESSARY?

related to the questions themselves. Such differences can be seen in the results concerning variables that do not reflect methodological variation. Informant characteristics included in the model were ethnic group, sex, age, and degree of schooling. I call these variables indicators of the various social roles informants can play. That social roles can lead to various types of cultural expertise has also been suggested by others, for example, John Roberts (1964) and Anthony Wallace (1970). Indeed, each of these variables has received extensive attention in the social sciences (i.e., in literatures on ethnicity, gender, ontogenetic processes, and acculturation).They therefore seem reasonable a priori criteria for dividing a population into groups that might exhibit different cultural belief systems. After controlling for any influence of the salience of different questions, variation in cognitive ability, or learning between interviews on the propensity to exhibit discrepancies, about 17% of explained variability was associated with social roles. This shows that individuals with different types of life history gain access to different culturally designated beliefs. Thus, social roles seem to affect the kinds of knowledge that an individual acquires about the food avoidance belief system. For example, an informant’s degree of experience in formal schooling improves his or her ability to repeat previous responses. Schooling has often been shown in the cross-cultural literature to have effects on cognitive abilities (Irvine and Berry 1988, 28). In Aunger (1996), I showed that informants with more schooling report a greater number and variety of food avoidances on average. However, since schooling does not reduce the tendency to make mistakes, the present results indicate that, like increasing age, greater exposure to schooling simply increases knowledge of the food avoidance system in general, and likely does not result in newfound cognitive abilities to deal with that knowledge. Ontogenetic effects are also evident in the results: younger informants have greater difficulty in repeating previous avoidance reports than do older informants. This suggests that older informants have greater knowledge of the food avoidance belief system, and therefore exhibit reduced probabilities of making mistakes in their responses. Finally, the results also show significant variability in the probability of repeated responses along the lines of ethnic group and sex. The horticulturalist ethnic groups are more likely to make mistakes than the foragers, probably because they have more elaborate food avoidance belief systems (Aunger, unpublished data). Men have a significantly higher 57

CHAPTER 3

propensity than women to forget their avoidances, probably because women believe that more severe consequences follow from violation of the taboos they report (Aunger, unpublished data). Since culturally dangerous beliefs such as food taboos probably have greater cogency than personal rejections of food (e.g., due to stomach upset), women are less likely to forget those taboos. We may conclude that the reason these different groups (i.e., malelfemale, SudaniJBantu) exhibit different probabilities of discrepant responses is that they have significantly different types of avoidance to report. Even after controlling for a wide variety of possibly confounding effects-from variability in reasoning and recall abilities to the salience of the stimulus used to elicit responses to interviewer and situational effects-individuals of different social categories still exhibited different probabilities of making discrepant reports concerning their food avoidances. This indicates that responses are more salient on average for some informants than for others (i.e., their search for responses to a question more reliably turns up the same chosen response). When informants have a higher proportion of more salient responses (e.g., dangerous taboos), they are less likely to make mistakes in reporting such avoidances or to forget them when asked repeatedly about them. Since responses are reports of food avoidances, this in turn implies that such informants tend to have a higher proportion of more salient avoidances to report. Because the set of questions (topics) for all informants was uniform, this implies that different informants have different sets of food avoidances. The variability in individual sets of food avoidances is not equivalent to the variability in individual sets of responses made by informants, because biases in responses have been controlled for in the analysis. Nevertheless, the substantive point is that this result establishes that there is meaningful variability in belief in this p~pulation.~ The social role effects indicate there are significant differences between individuals in their privately held beliefs with regard to the same question. Since many of these food avoidances are culturally transmitted taboos, it appears that socially sanctioned belief systems can nevertheless exhibit significant intra-population variability. In such cases, various subpopulations simply cannot be characterized by a single set of beliefs, as has been ethnographers’ traditional practice. Nevertheless, for both kinds of discrepancy, social roles account for only about 17% of variability: the majority of the observed variability in this ethnographic data is methodological in origin. 58

IS REFLEXIVITY NECESSARY?

This has its own implications for ethnographic practice, suggesting that studies that do not control for such influences on data may misrepresent ethnographic reality.

What ItAll Means

It is important to note that the task being asked of respondents was merely to repeat an answer to a quite straightforward and highly specific question (“DOyou have any restrictions against eating X?”). Food avoidances are either highly personal beliefs, reflecting experience with foods, or dangerous taboos that should be memorable. The salience of these beliefs should thus be quite high overall, especially since they involve highly valued foods like meat. This may not necessarily be true of other types of cultural traits, which may reflect less strongly held opinions or preferences. Thus, the importance of methodological variability in ethnographic data may be underestimated by the present case compared to that likely to be found in other kinds of belief systems. From this perspective, the conclusion of this study that methodological problems are significant becomes an even more forceful warning against the standard ethnographic practice of relying on single, uncontextualized responses as the truth, not only for the informant in question, but also for the larger group that he or she is said to represent. As suggested by critics of standard practice, then, interview responses should be viewed as strategic replies to a particular context perceived by the informant, who is considered a dynamic agent attempting to manipulate a situation involving other such agents. Variation in responses thus provides evidence about how the interactants view the interview situation (Kane and Macaulay 1993,24). Further, a myriad of factors influence responses, from the fact that the informant was hired to the weather that day. However, the specific effects of each situational factor simply cannot be determined. In addition, what informants believe privately (i.e., what they represent to themselves as their belief) can never be known with certainty. The only way to gain an understanding of the major factors governing responses is to investigate variation in verbal behavior considered as “speech acts” (Austin 1962). Responses that recur over a wide variety of contexts without being contradicted by the respondent’s other behaviors can be fairly considered an accurate representation of his or her belief, or at least characteristic of that informant. 59

CHAPTER 3

Another important implication of this study is that an individual unfamiliar with the culture of the informant is less likely than someone who is raised in the same cultural group to be able to elicit replicable responses. The model results suggest that increased degrees of similarity in cultural background between interviewer and informant improve the consistency of responses. Nonetheless, increased experience in a cultural domain makes an anthropologist’s performance more similar to that of interviewers who share cultural backgrounds with their informants and improve the anthropologist’s ability to elicit reliable beliefs from informants. These results suggest that effective interpersonal communication, especially in the artificial context of an ethnographic interview, requires attention to many subtle cues grounded in implicit cultural rules for social interaction, as sometimes recognized in the literatures on sociolinguistics and pragmatics (e.g., Duranti and Goodwin 1992). The results presented here point to the significance of a wide variety of factors in the determination of discrepant reports, many of which are not usually considered. Question-dependent effects are the specialty of psychometric and cross-cultural psychological research on intellectual ability (see e.g., Hambleton et al. 1991), but have rarely been investigated with respect to other types of informant responses, such as culturally acquired beliefs. O n the other hand, while there is a small literature in cultural anthropology on intra- and inter-informant variability (for a review, see Boster 1987; Pelto and Pelto 1975), such studies do not consider the effects of interviewers on responses. Sociological surveys (e.g., Blanc and Croft 1992) typically are concerned with interviewer effects, but do not often look at question-dependent effects. Another virtue of the present study, therefore, is that it allows the simultaneous treatment of all these effects in a single model, thus determining their relative significance. Only when comparable studies are undertaken in other groups can the endemism of the kinds of methodological difficulties uncovered here be determined. The results indicate that most of the statistical variability in responses reflects variation in informant knowledge: real changes in belief are associated with changes in the topic of the questions themselves. A particularly intriguing result is that the anthropologist performed less reliably than native interviewers, suggesting that one’s cultural background is an important determinant of the ability to elicit replicable responses from informants. 60

IS REFLEXIVITY NECESSARY?

The overall significance of the statistical models indicates that they do not explain a considerable proportion of the variability observed in responses. This implies that unspecified situational factors were also important influences on informant responses. Few existing ethnographic studies account for all of these sources of variability in primary data; in this respect, they are problematic. I therefore advocate greater attention to these problems in future ethnographic practice as a way of re-legitimating ethnographic reports. One implication of the demonstrated significance of purely methodological factors is that all analyses using ethnographic data should explicitly account for the way in which the data were collected. Only by using such reflexive methods can researchers assess the quality of the information they analyze (Pelto 1992). While isolating the sources of variation in ethnographic data cannot be the sole objective of cultural anthropology, it is a necessary first step whenever there are questions about the methodological purity of data. In such cases, analyses that do not distinguish between the proportion of variability that is purely methodological and that which represents meaningful variation can lead to inappropriate conclusions. By partitioning the observed variability in responses to a wide variety of factors that tend to bias what informants say during interviews, the present type of analysis also isolates phenomena that are of intrinsic interest from a non-methodological perspective. For example, in the present case, individuals with a larger total store of food avoidances to remember found it more difficult to respond consistently to questions regarding particular avoidances. This result has obvious import for cognitive psychology. Such substantive results can provide important clues for future research into the psychological and social consequences of a belief system. Finally, since most ethnographers have used interviewing under similar field conditions, this analysis carries the somewhat grim message that serious methodological and interpretational problems probably characterize the majority of traditional ethnographic data. Indeed, the anthropologist is the most unreliable interviewer because of his or her relative lack of familiarity with the culture of the informants. However, there is the more hopeful message that increased experience with the cultural group under study can improve one’s ability to elicit replicable responses from informants. O n the substantive question of whether there is meaningful variability in beliefs related to particular foods, the results indicate that even after 61

CHAPTER 3

a wide variety of methodological problems have been accounted for, intracultural variation in food avoidances does exist in these Ituri groups. Most important, the fact that meaningful variation in belief was discovered in this population suggests that ethnographers should consider the use of analytic and explanatory approaches that recognize the significance of intrapopulation variation in beliefs and behavior (see Poggie et al. 1992 for a similar statement). Once the purely spurious (i.e., random and methodological) variability has been abstracted, meaningful differences in informant beliefs account for less than one-third of the observed variability in reports of belief. Indeed, the preponderance of variability in what informants said to their interviewers was unrepresentative of differences in informants’ own beliefs, but rather reflected biases produced by various methodological factors. The major conclusions of this study are therefore that interview responses are quite unreliable, and that the majority of variation in reported beliefs is not meaningful. However, meaningful variation in belief about specific topics does exist, even within ethnic groups (e.g., along the lines of age and sex). It is thus inappropriate to characterize food avoidance beliefs in this population without referring to this individuallevel variation. By glossing over intra-cultural variability through the use of typological language and group-level explanations, ethnographers may misrepresent the cultural systems they study. As we will see in the next chapter, some have suggested that we can adopt just this traditional route to ethnographic description, albeit using new scientific methods. But chapter 4 shows this route remains fraught with problems, despite methodological advances. Notes 1. Note that the kind of data produced by the method advocated here can be represented as a matrix of codes representing the responses of a set of informants to specific questions. Since it is possible to code information from quite unstructured interviews into responses to a specific set of questions, I do not distinguish between types of elicitation procedures here. From this perspective, an informal interview format is simply likely to result in a greater proportion of missing data than a more structured format. 2. Note that here I adopt a “production system” approach to mental representation or cognitive architecture, which has a long and distinguished career in artificial intelligence (Newell 1990). 3. Alternatively, this result might also be interpreted to show that there is relatively little influence of informant experience at being interviewed (in contrast to interviewer experience). 62

IS REFLEXIVITY NECESSARY? 4. To some degree, recall, reasoning, and memory can all be influenced by the vagaries of circumstance as well. So it is never possible to tell for sure what someone believes. Alternatively, one might say that what someone “believes” is always accurately reflected in what they report at any given moment, and that this belief may change rapidly to reflect changing circumstances. This is a question best left to philosophers and semioticians. What the present approach run tell you is that not all variation in reported belief in a group is purely methodological in origin.

63

4

THE WRONG WAY OUT: TYPOLOGY AND IDEALISM

Culture as it is ordinarily constructed by the anthropologist is a more or less mechanical sum of the more striking or picturesque generalized patterns of behavior which he [sic] has either abstracted for himself out of the sum total of his observations or has had abstracted for him by his informants in verbal communication. .. .Cultures, as ordinarily dealt with, are merely abstracted configurations of idea and action patterns, which have endlessly different meanings for the various individuals in the group. -Edward Sapir, “The Emergence of the Concept of personality in a Study of Cultures”

T

he primary goal of ethnography has typically been to present a coherent picture of life in cultural groups (Geertz 1988). Th‘is requires a process of representation. Historically, ethnographers have created an idealization of that life by constructing a uniform picture from selected field notes and other data. Two tactics are used to create these depictions: Culture is made into an abstract structure external to the relevant population of individuals (e.g., Durkheim’s %ocialfacts” ); or Culture is personified in the form of a single hypothetical, idealized individual (e.g., the “culture and personality” school, ethnoscience). The first representation tends to serve as an isolated causal model of the social system (i.e., how that society works). The second form of ideal, em64

THE WRONG WAY OUT: TYPOLOGY AND IDEALISM

bodied in the “ideal informant” concept, provides a normative description of life in the cultural group (i.e., how people should behave). Max Weber ([1949] 1971), who is particularly clear about the process of idealization, calls either of these forms “ideal types.” They are constructed through the analytic accentuation of certain elements synthesized from many exemplars to make the type’s characteristic features clear and understandable. He emphasizes that this is a purely logical operation to be distinguished from aesthetic or moral appreciation of the construct as a model of what ought to exist. “It is a matter here of constructing relationships which our imagination accepts as plausibly motivated and hence as ‘objectively possible’ and which appear as adequate from the nomological standpoint” (Weber 1971,498). For Weber, it is important to distinguish between two very different claims about such idealizations,whether in the form of a normative model or an analytic construct: does the idealization exist in the mind of the researcher, or in the minds of the people being studied? As ideas,each of these ideals has a distinct existence in time and space. In the form of a norm, the ideal can have a cognitive reality, and guide the ethnographer’s imagination during writing. “Such presentations are of great value for research and of high systematic value for expository purposes when they are used as conceptual instruments for comparison with and the measurement of reality” (Weber 1971,503).As an analytic construct, a claim is not being made that each and every individual in the study population had such a conception in his or her mind. Rather, the formulation is forced on the researcher, because those “ideas” which govern the behavior of the population of a certain epoch i.e., which are concretely influential in determining their conduct, can, if a somewhat complicated construct is involved, be formulated precisely only in the form of an ideal type, since empirically it exists in the minds of an indefinite and constantly changing mass of individuals and assumes in their minds the most multifarious nuances of form and content, clarity and meaning. (Weber 1971,501)

Thus, in either form, an idealization is defended as a particularly efficient and persuasive means of representation-hence the popularity of the idealistic strategy in the social sciences. At least until recently, idealism has held sway in ethnography as well (Vayda 1994). It appears a good avenue to pursue: abstract constructions 65

CHAPTER 4

can be achieved rigorously; they can even be replicated by independent ethnographers. The ethnographic construction is based on careful consideration of variation within groups. AU of the relevant scientific criteria appear to be in place. However, as I suggested in chapter 2, idealistic methods are not particularly adept at being reflexive. Nevertheless, perhaps we can legitimately make a detour around this requirement and achieve our end goal of a highly appreciated and valued ethnographic project by using rigorous idealistic methods. Scientific methods, after all, are the sine qua non in many circles, and a scientific idealism, since it is already available and being practiced to some acclaim, should be seriously investigated as a route to restoring ethnography to the hallowed ranks of authentic academic endeavors. Perhaps we have here a shortcut to what everyone agrees is desirable: the restoration of ethnographic weightiness. Scientific idealism, after all, combines age-old typological representations with modern-day quantification. Is it possible we can simultaneously attain the best of both worlds? In this chapter, I seek to show that what appears as too good to be true is in fact too good to be true. Scientific idealism as a strategy is fundamentally flawed. This is not a purely philosophical or academic matter, since both the means and goals of ethnography are influenced by the culture concept adopted in practice, whether implicitly or not. In the following, I first delineate a number of conceptual and methodological problems associated with idealism as an approach to ethnographic description. However, since purely theoretical arguments are often unconvincingespecially to those with different meta-theoretical commitments-I also pursue an empirical line of investigation. In particular, I use a recent approach with a set of idealistic core assumptions, cultural consensus analysis (CCA; Romney, Weller, and Batchelder 1986), to examine whether the central conceptual goal of idealism can in fact be achieved: finding the one true ethnographic depiction of a cultural group from among the conflicting reports and perspectives that constitute ethnographic data.

Cultural ConsensusAnalysis Cultural consensus analysis has proved popular. Since the method was first published a little more than fifteen years ago, over twenty-five independent empirical studies using CCA have been published, and the original article defining the approach has been cited over one hundred times 66

THE WRONG WAY OUT TYPOLOGY AND IDEALISM

(Romney, personal communication). As perhaps the currently dominant approach taken by those with a scientific bent in the cultural anthropological community (itself admittedly a minority), CCA is worth investigating further. Why is CCA so popular? Arguably, it is because it addresses a significant problem: ethnographers, although typically untutored in the truth of the local culture under investigation, desire to properly characterize the beliefs and values of the group they study (Batchelder and Romney 1989, 229). CCA appears to reliably address this problem of ethnographic ignorance because the method formally infers the underlying cultural consensus from the pattern of informant responses to a set of questions designed to probe knowledge in a particular domain. Cultural consensus analysis gives ethnographers the ability to simultaneously ascertain: (1)whether an unknown belief system constitutes a “high concordance code” (Romney,Weller, and Batchelder 1986,316); (2) what the consensual beliefs in a topical domain are; and (3) estimates of individual informants’ competence in this cultural belief system. Further, it performs these inferences based only on standard ethnographic data: responses by a small sample of informants to a suite of questions designed to tap knowledge in some particular topical domain. Formally, the method factors a matrix that measures the cultural similarity of all possible pairings of sampled individuals in the group. If the factor-analytic output exhibits several characteristics, the domain can properly be considered consensual. Under these conditions, the response to a question most likely to be culturally correct becomes that which gets the highest score when the frequency of each alternative response is weighted by the average competence of those who made that response (individual competence also being estimated by the method). These inferred beliefs can be reasonably applied to the group as a whole, with observed variation considered as methodological or idiosyncratic in origin (Romney et al. 1996,4704). CCA has a number of virtues that make it a best case for our examination of idealism as a strategy for achieving reliable ethnographic representations: as a formal model, it permits testing of its claims; its assumptions are made explicit; and it has been validated using simulated data against modest violations of its underlying assumptions (Romney, Batchelder, and Weller 1987, 164). Perhaps most important, CCA’s output can be interpreted as representative of the two main means through which cultural belief systems are idealized: either as an abstract whole (i.e., 67

CHAPTER 4

the correct set of beliefs held by no individual mind, like Durkheim’s “SOcial facts”), or personalized in the form of an “ideal informant” capable of M y appropriate behavior in every situation (e.g., Goodenough‘s [1965] perfectly competent, but nonexistent embodiment of cultural knowledge). However, to my knowledge, the central, idealistic claim of CCA-to provide a robust prediction of cultural knowledge-has not been tested using ethnographic data. I therefore report quasi-experimental tests to test CCA’s ability to reliably determine consensual beliefs, again using the database concerning the edibility of foods from the Ituri Forest (see appendix C for details). In the course of this investigation, I adopt a tactic that does not conform to standard CCA practice (because this is a methodological, not substantive, investigation), but that shows the fragility of the conclusions that can be derived from the method. In essence, I use a variety of criteria to select subsamples from the available data. Such manipulations represent reasonably natural variation in the database that might be obtained-given the vagaries of ethnographic fieldwork-and hence are legitimate tests of the method’s ability to produce the same consensus consistently for a cultural group. To select data subsamples, I first make assumptions about the nature of the data available for testing (e.g., I change sample composition by removing all the youngest informants). Second, I simulate other decisions required to formulate an empirical test (e.g., by selecting particular sets of questions that might be more representative of domain knowledge, or by changing the coding of responses). After each such manipulation, I then perform the standard consensus analysis on the resulting data to see if substantial disagreements in the characterization of the belief system result. In each case, CCA produces significantly different sets of beliefs that are to be considered consensual. Thus, this chapter will show that CCA exhibits significant operational and conceptual problems. I will argue these problems derive from the method’s implicit idealism. I further argue that the procedures I used to design my field study and to prepare the resulting data for analysis using CCA are not outside the bounds of common practice in CCA-based studies-the procedures involved introduce fairly natural variation into the data available for analysis. I therefore contend that the problems uncovered in the course of this investigation are likely to characterize any CCA study that undertakes similar testing of their data. And since the empirical data concern a classic belief system 68

THE WRONG WAY OUT TYPOLOGY AND IDEALISM

(food taboos) in a classic setting (a traditional population in central Africa), the results shown here constitute a considerable warning to those who would use CCA to ascertain when and how to characterize groups using any single set of cultural traits. Since CCA constitutes the best available method for finding consensus, the same conclusion would likely be true of any other idealistic method. I therefore believe that the ambition to characterize cultural groups in such a typological or idealistic fashion should generally be abandoned by cultural anthropologists. Let me now present the basic findings from this inquiry (see appendix C for a complete report of the empirical analyses conducted to support this summary).

Variations in Consensual Belief

It is likely that samples of differing sizes of informant belief reports will arise in ethnographic fieldwork. Does such variation influence the consensus found, the cultural beliefs uncovered by CCA?The answer is yes (see appendix C, “Typologizing Groups” section, for substantiation of this claim). For example, randomly selecting informant samples of different sizes from one Ituri ethnic group produces a degree of variation in the consensual response set that approaches in significance the variation produced by running CCA on samples from different Ituri Forest ethnic groups. Indeed, it is fairly easy to construct a subsample that does not exhibit the characteristics necessary for food avoidances in the Ituri to be called a consensual belief system at all, even though all of the other analyses conducted indicated that it does constitute such a system. It is obviously a problem of the most fundamental kind if uncertainty extends to the basic determination of the analysis: whether a consensus exists at all. It is traditional practice to use self-identification as the primary consideration when determining the relevant group on which to conduct analysis. Thus, an ethnography will typically concern an ethnic group of some kind. But taking the informants’ word about their ethnic identity cannot be used as a secure way of finding the group within which consensus is most likely to be reached. At least in the Ituri, a randomly selected multiethnic sample exhibits higher measures of consensus than several of the ethnic groups considered independently. That is, there is greater sharing of belief in a sample composed of individuals taken from 69

CHAPTER 4

the population as a whole, across ethnic group boundaries, than when individuals are selected from one ethnic group. Which criterion should be accepted when hunting for an ethnographic population, for the individuals who belong together: the people’s own sense of kinship or the output from a statistical analysis? Consensus on important cultural beliefs such as food taboos, as measured by CCA, doesn’t seem to be the measure by which people organize themselves into groups. Similarly, using CCA in a study of Welsh towns, Carol Trosset and Doug Caulkins (2001) find that the individuals closest to the cultural ideal of “Welshness” are actually found to derive from multiple ethnic groups. Many individuals identify themselves as Welsh, but do not personally adopt the beliefs and values found by CCA to be characteristic of that group, while others in the same area who are not Welsh hold typically Welsh values. Penn Handwerker (2002) also finds that ethnicity does not correlate well with the dimensions of cultural variation in his analyses of data on parent-teacher working relationships in Puerto Ricans and Connecticut Yankees; instead culturally similar individuals cut across these communities. Should ethnographers be using ethnicity as their gold standard by which to judge shared culture? Handwerker (2002,118) argues that, by looking for similarity within ethnicities we “impose cultural differences by assumption, not evidence. To avoid this error, ask who shares what with whom; look for similarities and differences among your informants.” Perhaps Fredrik Barth (1969) was right when he famously argued that ethnicity is more about establishing borders between Us and Them than determining shared cultural content within the group. Ethnographers typically don’t interview children-because they are immature, still learning their cultural values-or older people, who may be forgetful or showing the onset of senility. In the Ituri, I interviewed both categories of informants. However, if you eliminate either of these age categories from consideration, then the projected consensus changes substantially. Again, traditional ethnographic practices, which are typically extended to consensus analysis, can falsely characterize a group. The focus on adults in the prime of life as experts in their culture is strong, but ultimately self-defeating if ethnographers want to characterize the population as a whole, or to get a sense of the dynamic means by which culture changes from generation to generation. 70

THE WRONG WAY OUT TYPOLOGY AND IDEALISM

Less fundamental, perhaps, is a troubling technical problem that often arises in ethnographic research: we cannot go straight from what an informant says to statistical analysis in every case. Sometimes we have to categorize responses, or turn “I can’t eat squirrel when I’m pregnant” into a code, for example, that individual has taboo type l? What happens if we adopt alternative methods for coding the same verbal responses from informants? It turns out that if we use a more specific coding scheme, with more dividers between categories, then we find that the suite of taboos declared consensual can change dramatically. We thus see that just about any sort of change in the basic data winds up changing what we conclude about the culture of the group under study in significant and unpredictable ways. We can’t prejudge what the likely influence of a particular foible of the available data w ill be on the consensus that is found by the statistical operations underlying CCA. The process is not transparent. This can be seen as a virtue: we submit our inputs, the computer grinds up the data and spits out a solution. The process is not influenced by our predilections or expectations; it is objective. However, it also means that we have outputs that we cannot rely upon because they are arrived at through a black-box procedure that cannot be second-guessed. Roy D’Andrade (1987, 194) has argued that it is legitimate to summarize cultural variation as a modal tendency, despite the disadvantage “that it takes a purely statistical characteristic and treats it as if it had some kind of reality,” because most domains exhibit a bimodal distribution: a particular belief tends either to be held by the great majority of members of a group or to be exhibited by a small fraction of individuals. So “one feels more justified in saying that some item is a part of a culture if it is held by 97% of the population rather than 51%. . . . [However] it would be much more satisfjring theoretically if there were something about the more highly shared items that indicated that they have some special kinds of psychological or social characteristicssomething that marked off these items from the less highly shared items beside frequency” (D’Andrade 1987, 194-95). Unfortunately, it seems that CCA cannot be relied upon to reliably provide the set of traits with the desired qualities that mark them as truly cultural. The set of consensus responses responds too flexibly to changes in data collection practices, and so is an idiosyncratic, rather than robust, characterization of a group. 71

CHAPTER 4

Ideal Informants

CCA can be used to construct both types of representation of a cultural ideal identified at the beginning of this chapter. Having found a number of problems with the measurement of consensus, the question remains whether CCA is also troubled by difficulties surrounding its implementation of the notion of competence and an ideal informant. In this section, I ask two questions about CCA’s estimates of individual competence in the consensual belief system. First: How stable are the measures of competence? The answer is similar to that for consensual responses: an individual’s cultural competence can vary significantly, depending on with whom they are grouped. In effect, the degree to which their set of responses to the interview questions approaches that considered representing the group is a function of group composition. Of course, this is sensible. The problem is that it is not clear which group provides the most relevant comparison. Remember that ethnic groups do not necessarily exhibit the highest consensus measures, so which group the individual should be compared to is not clear, given that the natural choice, people’s self-identification, is not available. Some criterion external to CCA itself must be used to provide a theoretical rationale for the selection of a particular group. But this will likely be specific to each study, and the type of question it attempts to address. Putting back into the hands of the ethnographer the choice of who to count and when returns us to old, subjective principles of ethnographic reporting. Second: Are informants whose knowledge base most closely approximates that of the group consensus the “expert” informants sought after by ethnographers since time immemorial? D’Andrade (1987) argues that the work of cultural consensus theorists, such as Boster and Weller, has shown that “good people give good answers,” assuming, as do cultural consensus theorists, that “good” answers are modal answers. The individuals estimated to have high competence also have the desirable characteristics that make it appropriate to call them “ideal informants” or “experts.”The good qualities of such informants are as follows: (1) Reliability: Persons who are more likely to give modal responses on a task are more likely to be reliable, that is, to give the same re-

72

THE WRONG WAY OUT: TYPOLOGY AND IDEALISM sponses when the task is presented at a later time [I have previously discussed the value of this criterion and found it wanting]. (2) Consistency: Persons who are more likely to give modal responses on some task are more likely to give responses that are consistent with each other [that is to give logically consistent responses that exhibit the transitive property: A > B, B > C, A > C]. (3) Normality: Persons who are more likely to give modal responses on some task are more likely to have had normal experiences with respect to the material symbolized by the task. (4) Education, intelligence, and experience: Persons who are more likely to give modal responses on a task are more likely to be better educated with respect to that task, are judged more intelligent with respect to their ability in that task domain, and tend to have more experience in that task domain. (D’Andrade 1987,196)

Thus, the contention of CCA researchers is that high competence individuals are the best informants because they exhibit a variety of other desirable qualities-l But is it true in the Ituri that the ideal (high competence) is correlated with these other kinds of expertise or measures of capability? The answer is no. Based on a variety of methods, I find that culturally competent individuals in the Ituri are not those with higher intelligence, greater schooling, culturally designated roles as keepers of traditional knowledge, normality of experience, or central roles in social networkseach of these characteristics having been found to characterize high-competence informants in at least one previous CCA study (see appendix C “Cultural Competence” section).The correlation between measured competence and these reasonable measures of expertise has been claimed to validate CCA (Romney 1994,270), but this validation is not appropriate in the Ituri example. Thus, just as was the case with the other measure of consensus-consensual responses-significant and worrying variation is uncovered about the sources of those responses, the informants themselves, by simple, naturalistic manipulations of the ethnographic database. Garro (2000,302-3), in her discussion of CCA, asks: “ISthe variability [in cultural groups] to be attributed to differences in how much individuals know about the domain, with the more ‘competent’ individuals

73

CHAPTER 4

knowing more about the cultural knowledge pool . . . [or] can variation be linked to social and personal experiences that contribute to differential patterns of reference to aspects of these widely shared [i.e., consensual] explanatory frameworks?” Handwerker (2002, 109) puts the same point more forcefully: “Only individuals learn, and individuals embody and constitute the only source of cultural data. . . . For a specific person, culture, because it is constructed in that individual’s mind out of the unique set and sequence of experiences that mark the trajectory of the person’s life, embodies who that person is as an individual.” In sum, “everyone constitutes a cultural expert in what she or he knows, feels, and does” (Handwerker 2002, 111).Together with Garro and Handwerker, I argue that we should interpret cultural variation as being due to real causal influences on individuals, not just their lack of ideal qualities. We should be thinking of individuals not as deviants from some nonexistent standard of value, but as having had rich lives that make them unique.

Cultural Systems # Consensual Systems Indeed, if the scope of argument is broadened to comparisons between other studies in the literature, similar, confirming problems with CCA come to light. In particular, there are cases involving systems that, by almost any definition, would be called cultural but that do not exhibit the necessary degree of unanimity in belief to warrant a consensus. Alternatively, there can also be a relatively high degree of measured consensus in a cultural group despite the lack of a cultural system for the determination of those beliefs. I present examples of each possibility in turn. Two cultural consensus studies (Weller 1984; Boster and Weller 1990) concerning humoral beliefs (i.e., values such as “hot” or “cold” applied to foods) qualify as examples of the first possibility. It is very unlikely that individuals constituting a large fraction of a cultural group would independently invent a system of valences for foods from their own personal experience. Cultural transmission between individuals is almost certainly the primary means by which humoral beliefs are replicated, since they derive from Hippocrates’ notion of bodily “humors,”which then diffused to particular cultural groups (Mathews 1983, 827). However, case studies consistently show that there is an insufficient degree of consensus in hu74

THE WRONG WAY OUT: TYPOLOGY AND IDEALISM

moral beliefs to call them cultural. In such cases, cultural consensus theorists are left in a quandary as to what such beliefs should be called (e.g., Weller 1984). O n the other hand, it is also possible for individuals in a cultural group to show a significant degree of uniformity in response to interview questions, despite the lack of a socially defined and culturally transmitted system of belief about that domain of experience. For example, Boster and Weller (1990) found a degree of consistency between American informants in the assignment of valences to foods equal to the consistency in a group with a culturally explicit humoral system. Without the knowledgederived from being members of that cultural group themselvesthat no cultural system for valence assignment exists in America, the authors would have had to argue that a humoral system exists in American culture. As Boster and Weller (1990, 178) suggest, the American group appears to have consistently applied shared cultural knowledge concerning food spiciness and the temperature of prepared foods to the unfamiliar valence assignment tasks. That is, the American informants applied cultural models from related domains of experience to the artificial task Boster and Weller asked them to solve. Similarly, Romney, Weller, and Batchelder (1986, 331) suggest that mean competence should also serve as a criterion of consensus, since some studies show a low competence despite a strong indication of consensus (in the form of a statistical measure called an eigenvalue ratio-see appendix C for details). For example, they found a relatively high eigenvalue ratio in their study of a “general information task” among American undergraduate students, but a low informant-by-informant average correlation, and are therefore hesitant to call it a cultural pattern, presumably due to a feeling that “general information” is not a natural category or domain of knowledge (Romney, Weller, and Batchelder 1986, 332).2 It therefore appears that significant inter-informant agreement is a poor indicator of the existence of cultural systems of belief. However, a primary use of cultural consensus modeling in an anthropological context is to infer that a belief system is in fact cultural when no a priori expectations are available (Batchelder and Romney 1989,229). Given the unreliability of consensus in a sample of informants as an indicator of socially transmitted beliefs in a domain, CCA seems unsuited to revealing the existence of a previously unknown cultural belief system, or confirming 75

CHAPTER 4

the existence of a known system. Thus consensus in beliefs cannot be used to describe the strength of a cultural system, much less determine whether such a system exists at all. The other primary measure derived from CCA-competence-also cannot be used to determine whether a belief system is consensual. D’Andrade (1995,214) points out that cultural conformity effects could be responsible for the correlation between competence and expertise: by definition, those who unquestioningly adhere to cultural standards are more likely to give the normative response, and to give it consistently and without hesitation. Nevertheless, D’Andrade (1989a, 122) argues that because “one can be an expert in a cultural system implies that such cultural systems are indeed real.” However, individuals can show “expertise”(i.e., exhibit a higher percentage of consensual responses than other individuals, while also being more normal, intelligent, etc.) in noncultural systems as well, such as word association tests “even when there are no right or wrong answers, and no pressure of conformity, and not even an obvious kind of knowledge involved, the same pattern [of association between competence and other desirable individual qualities] is found” (D’Andrade 1995,215). Due to this inability to distinguish between the significantly different consensus sets and competence values that CCA produces, neither competence values nor especially the “answer keys” to cultural knowledge produced by CCA can be trusted. Thus, even if the results from the Ituri study (discussed above) are ignored, I believe published CCA studiesconducted by researchers trained in the method and that have passed peer review-make the same points: there can be significant intra-cultural variation underlying an apparent consensus (Johnson and Griffth 1996); a belief system can exhibit high consensus (Boster and Weller 1990) and people can be “expert”in such knowledge (D’Andrade 1995), even though it is not a.cultural system, and vice versa (Weller 1984). This inability to discriminate between consensual and nonconsensual systems is significant because CCA researchers argue the primary objective of the method is to determine when it is legitimate to aggregate to the “majority view” (Borgatti 1994,275; cf. Weller et al. 1993,115). They consider this move legitimate when there is consensus because cultural systems are defined as commonly shared beliefs (Romney, Weller, and Batchelder 1986,316) or shared cognitive representations (Romney et al. 1996). Borgatti (1994,276) believes consensus theory is an important the76

THE WRONG WAY OUT: TYPOLOGY AND IDEALISM

oretical development because it provides clear criteria for when and how to aggregate data, “which is the fundamental operation of analysis.” However, even when no consensus is found, it is argued, CCA remains useful. For example, Doug Caulkins and Susan Hyatt (1999), Garro (2000), Jeffrey Johnson and David Griffith (1996), Juliet McMullin and colleagues (1996), and Susan Weller et al. (1993) believe CCA can be used to investigate intra-cultural variation. But when used for this purpose, individuals (via their competences) are compared to an idealized representation of the group, not each other (unless the agreement matrix itself is analyzed, but this is not typical practice). The argument is that individuals deviate from this statistical construct in ways that can be analyzed to provide new insights into the sources of such deviation. A. Kimball Romney (1994,269,273) goes so far as to claim that “clearly detailed studies of intracultural variability depend upon estimates of cultural competence” and that “the application of the model makes possible a far deeper understanding of individual differences in cultural knowledge than heretofore.” But I have just argued that neither consensus nor competence can reliably distinguish between what CCA researchers themselves would call cultural and noncultural belief systems. So the point at which aggregation to specific group-level values becomes legitimate is under-determined. The set of values to which individuals should be compared is therefore also in question, and the project of characterizing deviance jeopardized. As Garro (2000,312) concludes: Uculturalconsensus analysis cannot take us very far in understanding either intracultural variability or how cultural knowledge is interrelated at a cognitive level.’’ Further, the method itself makes no direct reference to any theory of consensual belief; this pure instrumentality leaves CCA without the kinds of a priori expectations that would permit a researcher to choose the correct characterization of variation or consensus from among the method’s outputs. O f course, theoretical concerns and background knowledge acquired through field experience play important roles in constraining the analytical strategies employed by any ethnographer. Certainly, the studies of humoral belief cited above showed that use of CCA does not necessarily clarify whether people in a particular group attach social value to choices in a domain of knowledge. So, in fact, researchers must have independent knowledge of whether the system they study is consensual (that is, cultural) befOre they conduct their analysis. Even given the perception that a consensus should be expected, I have argued that the choice 77

CHAPTER 4

among possible representations of that consensus is uncertain. I therefore submit that the standard use of CCA is not a valuable addition to the panoply of techniques for the description of cultural groups. In a similar spirit, Handwerker (2002) suggests we abandon CCA altogether and use statistical models (such as factor analysis and multidimensional scaling) designed to investigate individual-level variation directly, rather than attempting to compare individuals to the “right answers” in some kind of academic test framework like CCA.3

The Failure of Idealism This chapter has investigated the ability of idealistic methods (particularly cultural consensus analysis) to reliably characterize both the consensual beliefs of cultural groups and the competence of individual informants in a belief system. The data on which sensitivity tests of CCA were performed consist of standard responses to formal interview questions (concerning cultural constraints on the edibility of foods) in an ethnographic context. However, test results were disappointing. Manipulating the sample of data available for analysis produced significant variation in both consensual beliefs and individual competence. Measures of consensus are a function of the available sample of informants, the sample of questions, and the scheme used to code responses. The most fundamental question addressed by the cultural consensus method is whether consensus exists in a given belief system. A secondary goal, if a consensus does exist, is to specify that consensus through the (not necessarily modal) set of beliefs characteristic of the group. Therefore, perhaps the most unexpected and debilitating evidence against the approach would be tests showing both consensus and a lack of consensus with respect to the same belief system in the same population. This is just what happened in the Ituri case. CCA cannot be said to reliably determine consensuality, since the same sample of informants (the largest of any published CCA study) can show both consensus and nonconsensus from the same suite of questions, depending on the type of coding scheme (binary or multiple choice) and concomitant estimation technique used. Further, the significant variation seen here in consensual beliefs for the Ituri ethnic groups attests to the fact that the secondary goal of reliably characterizing a cultural group by some particular suite of responses is also unrealized: results are not stable, at least not in this case. Additional con78

THE WRONG WAY OUT: TYPOLOGY AND IDEALISM

tradictions were seen between tests on the same multiethnic sample, with CCA suggesting there was considerable consensus at the regional level, while analysis of variance indicated that variation in belief between ethnic groups is in fact significant. These conclusions are mirrored by Johnson and Griffith‘s (1996) recent study, in which they performed a variety of analyses similar to those conducted here on a very different cultural system and population: beliefs about the relationships between coastal pollution and the safety of seafood among Americans. They show that when this cultural group is broken down by ethnicity, income, and residential area, consensus sets vary meaningfully (using an equivalent analysis to that conducted here). They also demonstrate that there can be relatively high overall consensus within a large sample that masks a lack of consensus in a number of these subgroups. Thus, when similar analyses are conducted on other domains and populations, similar difficulties appear. I therefore conclude that the Ituri case cannot be seen as parochial or idiosyncratic: any CCA study that performs similar sensitivity analysis will likely encounter similar difficulties to those found in the course of this study. There is no norm to share food taboos in Ituri society. Rather, individuals are supposed to acquire a comprehensible set of taboos from their parents and then put them into practice, whatever constraints these taboos may impose. This transmission norm determines from whom a belief should be learned, not its content. The lack of social restriction on content produces the significant intra-cultural variation I have previously documented in this belief system (Aunger 1994a, b, 1996, unpublished data). It is therefore confusing to find that CCA suggests there is a significant degree of cultural consensus in the Ituri. Indeed, the Ituri study produced some of the highest measures of consensus in the social science literature. Of course, it is true that scientific concepts need not be identical with “folk” notions-the contrast between everyday or Newtonian physics and quantum mechanics being an obvious example. However, the set of characteristics by which Romney, Weller, and Batchelder (1986) define consensual beliefs are those necessitated by the mathematics of the method itself, and do not derive from an independent consideration of what cultural consensus means. Because CCA-based measures of consensus are not theoretically derived, they cannot claim to override ethnographic impressions nor the commonsensical idea that if there is significant variation within families, then intra-cultural variation must be real. I therefore con79

CHAPTER 4

clude that CCA’s general assessment that there is consensus in the system of Ituri beliefs concerning food gives one a mistaken impression about this ethnographic system. Further, in no case did average competence measures descend into the unacceptable range. According to CCA, informants were always competent, on average-but about significantly different belief systems-and regardless of whether the system was consensual (according to other CCA criteria). This makes a mockery of the notion of cultural competence, the other primary point of information provided by CCA. Nevertheless, it remains the case that standard tests have produced standard results here: high measures of consensus and competence, correlations between competence and other desirable qualities (e.g., education). It has only been the interpretation of these results, and more detailed examinations of their cultural context, that has reversed the conclusions typically derived from the tests. Tests not typically conducted have added contradictory or confusing complications to the general picture. In essence, despite the formality of the procedure itself, cultural consensus analysis remains an art form because the method’s output varies significantly with changes in the questions selected to represent domain knowledge, and the set of informants selected to represent the cultural group. As such, it does not appear to represent an improvement over standard interpretive ethnography in producing authoritative characterizations of cultural groups. The real problem lies, I believe, with this goal itself. (There is nothing formally wrong with CCA’s methods.) The enterprise of definitively labeling cultural groups, which underlies CCA, is misguided in my view. Romney (1999) counters that statistical models like CCA cannot have philosophical implications, because they are merely instrumental devices bent to any need. But others disagree. For example, Andrew Abbott (1988) maintains that the General Linear Model-perhaps the most popular, traditional approach to multivariate statistical analysis-influences researchers’ ontology as well as the causal model of the world in fundamental ways, primarily due to its division of phenomena into characteristics or variables rather than whole events. For example, gender can be objectified independently of its instantiation in individuals; it is then no longer a quality of someone but a thing in itself. In particular, the use of typical statistical procedures like the generalized linear model leads to a viewpoint that Abbott (1988) calls “general 80

THE WRONG WAY OUT: TYPOLOGY AND IDEALISM

linear reality,” characterized by a number of largely unrecognized but unrealistic assumptions (Abbott 1992,433-34): 1. The social world is made up of futed entities with varying attributes (demographic assumption). . . . 2. What happens to one case [e.g., individual] doesn’t constrain what happens to others, temporally or spatially (case wise independence assumption). 3. Attributes have one and only one causal meaning within a given study (univocal meaning assumption). 4. Attributes determine each other principally as independent scales rather than as constellations of attributes . . . (main effects assumption). . . . 5. Things happen in discrete bits [of time] of uniform length and are not aggregated into overlapping “events”of varying length (continuity or uniform time-horizon assumption). .. . 6. The order in which attributes change does not influence what changes occur; all cases follow the same “causal narrative” or model (non-narrative assumption).

In essence, general linear reality assumes that individual cases can be disaggregated into sets of attributes, each of which can be treated independently. These attribute sets are placed in an abstract, artificial alignment with one another, move in lockstep through a “space”without time or distance, are subjected to similar processes of alteration, and react to these processes in exactly the same way. Since cases with particular combinations of attributes do not exist in this conceptual space, the units of analysis cannot transform into qualitatively different things during the analytical process, nor are any units added or subtracted from the model. Outcomes cannot be sequence-dependent, because there are no temporal steps in the analysis: a case’s value after having jumped to the next analytical time frame is completely determined by its previous state. Similarly, I believe it is reasonable to characterize CCA as having idealistic implications, and that the pattern of adoption of CCA by researchers in different disciplines is conspicuous evidence of its idealistic nature. For example, those studying complex, stratified, highly heterogeneous societies (e.g., cultural sociologists) have no interest in typologizing their subjects, and do not use CCA. They simply survey the relevant population and investigate what variability they find within it for clues as to 81

CHAPTER 4

the causes of that variation. The standard cultural anthropological goal of characterizing small, relatively homogeneous groups by interviewing a very small number of informants seems pointless from their perspective (and I would argue that even traditional anthropological populations deserve social survey treatment: identification of the distribution of cultural traits by the various dimensions that distinguish subgroups). One might argue that awareness of CCAls availability hasn’t yet leaked across disciplinary borders; however, its use by psychologists (e.g., Alvarado 1996, 1997), medical doctors (Moore et al. 1997; Chavez et al. 1995), and linguists (e.g., Jameson and Romney 1990) also interested in uncovering human universals suggests that absence of evidence is evidence of absence in this case-after all, these are not even fellow social scientists. Thus, those with idealistic ambitions use CCA, while those without, don’t. Even prominent social anthropologists argue that few would any longer subscribe to the idea . . . that cultures exist “out there” as things that human beings might be said to live in: each one a neatly bounded, perfectly shared, and historically cumulative body of acquired tradition. It would be more in accord with present anthropological thinking to say that people live culturally rather than living in cultures, and that cultural life consists in a medley of multiple voices, as often discordant as in unison, characterized by an imperfect sharing of knowledge, founded as much on misunderstanding as on common understanding, and perpetuated through an improvisatory, often playfd and always creative, dissembling of convention. (Ingold 1993,526) In the end, there is no need for CCA except to characterize consensus. If one does not intend to use the consensual representation either as a marker for the group as a whole or as a point against which individuals are compared, then other, more standard statistical techniques are available for investigating intra-cultural variation directly. However, as we have seen, CCA can’t be reliably used to find consensus. Thus, finding significant variation in consensus is not a purely methodological issue-to be corrected by context-sensitive use of the method and by making the right choices in research design and implementation-but rather annuls the utility of cultural consensus analysis. Significant problems of reliability and interpretation have been uncovered in this investigation of CCA. The cultural consensus literature itself further corroborated the conclusion (based on the Ituri study) that 82

THE WRONG WAY OUT: TYPOLOGY AND IDEALISM

CCA does not provide sufficient internal clues to dependably determine consensual beliefs. Since this represents a failure of the best available method for depicting life in cultural groups using idealized constructs, I believe the problems highlighted in these investigations are significant. In fact, I posit that the road to idealized ethnographic representation is necessarily fdled with both practical and theoretical potholes. Since the primary goal of most ethnography has traditionally been to depict life in cultural groups using idealized constructs, I believe ethnography must find another ambition.

Implications for the Notion of Culture There are further reasons not to adopt an idealistic approach. These have to do with the implications of such an approach for the central notion in anthropology: culture. What is culture if you adopt idealism as your goal? From a philosophical point of view, the most significant feature of current definitions of culture is the fact that they presuppose either a realistic or an idealistic approach. . . . [Idealists] regard culture as a heritage of ideas that have a transcendent reality independent of the individuals or societies which happen to bear them. . . . [Realists, on the other hand, are those who] conceive of culture as an attribute of human social behavior and usually define culture in terms of acquired habits, customs, and institutions. (Bidney 1967,23-26)

Thus, idealism is both historically and naturally linked to the idea of culture as shared knowledge, while realism has been affiliated with a conception of culture as socially learned information. Many of the conceptual problems with an idealistic approach to culture are reflected in, and perhaps spawned by, its definition of culture. Definitions of culture as shared knowledge have been around since the beginning of anthropology (see the reviews by Kroeber and Kluckhohn 1952 or Keesing 1974). As Robert Borofsky (1994,331) notes, most current introductory textbooks continue to emphasize the shared nature of culture. Further, since cognitivism gradually conquered behaviorism in the 1960s, culture is now generally viewed as “in the head” (DAndrade 1995). This tends to exclude artifacts and other material manifestations of cultural knowledge from the culture concept. Thus cognitivism, together with the shared knowledge constraint, makes culture equivalent to widely shared beliefs and values. The notion 83

CHAPTER 4

of culture becomes further refined in ethnographic practice when only the coherent and consistent aspects of cultural knowledge are emphasized (in line with idealistic objectives). In effect, culture becomes consensual knowledge. One contemporary stream of culture-as-consensus theory, processual in nature, is interested in why some beliefs and attitudes come to be widespread. A representative is Sperber (1991, 1996), who argues that culture theory should explain why only some of the beliefs invented by human imagination become widely distributed. He postulates that widely shared beliefs are those that are easily communicated and minimally transformed in the process-hence allowing many copies of those traits to be current in a population. A more restrictive version is that only widespread normative beliefs are cultural. D’Andrade, for example, believes that what makes knowledge cultural is not just whether it is shared, but whether ignorance or inappropriate use of that knowledge is sanctioned by others in the group: “What makes something a cultural model is the use of it, not how it is learned” (198913,824). However, a definition of culture as shared, whether normative or not, has a number of conceptual problems. First is the problem of origins. Any newly minted belief necessarily begins at low frequency in a populationa frequency of one. Some beliefs become popular, so today’s marginal belief is tomorrow’s cultural convention. Do beliefs become more “cultural” as they increase in prevalence from necessarily humble beginnings? This reliance on degree of prevalence to define culture also spawns operational difficulties. For example, it suggests that the nature of beliefs changes once they reach a critical degree of prevalence. Thus, it is wrong to argue that “at what point in the continuum of sharedness we decide to call a given schema [or model] ‘cultural’ is simply a matter of taste’’ (Strauss and Quinn 1994,293). What is this point-51%, 75%’ or 90% of the population? And what is the nature of this change? For those who adopt the normative view of cultural consensus, this crucial point is when belief becomes sanctioned. But problems may arise if the criteria of normativeness and popularity conflict. Some beliefs are acquired because they are rare (a bias toward beliefs of relatively low frequency in the group). Is such an “anti-norm” also a norm, even if it never leads to high sharedness?What if, even after a belief has become popular, most individuals acquire the belief simply on its merits rather than be84

THE WRONG WAY OUT: TYPOLOGY AND IDEALISM

cause it is sanctioned? And if norms are defined by the strength of commitment, why can’t the source of commitment be nonsocial? Eccentric personal revelations such as belief in having been abducted by extraterrestrial aliens can nevertheless change an individual’s life in myriad profound ways (Mack 1994). Requiring cultural beliefs to serve as norms involves many complicated distinctions. Second, the predominant justification for defining culture as shared belief is that it facilitates communication and hence serves as the backbone for social behavior. The underlying assumption is that “people must share some degree of understanding if they are to communicate effectively with one another, if they are to participate in the same tasks” (Borofsky 1994, 331). However, social interaction does not require commonality of beliefs, but rather general role-playing abilities that lead, iteratively and mutually, to the creation and coordination of complementary expectations in particular social contexts (Wallace 1952; Swartz 1991). Third, it can also be inappropriate to characterize groups by some single value when intra-cultural variation is significant. “For a long time there has been a minor scandal at the heart of the study of culture. . . . Culture is shared knowledge and belief; but when we study human groups, we find that there is considerable disagreement concerning most items” (D’Andrade 1987, 194). In such cases, it is questionable whether any summary measure based on agreement can serve to represent a group. In statistical parlance, deviations from the mean dominate any central tendency: too much of the distribution lies in the tails, so the mean or mode represents only a few of the group’s members. It then seems perverse to characterize an entire group by such a minority position. Fourth, it is not a belief‘s normative status but its ability to replicate itself in the minds of others that matters to cultural dynamic^.^ Those beliefs that remain only in the heads of those who originate them die with their originator; they do not enter the social circle and are not transmitted to subsequent generations (Cavalli-Sforza and Feldman 1981; Boyd and Richerson 1985). If one admits that cultural dynamics are of interest, then idealistic approaches fall short. Fifth, since variation is a necessary precondition for endogenous change (homogeneous objects require exogenous forces to be perturbed), idealistic investigations of cultural change must invoke deus ex machina. Postulating such transcendental entities (including ideal informants or groups) is unparsimonious, both in the sense of involving an extra step in 85

CHAPTER 4

analysis after describing variability in data and ontologically, since these entities exist in addition to the individuals that make up groups (unless one is willing to relegate individuals to the unreal). Further, to reify the group is to commit the “culturalistic fallacy” (Bidney 1967; Vayda 1994, 320). Sixth, the importance of consensus could also be defended if a belief‘s degree of sharedness significantly influences how people behave. However, if decision making is based on the subjectiveperception of consensus, then a robust finding from social psychology is that individuals tend to overestimate the degree of sharedness of opinion (the false consensus effect-see Marks and Miller 1987 for review). Since an individual’s social network is usually composed of like-minded people, the error consists of generalizing correct knowledge about the beliefs of associates to larger social groups. The ubiquity of this effect suggests that social order can persist even when intra-cultural variability is relatively high-indeed, this bias may increase one’s sense of community with a large group, despite necessarily limited social contact, because people assume that unmet individuals are also like themselves. Seventh, as Pascal Boyer (1994) points out, the ethnographic tradition of describing cultural groups using ideal knowledge systems is problematic because such portrayals do not describe the thoughts that people actually have in their minds, but rather those that the ethnographer infers are necessary to make sense of what people say and do. The problem is that the ethnographer then treats these constructs as “direct, literal descriptions of people’s mental representations,which of course leads to rather extravagant interpretations”(Boyer 1994,51). In fact, ritualized behavior and some culturally informed beliefs are purposely counterintuitive to demand attention, to set them apart from the backdrop of ordinary life. Thus, many “collective representations,”“world-views”and “[primitive] modes of thought” have to be considered scholastic concoctions rather than descriptions of psychological realities. Indeed, considerable effort has been expended in symbolic anthropology developing sophisticated representations of culture that exist only in the anthropologist’s mind. This undertaking was sustained by the belief that these ethnographers were describing ways of life characterizing whole cultural groups. However, what people actually know does not necessarily exhibit the qualities of idealized systems, such as logical consistency and completeness. Cultures do not smoothly hinge at the joints between belief systems (Barth 1993). Indeed, we probably should not call topical do86

THE WRONG WAY OUT: TYPOLOGY AND IDEALISM

mains of cultural knowledge (e.g., kinship terminology) systematic because, if we admit that culture is in the head, then cognitive science tells us there are a variety of limitations on the ways in which people can represent knowledge. Human reasoning relies upon a variety of domain-specific heuristics and biases that deviate from the standards of logic (Hirschfeld and Gelman 1994). Further, knowledge acquired from others will have additional idiosyncrasies resulting from problems of access and interpretation of information outside personal experience. The typical ethnographic picture of an integrated, coherent, and stable body of knowledge from which appropriate deductions can always be made is therefore not cognitively realistic (Boyer 1994). Constructing idealistic representations in ethnography involves fding in what a number of informants report with implications that may not depict what any collection of individuals actually knows about their culture. Idealism justifies such inferences by arguing that what individuals actually believe is only an imperfect reflection or manifestation of the abstraction, which is more highly valued because of its logical qualities. Some may be unimpressed with the apparent absurdity of describing cultural groups using traits no member of the group need possess, and so continue to pursue the idealistic objective of a single best ethnographic representation. However, methodological and interpretational problems also beset attempts to reach this goal. These problems begin with the fact that the amount of knowledge required to be M y competent in all social tasks exceeds what single individuals can remember (Roberts 1964). As a result, there is necessarily (mostly role-based) specialization in cultural knowledge (Swartz 1991). The methodological consequence is that no individual can serve as an ethnographer’s sole informant, with every answer trusted to be correct. Rather, true cultural knowledge-even within restricted domains-must be cobbled together using some principle from the error-ridden information provided by a number of informants. But who can one trust about which aspect of culture? Each informant’s knowledge is imperfect in some unknown way. A crucial assumption is required to escape from this potential hall of mirrors. The standard recourse in ethnography has simply been for ethnographers to rely on their own corpus of knowledge and understanding, acquired through participant observation, to discriminate between alternative characterizations of cultural knowledge. The ethnographer’s judgment can be given a stamp of authority through effective use of a confident writing style (without cavil or hedging), which induces in the reader a sense of consistent 87

CHAPTER 4

and coherent lifeways in the group portrayed (Geertz 1988). Imbuing an ethnographic report with such features, however, requires a process of idealized abstraction fiom the varied mass of observations and information ethnographers usually collect. However, this method is subject to a fatal flaw: the argument that the ethnographer cannot rely on any single individual to tell the whole truth pertains to ethnographers themselves; ethnographers are fallible people too (Clifford and Markus 1986). A second recourse is to hand the job of idealization to a formal algorithm, which is presumably not subject to human foibles. This is the approach adopted by cultural consensus analysis (CCA; Romney, Weller, and Batchelder 1986). In particular, CCA assumes “that the correspondence between the answers of any two informants is a function of the extent to which each is correlated with the truth” (Romney, Weller, and Batchelder 1986,316). Competent informants, each with good knowledge of a cultural domain, should therefore exhibit a relatively high degree of concordance with each other. Cultural competence is then an individual’s degree of agreement with other informants, and true knowledge becomes consensual knowledge. Further, since culture is defined as the information pool that is shared (Romney, Weller, and Batchelder 1986,316), consensual knowledge again becomes cultural knowledge. CCA is still an idealistic approach: the two kinds of output that CCA produces-the consensual knowledge set and informant competence measures-constitute examples of the two kinds of representation that ethnographers have traditionally used to create idealized depictions of cultural groups.

Group Mind? More recently, Romney and his associates have claimed to provide us with a statistical method for rigorously comparing the mental representations of cultural domains that individuals have in their heads. This representation is conceived as a space in which semantic terms, as the elements of a cultural domain, can be related to one another in terms of distance. We construct a model of culture by aggregating the individual cognitive representations, derived from judged-similarity tasks, into an optimal “compositepicture.” . . . [This composite is achieved by] recent advances in the measurement and scaling of the structure of semantic domains88

THE WRONG WAY OUT TYPOLOGY AND IDEALISM

for example, animal names or kinship terms-as inferred from pair-wise judgments of similarity. . . . [Aggregating over all terms] allows the precise measurement of the degree of overall sharing among the members of a culture as well as the extent to which each individual participates in the common understanding. (Romney and Moore 1998,315) Romney and Moore (1998,332) argue that these composites, thanks to the high degree of sharing within cultural groups, are a “better estimate of ‘what is in the mind of the individual’ than an estimate based on the subject’s own responses.” I argue, on the other hand, that a better estimate of what an individual believes is what would be predicted after accounting for situational variability in the elicitation process. (This is the foundation for the Reflexive Analytical approach discussed later.) This method would predict some degree of real intra-cultural variation, whereas all of Romney’s predicted responses are the same: the cultural ideal. Romney, in essence, says that all variation in response within cultural groups is due to the noise of personal experience (since he doesn’t admit any elicitation bias). This variation is not real because it is not shared between informants. H e says, “The degree of sharing is remarkable; one can confidently assume that every individual shares the same structure. One implication of this is that researchers can prudently use cultural definitions of cognitive representations in applications predicting individual cognitive behavior. . . . Every normal member of a culture shares similar cognitive structures for common semantic domains” (Romney and Moore 1998,332). This is an astounding claim to make because it argues that all individual deviation from the constructed composite is just error: any individual’s beliefs at variance from the cultural ideal are considered idiosyncratic noise rather than meaningful, situated dissent. The only difference, then, in this new approach, is that it seeks to characterize the group not by a point along a single axis (e.g., the groupconsensual response), but rather with a set of structured relations among dimensions within a conceptual domain (e.g., response frequencies in different classes of informants). This remains an idealistic ambition; we have just moved from a consensual representation in terms of a set of independent points (the right answer to a question), to a set of points represented in a common space, and so related to one another as a structure. It is a representation of the “group mind,” if you will. Romney and colleagues (Romney et al. 1996; Romney and Moore 1998) are still considering only 89

CHAPTER 4

a single structure abstracted from individual-level variability that is perhaps characteristic of no one in the cultural group, but nevertheless is considered to represent it. It therefore remains true that only two choices are available to an idealist: culture can either be made into an abstract structure external to the relevant population of individuals or be personified in the form of a single hypothetical, idealized individual. The first representation of culture tends to be about how culture works in that society, and corresponds to the consensual belief-set made up from bits and pieces of informant testimony. The second form of ideal, embodied in the ideal informant concept, provides a normative description of life in the cultural group. In CCA, the scale of cultural competence defines this personified ideal; a highly competent informant is expected to have a number of personal virtues (such as intelligence and experience) associated with this social ability (D’Andrade 1995,212-13). The question that remains is whether CCA, by avoiding the primary methodological problem associated with idealism (the fallibility of human judgment), also avoids the central conceptual problem (providing a robust determination of the single best ethnographic representation of a cultural group). CCA has proven extremely attractive to researchers, as the increasing proliferation of studies using CCA in the social science and psychological literatures attests. This is probably because its goal has been to rejuvenate and legitimate the traditional ethnographic project: to paint a true representation of cultural life using relatively few informants, independent of the foibles of both anthropologist and informant. CCA could be seen in this light as a potential panacea for the postmodern malaise in cultural anthropology, returning us to an earlier age of certainty and objectivity. But my investigations of CCA (detailed in appendix C) have belied this promise. Instead, I have uncovered a variety of empirical and conceptual problems that issue primarily from the method’s implicit idealism. Even if a more reliable algorithm were to be found, it probably would not solve the basic problem: that there is no means besides individual predilection for choosing among the various possible representations that an idealistic, typological approach produces-even assuming it is justifiable to argue the domain in question is cultural, based on external ground^.^ This is because it is the decisions prior to activation of the algorithm (associated with study design) that significantly influence the character of the consensus, not the algorithm itself. Alternatively, other justifications for 90

THE W R O N G WAY OUT: TYPOLOGY AND IDEALISM

idealism besides expert consensus might be defended using independently recognized theoretical principles. For example, a democratic goal might represent a group by their modal beliefs and values. However, I would argue a fundamental difficulty remains. This difficulty necessarily follows from the combination of a particular desideratum (the one true characterization of a culturat group) with the fact that aspects of research design (such as the choice of an informant sample or set of questions to probe knowledge in a cultural domain) do not have a single best solution. This need to choose among representations without a clear rationale for choice is characteristic of idealism generally. I therefore believe the problems highlighted in these investigations debilitate the whole idealistic project. CCA is one way in which those with a scientific bent and ethnographic ambitions have moved: toward classification. CCA is replicable, and thus satisfies the scientific constraint on possible ethnographic practice. However, it fails to be reflexive. As we discovered earlier in chapter 2, it does not attempt to deal with the data collection situation appropriately because its intent is to abstract away from that situation to achieve a verifiable result: cultural consensus. CCA is one way of abstracting from the social nature of primary data collection, but it can only achieve this goal by abstracting from individuals as well. It winds up being neither individual nor social, but produces rather a metaphysical construct. I conclude that there appear to be reasonable grounds for the textualist critique of ethnography: you can’t have a truly valuable ethnography without it also being reflexive.

Notes 1. There is a practical difficulty with using competence as a criterion for informant selection: identification of high competence individuals still requires doing all the work anyway-that is, sampling the entire population with respect to their knowledge in the relevant domain. The point is to use these other qualities, which may be obvious during social interactions with individuals, as proxy measures of competence to identify desirable informants at the beginning of fieldwork. 2. It is interesting to note that the American students’ “low” average competence is .54, which is higher than a significant number of consensus values from other CCA studies (Boster 1991; Brewer 1995; Chavez et al. 1995; Johnson, Mervis, and Boster 1992). Nevertheless, these authors-plus Anita Iannucci and Romney (1994), with average competence less than .54 and eigenvalue ratios of around 2.51-claim consensus for their domains, whereas Romney, Weller, and Batchelder (1986) are loath to do so for the “general information task,” probably because that domain is not a named category of knowledge, but consists of randomly selected questions about history, sports, current events, and so forth. Obviously, Romney et al. doubt that the domain is real. However, Romney et al.’s 93

CHAPTER 4 “general information” could be seen as belonging to a pop culture domain, while Leo Chavez et al.’s (1995) breast cancer etiology might not be a domain for those outside the Western biomedical model, where folk categories may divide the world up differently.This only points out again that definition of a cultural domain can be fraught with operational difficulties and requires some theoretical rationale. 3. I want to emphasize that my point has not been to critique this particular method laboriously just to kill a horse and then beat it. Rather, I thought it best to present the concrete difficulties of interpretation that arise from an idealistic analysis of culture. Since it is one of the few formal approaches to cultural analysis, and has gained a substantial following, CCA simply provided the most effective foil for my more general argument against idealism. In particular, my complaint is not against such methods per se; they are statistically valid, as Romney (1999) notes. Especially since the present climate of opinion seems contrary to scientific approaches to the analysis of culture, I do not intend to give comfort to those who would argue against quantitative methods, nor to be seen as an advocate of purely interpretive work in this area (see Aunger 1995). However, it is important that the tools that are developed in anthropology be given a theoretical underpinning and include internal controls for methodological biases (e.g., due to the data elicitation technique and observer-based biases). I do not mean to deny the reality of cultural ideals either; they can be powerful motivators of action and belief. This can be seen in the recent work of William Dressler and colleagues (e.g., Dressler and Bindon 2000), who have developed a concept they call “cultural consonance.” This measures the degree to which the actual lifestyle of an individual approximates that of the cultural consensus for some domain, such as the “good life.” In effect, Dressler takes the consensus values determined from CCA, and then looks at individuals to see how closely their actual situation in life (belongings, social roles, and relationships) approximates this ideal. H e finds, in an African American community, that individuals whose lives more closely approximate the ideal measure higher on several indices of physical and mental health: in effect, “the more closely individuals approximate in their own behaviors the shared expectations of local cultural models of these domains, the better their health status” (Dressler and Bindon 2000,247). This is a useful way to assess the quality of life in endogenous terms. My argument with the standard CCA approach lies in the use of CCA to attribute ideals to individuals who do not report them, with the tendency of CCA users to suggest that statistical constructs are ontologically real, and held in the minds of individuals. (Note that Dressler asked informants not to respond with their own beliefs and values, but with what they thought the group ideal values for a ‘‘successful life” would be.) Importantly, CCA can be used to investigate cultural diversity, rather than consensus (e.g., Caulkins 2001; Garro 2000). This use, it seems to me, can be perfectly valid, as it avoids the idealistic inferences intended by the method’s original authors. 4. However, normative beliefs may also be cognitively distinct from non-normative ones, in having different motivational foundations. D’Andrade (1989a: 114-18), for example, has argued that widespread beliefs gain new emotional support when they become socially enforced. In effect, individuals begin to think, “I will believe X because I see that most others around me believe X,” rather than believing in X due to some intrinsic quality of that belief. In such cases, the normative (i.e., shared) status of beliefs can also affect cultural dynamics, if the motivational change associated with normative status results in a new bias to adopt those beliefs. Once a belief becomes normative, its rate of further reproduction may increase because of the higher probability that it will be both transmitted

92

THE W R O N G WAY OUT: TYPOLOGY A N D IDEALISM and adopted by individuals. However, it is not the psychological difference (i.e., salience) itself, but the adoption bias that matters to cultural dynamics. 5. Admittedly, as with any research tradition, there is an art of practice into which users of CCA become enculturated, and in this empirical study, I violate a number of the unwritten rules of that practice. For example, the construction of samples, recoding of data as missing, and use of multiple coding schemes from the same set of data, are irregular. As a result, many more results are presented in appendix C than are considered legitimate by those who advocate the approach. However, the art of CCA does not cover all of the necessary decisions in research design, so the sensitivity of consensual representations to such decisions remains a problem.

93

5

REFLEXIVE REALISM: A NEW WAY OF DOING ETHNOGRAPHY

Anthropologists behave like scientists to the degree that they publicly acknowledge the role of the producer and the process in the construction of the product. . . . [Bleing reflexive is virtually synonymous with being scientific. -Barbara

Meyerhoff and Jay Ruby, Introduction to A Crack in tbe Mirror: Rejexive Perspectives in Anthropolou

R

egardless of the skill with which ethnographers conduct their fieldwork, they often find that when they ask more than one informant the same question, different answers are elicited, even when (or especially when) the question bears on cultural beliefs or values. How should we interpret this variability? Does it mean that different informants have different beliefs concerning the topic? Should it be interpreted as spurious deviation from a culturally normative response arising from situational vagaries (such as the weather on the day the question was asked)? Alternatively, have some informants been unable to learn the appropriate response for some reason? O r does this variability result from a misunderstanding of the question a n d o r the anthropologist’s inability to comprehend the informant’s response? Similar interpretational difficulties arise when a single informant responds differently to the same question on separate occasions (as we saw in chapter 3). What is a conscientious ethnographer to do?

Dealing with Cultural Variation

As noted in chapter 4,the history of anthropological thought is rife with typological notions of culture. Definitions of culture have reflected this: 94

REFLEXIVE REALISM: A NEW WAY OF DOING ETHNOGRAPHY

for example, culture is “the sum total of knowledge, attitudes and habitual behavior patterns shared and transmitted by the members of a particular society” (Linton 1940) or “the shared systems of meaning that underlie the ways in which a people live” (Keesing 1976,139). The dominant class of anthropological theory, functionalism, is appropriate for explaining such definitions of culture. Functionalism typically implies cultural unanimity because the hypothesized function serves to achieve some goal considered fundamental to the common lot of humankind (or at most differentiated by sex). The (often implicit) argument is that cultural beliefs are like a language: without commonality of form and content, there can be no communication of meaning. Everything from cosmology to kinship terminology to art forms should thus be invariable between individuals to ensure mutual interpretability. This commonality of belief leads to standardized cultural practices that in turn minimize the coordination problems associated with social interaction. The inferences needed to behave properly are regularized and can thus become subliminal: “It’sjust the way we are.” Any sort of comparative analysis also seems to necessitate the typologizing of cultural groups. Intra-cultural variation is simply not expected from a functionalist perspective. Another popular type of theory in the social sciences, structuralism, admits some degree of variation, but only between classes of people. Structures (e.g., social institutions) are presumed to arise in human groups as the precipitate of joint activities. These social-level entities then have an independent ability to constrain human action. Members of social groups have different types of interaction with these entities because they are assumed to have desires that vary according to, for example, their social roles. As a result, they do not come into contact with the same kinds of cultural information. Any residual variation in beliefs or behavior is just noise, due to the idiosyncratic nature of each individual’s cumulative experiences in life. However, some cultural belief systems exhibit a degree and pattern of variation that cannot be explained even by structural variation. The most extreme possibility is that every individual in the cultural group has a set of beliefs in a particular domain that is unique in some significant, potentially meaningful way. Certainly, this is true of the case study I have been investigating in this book, and this extreme degree of variability may be characteristic of other cultural systems as well.’ 95

CHAPTER 5

Why then has typological thinking continued to dominate cultural anthropology? A number of reasons have been suggested: Dominant conceptions of culture as shared systems of thought, or ability to perform correct behavior, are typological (Mathews 1983, 826; Pelto and Pelto 1975,7).

A method has been lacking to study variability (Mathews 1983, 826). Reliance has typically focused on a few key informants (Pelto and Pelto 1975,7). Stereotyping is a basic human psychological need (Pelto and Pelto 1975, 6). Contrasts between Western and “exotic”cultures were so great as to dwarf within-cultural variation in either (Pelto and Pelto 1975,7). Sophistication has been lacking among anthropologists in quantitative methods of data analysis (Mathews 1983, 826; Pelto and Pelto 1975,8). There has been a general fear of reductionism in anthropology (Mathews 1983,826). Each of these reasons is plausible, and several may apply in a given case. As Pertii Pelto and Gretel Pelto argue, many researchers have felt that cognitive consistency and homogeneity of belief are required for the maintenance of social organization, whereas others (e.g., Wallace 1970) have argued that diversity of belief and knowledge is a prerequisite to social role specialization and cultural elaboration. Economic and social role specialization is likely to lead to diversity of cultural belief, but as will be seen in the following, a great deal of cultural variability can still be found even in traditional societies with an egalitarian ethic. John Roberts (1951) likewise found a great deal of variability between households in traditional Zuni society. One reason that ethnographers have continued to ignore intra-cultural variation is that the theoretical or empirical tools needed to determine the relative importance of different sources of variability in 96

REFLEXIVE REALISM: A NEW WAY OF DOING ETHNOGRAPHY

primary ethnographic data have not been available. As a result, although many anthropological studies have tacitly recognized the existence of variability in the beliefs and behavior of different individuals from the same cultural group, most ethnographers have chosen to ignore that variability when conducting their analyses or presenting their results (Pelto and Pelto 1975,2). Instead, ethnographies have traditionally been constructed on the assumption that group-level processes dominate, with individual-level variability being ephemeral. However, chapter 3 showed that when significant intra-cultural variability in belief does exist, the use of categorical statements to characterize cultural groups as a whole could seriously misrepresent the ethnographic reality being described. Chapter 4 showed that ignoring this variation can lead to significant theoretical problems as well. This suggests that ethnographers will simply have to deal with intra-cultural variability to regain authority. As Handwerker (2002,107) argues, “To the extent that we perpetuate by assumption the construct of ‘cultural group’ bequeathed us by structural-functionalist ancestors, we beg what may constitute the most important questions ethnographers can address in the early 21st century: ‘What groups, by what criteria, where are their boundaries, and how can we know any of this?”’ The crucial question for ethnographers now-given our inability to ignore the failure of idealism as a strategy-is how to properly represent cultural variation. Only by doing so can ethnographers increase the reliability, and hence authority, of the depictions they offer. But how can this elusive goal be accomplished? The solution suggested by textualist ethnography has primarily been to present the variation in relatively raw form, so that readers can make of it what they can. Experimental or interpretive ethnographies merely place the burden of analysis and interpretation squarely on the reader’s shoulders. But this is a burden most readers are ill equipped to assume. In addition, the recent turn to a more self-conscious, experiential ethnography does not solve the basic problem of increasing the credibility of ethnographic reports. One reason is that the textualists’use of variable modes of presentation and viewpoints does not necessarily allow the reader to determine the quality of each different kind of material in the report. A “Rashomon effect’’ may result from the concatenation of narratives: as in Akira Kurosawa’s famous movie titled Rashomon, a number of viewpoints on the same event may be recounted by different participants, but each 97

CHAPTER 5

voice in the ethnography will be slanted in a way that may be unknowable to the reader due to a lack of an overarching contextualization (see Cronk 1993). Readers of experimentalist ethnographies also do not know whether all the relevant information has been presented to them because the traditional reliance on participant observation means that the set of information acquired during fieldwork is necessarily ad libitum. In addition, there is usually no explicit procedure by which the ethnographer selects textual fragments (e.g., field notes, transcriptions, or archival materials) for presentation in the report. George Marcus (1998, chapter 8 ) claims that the true anthropological stance must be subjective; he gives up the commitment to sustain any sort of objective framework into which an ethnographic report can be placed. He contrasts this “anthropological reflexivity” with “sociological reflexivity” (and cites Bourdieu as an example), which tries to get by with simply being aware that social science is generated by and becomes itself a social phenomenon. I counter that the trick is to explicitly recognize the social nature of social science without losing the commitment to an objective framework of ethnographic representation. It is this middle course I wish to pursue in the rest of this book. In contrast to both traditional and textualist ethnography, then, the approach I favor results in a scientific report. Based on formal data collection procedures and analysis, the report includes descriptions of methods and results, reflections on these results, and other, more traditional materials, if applicable. What I call the “Reflexive Analytical approach” does not draw attention to the report itself, as do the textualists, but also it does not attempt to obscure the origin and nature of ethnographic statements, as does traditional ethnography. This approach, based on an explicit consideration of the data collection process as a situated human interaction, includes attention to data elicitation effects, and reports on variation in informant beliefs or behavior, as well as the effects of the context of data collection. The empirical analysis using this approach discussed in chapter 3 has shown that reflexive analysis can effectively separate purely methodological biases in interview responses from answers that more truly reflect the beliefs of informants in a typical ethnographic situation. In this fashion, the Reflexive Analytical approach establishes data quality and achieves a relatively high degree of reliability. Thus, by using the Reflexive Analytical approach to ethnographic data collection and analysis, the ethnographer can simultaneously solve 98

REFLEXIVE REALISM: A NEW WAY OF DOING ETHNOGRAPHY

the two related but distinct problems of determining whether there is substantive intra-cultural variability in belief, and whether particular responses are reliable. First, by partitioning variability in interview responses to different influences (e.g., by using a statistical model), the method determines whether all the observed variability is purely methodological, or represents, at least in part, meaningful differences in belief between individuals in a cultural group.2 The ethnographer can thus address the central problem, identified by Boster and Weller (1990), mentioned above, of determining the sources of variation in basic data. This was, in effect, what was accomplished in chapter 3. How the Reflexive Analytical approach helps achieve the reliability goal is described in the next chapter.

Reflexive Analysis

I have put much emphasis on context. The various contributors to Roy Dilley (2002) try to find ways to make the idea of context more explicit, or at least more bounded, and thus to make the treatment of context possible. My approach is different: to acknowledge that aspects of context will always remain unknown, but that others can be inferred-and even measured-by assuming they exist and then using statistical models to estimate their magnitude. Can you reliably get at the effects of context? Yes, if you use the right data collection protocol: repeat interviews with different interviewers, and so on. This provides a lot of insight into the reliability of your basic data. From this point, you can proceed with firmer convictions. The basic methodological problem in the human sciences therefore consists in isolating the effect of the observer from what is observed within a situational context. The Reflexive Analytical approach, because of its emphasis on reflexivity in an analytical approach to ethnographic description, is therefore based on the concept of a data collection situation, as discussed in chapter 2. But there are further problems we need to address. The standard transformation of field notes into an ethnographic narrative takes place through a secret, implicit, and largely subconscious process. This magical transformation is hardly conducive to validation by others. The authority of ethnographic representations can only be reestablished if the means by which they are constructed is made visible. This visibility is maximized by making the construction process effectively independent of the researcher. 99

CHAPTER 5

For this purpose, an impartial intermediary is required. Formal analytical methods are designed to perform this function. Through the use of formal methods, ethnographers (and certainly readers) need not infer significant patterns and processes directly from undigested data. Formal methods can assist in the interpretation of data through their ability to discriminate patterns in complex datasets. Such methods force researchers to deal with the output of a determinant procedure and to bring about some conformity between what the procedures tell them and their expectations arrived at prior to the analysis. However, the types of patterns to be sought are determined by the researcher, guided by theoretical expectations and the ethnographic insight gained through personal experience in the field. An important criterion for reflexive analysis is therefore that formal methods are used t o analyze data. Such methods include any procedure that looks for patterns in data and has an outcome not predetermined by the investigator. Fulfillment of these criteria almost always involves some form of reduction of the original complexity of the data. Mere management of data is thus not a formal method by this definition. A dominant type of formal method is quantitative analysis using statistical models. In effect, analysis is performed by a machine-based algorithm embedded in statistical software. However, formal methods would also include a variety of exploratory data analysis and some data presentation methods (e.g., graphic displays of spatially represented data). All such methods are analytic in the etymological sense: cases are broken into their constituent parts but only in order to explain the ways in which they fit together; interpretation is required to understand why only certain configurations of parts are realized in the world. Formal methods based on an explicit treatment of the situation in which data collection occurs can actually take advantage of the intersubjective nature of social events. As long as the collection protocol features a variety of relationships between observers and subjects, similar (ie., semi-replicated) events can be compared. This allows the analytical procedure to distinguish those aspects of events due to social interactions from those due to background conditions (i.e., nonsocial causes such as physical aspects of the world at the time and the private motivations of the participants). In this way, the means by which elicitation process itself-biases observations can be ascertained, so the observed phenomenon is ef100

REFLEXIVE REALISM: A NEW WAY OF DOING ETHNOGRAPHY

fectively divided into its two primary aspects: those derived from the observer or situation and those based in the subject itself3 The point is that noise is introduced into basic ethnographic data by contextual factors that influence the process (such as an interview) by which information is originally elicited. However, this noise can be removed, post hoc, by an analytical process involving the use of formal modeling. A major virtue of such a test procedure is that it does not favor one answer or the other to the primary questions at hand. The use of this approach to produce my own dataset made the specialized analyses reported in previous chapters possible. The Reflexive Analytic method essentially attempts to purify the data of its social nature, to get at the individual truths underneath. In sum, three criteria must be met by any approach to ethnographic research that is consistent with both textualist concerns and scientific standards: analytical methods must be both reflexive and formal, and the framework for analysis must be “situationalist” (based on recognition of data collection as an intrinsically social process). This requires the Reflexive Analytical approach to combine the introduction of variability into each aspect of the data collection protocol formal methods (such as multivariate statistical analysis) to isolate the many aspects of the data collection situation that can influence observations, and the incorporation of elicitation effects into the analysis (i.e., reflexivity). In the context of the analysis itself, the Reflexive Analytical model admits systematic bias effects due to measurement error (attributed to the interviewer), interaction effects between the interviewer and informant, contextual effects, and variability due strictly to informant cognition (e.g., beliefs, reasoning). Thus, the data collection situation is fully and explicitly specified by this a p p r ~ a c hThis . ~ is because the Reflexive Analytical model explicitly deals with variation introduced by the data collection situation in the process of analysis, which forms the foundation for the eventual ethnographic report. This is a form of reflexivity. As Davies (1999,4) 101

CHAPTER 5

says, “[Rleflexivity, broadly defined, means a turning back on oneself, a process of self-reference. In the context of social research, reflexivity at its most immediately obvious level refers to the ways in which the products of research are affected by the personnel and process of doing research.” It should be clear how the Reflexive Analytical approach satisfies these criteria. From an operational point of view, Frederick Steier (1991,2) argues that, to be reflexive, “the research process itself must be seen as socially constructing a world or worlds, with the researcher included in, rather than outside the body of their own research.” Certainly, the Reflexive Analytical approach places the observer into the data collection situation, and then analyzes what transpires from the interaction of the observed with the ob~erver.~ I therefore argue that because this approach is reflexive, reflexive analysis satisfies the explicit demands of the textualist critique of traditional ethnographic practice and therefore represents an advance over earlier practice. As a result, the abandonment of scientific approaches to the description of cultural groups cannot be taken as a necessary consequence of the demise of “objective”ethnography.

Reflexive Analytical Methods Exactly what kinds of formal methods are consistent with implementing a Reflexive Analytical approach? The basic requirement is that data collected using a given collection regime must be analyzed in a fashion that discriminates between the purely methodological variability in observations and those aspects that are representative of the subject. This will generally require use of statistical methods for data reduction. Qualitative analytical methods that simply represent data-including taxonomic trees, sociograms, and other diagrammatic forms-cannot characterize the context in which the data were collected as a constraint on the outcome of the analysis (Denzin and Lincoln 1994). While there are rigorous rules for producing such representations, neither the identities of observers nor other aspects of the data collection situation can be made to appear as representational elements so their influence can be “cleaned” from the data. For example, the Reflexive Analytical approach can be formalized using generalized linear models (see McCullagh and Nelder 1989).These models extend well-known linear regression techniques and analysis-ofvariance to logistic regression and loglinear models, as well as multivari102

REFLEXIVE REALISM: A NEW WAY OF DOING ETHNOGRAPHY

ate dummy-variable regression (Del Pino 1989). The Reflexive Analytical approach can also be easily implemented using hierarchical linear models (Bryk and Raudenbush 1992; Hox and Kreft 1994), and less-familiar data reduction techniques for categorical variables such as loglinear models and optimal scaling (see Agresti 1990; Gifi 1990, Weller and Romney 1990). In many cases, however, ethnographers have not collected large samples (i.e., over one hundred observations) of representative data that would be amenable to treatment with standard multivariate statistical methods. This may suggest that implementation of the Reflexive Analytical approach is restricted in its applicability and much more expensive than traditional ethnographic research. It is true that the approach includes some novel elements (e.g., use of multiple observers) that cannot be avoided. However, larger samples become necessary only as the number of factors to be analyzed simultaneously increases, and random sampling is necessary only if the ethnographer wants to generalize results to a larger population (represented by the sample) or to use methods that assume an underlying distribution (i.e., parametric statistical techniques such as generalized linear models). Ethnographers may simply wish to determine what can be said about the sample itself, in which case nonparametric statistical techniques (i.e., tests not based on restrictive assumptions concerning variable distributions) can be used. These techniques make the Reflexive Analytical approach more amenable to the kinds of data traditionally collected by ethnographers. Further, the power and variety of available nonparametric techniques has recently increased significantly with the development of computer intensive randomization procedures (Edgington 1987). These methods provide significance tests for data that were collected without reference to particular sampling methods (although the resulting test values are specific to the sample). Because no reference is made to specific generating processes for the variable of interest, the distribution of values must be found empirically. Computers are generally used to permute data values randomly in order to generate a new dataset of equal size. The relevant test value is calculated for this new dataset. These steps are then repeated a sufficiently large number of times to produce an adequate sample of test values. Finally, the significance of the test variable can be calculated as the proportion of generated values that fall to one or the other side of that measured in the dataset actually observed. 103

CHAPTER 5

Randomization procedures are a fundamental aspect of combinatorial assignment methods, an integrative framework for nonparametric data analysis that constitutes an alternative to standard statistical analysis (Hubert 1987).The simplest formulation in this suite of techniques, the linear assignment model, can be used to compare data vectors (i.e., strings of data values). The classic vector comparison problem is determining the correlation or covariation between variables (as in analysis of variance or the nonparametric Kruskal-Wallis test) or between samples of the same variable (e.g., the t-test or the Mann-Whitney test). The linear assignment model thus covers much of the same ground as standard nonparametric tests and generalized linear models. The quadratic assignment model is an exciting development because it provides significance tests for the structural similarity between different two-dimensional data matrices (fully satisfactory parametric methods for comparing matrices have not been developed). This method provides a measure of the degree of similarity between the association patterns seen in two or more distance matrices. Quadratic assignment can thus be seen as a generalization of the chi-square test, which looks for nonrandom patterns within a data matrix (i.e., an association between categories of two variables). Methods for finding similarity between the structural patterns in higher-dimensioned matrices are also available. In general, these combinatorial assignment methods use the resampling procedures described above to generate multiple derived datasets on which the relevant statistical test is performed. This will produce an empirical distribution of test values from which the likelihood of the observed value can be determined. These computer-intensive approaches-made feasible by recent increases in affordable computer power-are both more general and more realistic for most ethnographic data than the generalized linear model because they allow the rigorous analysis of complex problems that would be difficult or inappropriate to treat under standard distributional assumptions. Further, an important virtue of these permutation approaches is that, unlike nonlinear techniques such as cluster analysis or scaling, they provide a means to assess the likelihood of observed outcomes. Any such statistical modeling has several virtues when applied to ethnographic research. First, a natural interpretation of reliability is built into such procedures. Reliability can be measured as the likelihood that an observation will be repeated, which, in turn, is a function of the probability that the observation was correct the first time (i.e., not due to mea104

REFLEXIVE REALISM: A NEW WAY OF DOING ETHNOGRAPHY

surement error). Since specifylng the criteria for repeatability of the recorded event is what permits assessment of the probability of measurement error, the estimation of an appropriate statistical model, when reflexive, accomplishes the goal of determining data reliability. Second, through the use of reflexive statistical models, a form of validation of the research can be achieved. Validation is a question not of measurement but of appropriateness: does the variable in question measure what theory suggests is the factor of interest? By excluding confounding factors, the context of assessment is constrained, and questions about the legitimacy of the variable measured are reduced. Thus, specifylng the experimental conditions for replication can provide the basis for validation of measurement as well. The investment required to complete a Reflexive Analytical ethnography is not trivial. This novel approach certainly requires more detailed data than usual in traditional ethnographic practice, and fieldwork is reformulated to some extent. Nevertheless, the results of the empirical study on food avoidances reported in chapter 3 suggest that the complete specification of the data collection situation is necessary to determine the quality of primary ethnographic data. In the next section, I detail just what changes to standard practice a Reflexive Analytical approach requires.

Necessary Changes in Ethnographic Practice During the recent bout of self-examination by ethnographers, little attention has been directed at data collection methods. The textualists’ solutions “do not include better ways of doing fieldwork, but different (better?) ways of writing ethnographies” (Wolf 1992, 136). As a result, the search for solutions to the problem of ethnographic representation has been unnecessarily limited. The textualists’ argument has proceeded on the implicit assumption that ethnographers should continue to rely on participant observation and the interviewing of key informants. But other modifications must be made to ethnographic practice besides the kinds of changes in analytical methods I have just described. Data collection methods must also change if the goal of ethnographic research is to characterize cultural variability reliably. In particular, if ethnographic data analysis is to rely on formal methods, then data collection procedures must in most cases themselves change to reflect the requirements of such methods. 105

CHAPTER 5

What specific changes to traditional ethnographic fieldwork does implementation of this approach to ethnography require? If, for example, standard statistical models are to be used for analysis, sample sizes sufficient to allow the statistical isolation of the various factors assumed to cause the phenomenon of interest must be collected. In effect, larger numbers of informants must be queried with respect to any given issue than standard ethnographic practice (which relies on a few key informants) would require. These informants will also typically be randomly chosen, rather than favored for their expert knowledge of a cultural domain. Further, variability in each of the factors identified by the particular specification of the data collection situation must also be introduced into the data collection protocol in order to isolate the influence of those factors. In the case of interview-based data, this means, for example, the use of multiple interviewers. At minimum, to implement a reflexive analysis, a new generation of ethnographers would have to: decide upon some replicable protocol for collecting data; use as many different types of people as possible performing the protocol in parallel with the anthropologist(s), particularly people from the study population itself; repeat the protocol with varying combinations of subjects, researchers, and intervals of time between repetitions of the protocol with a particular subject; randomly sample the subject population; and collect a sufficient sample size of similar data on informants to isolate methodological influences on the data (e.g., by statistical manipulation). Basically, the idea is to systematically vary the major influences on what informants say or do during data collection procedures (including changing the informants themselves), in order to determine the correlation between variation in those factors and variation in what informants report as their belief or value. The pattern of change in responses associ106

REFLEXIVE REALISM: A NEW WAY OF DOING ETHNOGRAPHY

ated with each factor represents the direction of bias due to that factor. The many more minor or indirect influences on what informants say (i-e., situational factors such as weather), which cannot be systematicallyvaried, should have a nearly random influence on responses, and thus do not represent significant directional shapers of data.6 This list is just the minimum set of changes necessary to bring about a reflexive analysis; other changes may be required to use particular methods. Nevertheless, this list represents a significant change in method from the lone ethnographer collecting all of his or her own data through participant observation and ad libitum questioning. Perhaps the most onerous change is the required use of multiple data collectors. However, the ethnographer is almost always surrounded by the people being studied. In most cases, some of them can serve as assistants. (Indeed, recruitment is likely to be problematic only when everyone in the area is illiterate--which is unlikely to be the case anymore, even in the most remote populations-or when there is some other social constraint.) This is one means by which the local people can become involved in research. This can align their interests more closely with the ethnographer’s own, which may diffuse possible tensions and reservations about the ethnographer’s presence. Nevertheless, I would like to emphasize that the types of daily activities undertaken by ethnographers would not change significantlywhen using the Reflexive Analytical approach: ethnographers would still spend the bulk of their time participating in local activities, talking to people, and searching out any historical documents or other records that might be relevant. Other aspects of the Reflexive Analytical empirical program only require more effort than usual to investigate any topic area. It is therefore likely that ethnographic studies will become somewhat more specialized when using the Reflexive Analytical approach. However, from an analytical viewpoint, implementation of the Reflexive Analytical approach will often require some expertise in conducting statistical tests. The development of this skill tends not to be part of every ethnographer’s training at present, but is likely to become more prevalent. The level of skill required to perform even quite sophisticated statistical analysis is also declining rapidly as modern statistical software on computers becomes ever easier to use. I therefore argue that this requirement is not excessively onerous. 107

CHAPTER 5

The approach advocated here can be fruitfully seen as the application of standard sociological practice to the ethnographic situation. Survey interviewing has long been the data collection technique of choice in sociological research on beliefs and attitudes (Sudman and Bradburn 1974). The primary differences between the sociological and anthropological use of surveying is an additional concern with the cultural backgrounds of informants and data collectors due to the cross-cultural nature of data elicitation and interpretation, and longer residence with the study population than is customary among sociologists. Survey interviewing becomes necessary, with random sampling and relatively large sample sizes, thus allowing the statistical isolation of various effects. Survey interviewing has long been the data collection technique of choice in sociological research on beliefs and attitudes. The job of the anthropologist during an ethnographic interview, like that of a pollster, is to find out what people believe about a particular topic. I suggest that the ethnographer use the methods of the sociologicalsurvey to do the same job: polling the opinions of those in the group being studied. The approach advocated here (reflexive analysis using formal methods and random sampling), can be fruitfully seen as the application of standard sociological practice to the ethnographic situation. In essence, reflexive analysis couples social survey collection methods, modified for the ethnographic field situation, with standard statistical analysis of the resulting data. Substantial benefits derive from the combination of survey and ethnographic methods that characterizes the Reflexive Analytical approach. Recognition of this fact has led a number of primarily sociological methodologists to advocate similar approaches (e.g., Agar 1980;Axinn et al. 1991; Freidenberg, Mulvihd, and Caraballo 1993; Sieber 1973; Smith 1987; Sprague and Zimmerman 1989). Detailed knowledge of the local culture arises from long-term participant observation in a community, while the larger-scale data derived from survey collection procedures can be subjected to statistical analysis. In particular, ethnographic insight provides the following benefits to survey-based data (see Axinn et al. 1991): a reduced proportion of nonresponse from informants (due to the build up of trust that comes when researchers live in the community for a significant period); 108

REFLEXIVE REALISM: A NEW WAY OF DOING ETHNOGRAPHY

a reduced bias in observations due to the elicitation process itself, attributable to direct and continuous supervision of field assistants (e.g., interviewers) by the researchers responsible for the study; a minimization of measurement errors (e.g., because researchers with ethnographic insight can design more appropriate questionnaires); and increased validity of the survey instrument (e.g., questionnaires), due to background knowledge and language skills developed by the researchers over time. In terms of analysis, there are further benefits: more meaningful interpretation of the data due to the ability to augment and modify the research design after some experience with the culture; and increased ability to interpret relationships between variables uncovered by statistical models, due to insight into the workings of the society, again derived from ethnographic experience. Many of these benefits will come as no surprise to anthropologists. Surveying everyone in the local community can also avoid problems of representation since the available sample in such cases is the population of interest; statistical estimation is then not subject to sampling issues (Axinn et al. 1991,202). Sam Sieber (1973,1354-55) argues that survey-based data can likewise contribute to the observations derived from fieldwork. For example: (1) surveys demonstrate the generality of field observations by showing how characteristic such observations are of the group as a whole, and (2) surveys can also correct for the holistic fallacy of assuming, based on limited participatory experience in the local cultural life, that all aspects of the society work harmoniously, or that the culture is a homogenous whole. By conducting surveys, ethnographers are more likely to become aware of variation in the group. Using survey and ethnographic methods together can also lead to substantively different results than would have been derived without the ethnographic insight. For example, William Axinn et al. (1991, 209) report the discovery of important social categories while living in the field. Once their statistical analyses reflected these culturally relevant divisions in social groups, they arrived at different, but more appropriate conclusions about the population than were reached without these distinctions. 109

CHAPTER 5

However, it is important to note that the benefits of this multi-method strategy are fully realized only when rigorous survey and ethnographic techniques are applied simultaneously and integrated at every phase of the data collection process. .. . It is the contact between the two approaches that allows the strengths of each to compensate for the weaknesses of the other. (Axinn et al. 1991,214) Let me hasten to add, however, that this discussion of survey sampling does not mean the Reflexive Analytical approach is limited to interviewing. Basing the Reflexive Analytical approach on a data collection situation makes it easily generalizable to a variety of data collection techniques. For example, responses to a written questionnaire can be seen as formally equivalent to a social interaction with a passive interviewer, since the researcher effectively serves as an interviewer-once-removed when devising the questionnaire. Further, little modification of the approach would be necessary to deal with ethological behavioral observations (Tinbergen 1951; Lorenz 1950) or time allocation studies (Gross 1984; Johnson 1979, since such procedures involve direct observation and hence social interaction between the researcher and the subject at the time of sampling. The only kind of data collection procedures not readily amenable to treatment using the Reflexive Analytical approach are ad hoc methods, such as participant observation, that do not involve replication of at least some aspects of observations. The Informant Selection Problem

One of the more interesting aspects of data collection using the Reflexive Analytical approach is the change in rationale concerning informant selection. The guidebooks for ethnographic data collection (Johnson 1990; Werner and Schoepfle 1987; Bernard 1988) have a variety of recommendations for field-workers concerning the selection of informants. However, all are concerned with how to get the most accurate information about a culture from the fewest informants. This is viewed as an empirical inference problem. Werner and Schoepfle (1987) and particularly Bernard (1988,177-78) tend to focus on personality traits in the ideal informant, suggesting that although a good deal of luck is involved, good informants are trustworthy, good storytellers, reflective, articulate, and observant. Johnson (1990), on the other hand, is concerned with elimi110

REFLEXIVE REALISM: A NEW WAY OF DOING ETHNOGRAPHY

nating the “luck factor” by describing methods for selecting informants that can be used in the field. He (1990,37-39) suggests a two-stage procedure. The first stage is selection of a potential pool of informants based on their “theoretical qualifications” or “representativeness”with respect to a number of variables determined either a priori (e.g., social status, knowledge, subgroup membership) or along dimensions that emerge during the course of the study (e.g., through use of multivariate reduction techniques like factor analysis). Johnson (1990,27) distinguishes “statisticalrepresentativeness” from “theoretical representativeness,” the latter of which is determined by the researcher’s question, such that selection of informants is based on a priori determination of those social segments pertinent to that question. The former is difficult to determine, since it depends on random selection from a known population. However, the relevant population may be difficult to determine, much less sample. The form of representativeness to be used in a particular study depends primarily on the availability of theoretical reasons for selection of a subpopulation for study, or secondarily on more practical concerns associated with being in the field, such as how sparsely potential informants are distributed geographically. From this set, during the second stage, personal decisions can be made of candidates with respect to their innate abilities (which include the traits mentioned by Werner and Schoepfle and Bernard). But if statistical analysis is used to estimate both specific and unknown contextual factors (and I do not know of any method besides statistics that can determine the effects of unspecified contextual factors), sampling of informants cannot be based on an informant’s expertise in a particular domain of cultural knowledge. This is because standard statistical models typically assume random sampling from a population. A rigorous treatment of contextual effects on primary data therefore requires that information representative of the population as a whole be collected. Thus, ethnographers must randomly sample the population of interest to them in order to take advantage of the complete, reflexive specification of context made possible by statistical analysis.’ Thus, my emphasis is entirely different from the traditional one. To document within-cultural variation in some aspect of a cultural system, one cannot rely on key informants and participant observation. Instead, the new goal is to collect as large a representative sample of information as possible from the relevant population being studied, in a strictly controlled manner suitable to inter-individual comparison. This will allow the 111

CHAPTER 5

pattern of variability in the population to be properly characterized with respect to the relevant qualities of the individuals. Random sampling is not standard ethnographic practice. With limited resources and time, anthropologists have traditionally attempted to gather high-quality information from the best informants, with the objective to characterize the culture of the group as a whole most efficiently (Johnson 1990, 28). Interviews with a relatively few key or expert informants, who by definition are not representative of the group from which they come (Johnson 1990,22), are assumed to be sufficient to accomplish such an objective.8 Furthermore, observations in traditional ethnographic practice are typically not collected systematically enough from different informants to be statistically comparable in the first place. However, misrepresentation is more likely to occur when ethnographers follow the traditional practice of asking only a relatively few (i.e., key) informants about any particular topic. If informants do not respond consistently to questions, then the certainty of any particular response is low. Building an interpretation of some aspect of culture based on a series of such weak links can severely reduce the likelihood of arriving at a valid conclusion. This argument thus indicates that reflexive analysis, achieved through a complete specification of the data collection situation, requires considerable changes to traditional ethnographic data collection methods. However, such changes would result in a more scientific (i.e., reliable) approach to ethnographic description than has typically been considered possible by those engaged in the recent debates about ethnographic practice. In conclusion, I argue that a scientific cultural anthropology is possible. However, it will have a number of features distinct from traditional ethnography-and different from the purely verbal accounts that have usually been derived from studies based on participant observation. Scientific ethnography depends on a realism that is both reflexive and considerate of variation-major tenets of the textualist revolution. At the same time, it is not exactly what postmodernists had in mind either-in particular, it is not purely narrative, as expected by the new wave in ethnographic writing. This is because reliability and thus authority are only achieved by accounting for variation due to extraneous forces arising during the data collection process itself, and in the decision making of the analyst. Unlike textualist approaches, reflexive variation-based analysis becomes reflexive through the formal elimination of spurious influences on 112

REFLEXIVE REALISM: A N E W WAY OF DOING ETHNOGRAPHY

primary data. The job of constructing a representation remains with the ethnographer, rather than the reader. Of course, those who define ethnography as a form of qualitative research (e.g., Davies 1999,4) will no doubt be unhappy with this solution, based as it is upon rigorous data collection and often quantitative methods of analysis. The intrinsic “softness”of ethnography, and its poetic evocation of lived experience, may seem to have been sacrificed to gain the cold, hard mantle of scientific respectability. However, it remains the case that good reflexive ethnography will depend on long-term participant observation to properly interpret and understand the hll meaning of the output from quantitative models. A reflexive ethnographic report will not consist solely of tables and charts. A significant role remains for traditional kinds of investment by the ethnographer in the lives of informants and their social group. The need for grounded experience is not lost. It is merely supplemented by concern for a core domain of belief or behavior. This domain is treated in a rather specialized manner to make it amenable to scientific investigation. Outside this domain, standard practices apply. The ethnographer must still get a feel for the place, through everyday involvement, to be able to appreciate the context in which the focal domain exists, to couch the report of that domain within a larger framework of ethnographic understanding. It is only the work on the focal domain that is replicable, hence comparable to similar work in other domains and cultures, and so M y scientific. Reflexive analysis is not a panacea. In particular, its practice cannot return us to the golden age of encyclopedic treatments of life within a cultural group. Nevertheless, its ability to solve the problem of ethnographic authority recommends it as an addition to professional practice in the social sciences.

Notes 1. The number of studies documenting significant intra-cultural variation is presently small (Pelto and Pelto 1975),so it is difficult to say one way or another. Ethnographers simply have not looked for such variability, and have tended to discount any evidence of it they may have encountered, since it does not conform to theoretical expectation. 2. It is important to remember that use of any method for validating a particular data collection protocol represents only the first step in empirical research. Once the nature of methodological variability has been determined from a relatively small dataset using the approach advocated here, the ethnographer may rely on a larger, independent sample of similarly conducted, but non-replicated interviews to conduct substantive analyses using standard ethnographic and statistical methods.

113

CHAPTER 5 3. Some will argue that this division is an abstract one, that events in the world cannot be disaggregated along such lines except through such artificial means as a statistical test. This is true enough, but, as I will argue later, it is a necessary first step in the process of isolating the causal processes of real interest. 4. The Reflexive Analytical approach has the ability to be used in either individual- or group-level analyses, unlike the approaches reviewed in chapter 2. There is no a priori commitment to derive group-level values from the primary data, as is intrinsic to the Informant Error school or the Culture-Centered approaches. However, such an aggregation could be achieved in a second analytical step in the Reflexive Analytical approach, if desired. Otherwise, the data can be analyzed at the individual level. This is because purumefer values are estimated with the Reflexive Analytical approach (e.g., functional operators defining aspects of the relationship between the informant observation and various aspects of the data collection situation). In the other ethnographic approaches, it is variablevalues for characterizing the group that are estimated. This difference arises because the Reflexive Analytical approach attempts to determine the influence of the data collection situation on observed response values, whereas all existing approaches attempt to infer unobservedvariables (e.g., group-level cultural norms) from observed ones (e.g., informant responses). 5. Doubts may remain about whether reflexivity as treated in this book is what the textualists mean in their use of this term. Norman Denzin and Yvonna Lincoln (1994,480), following Martyn Hammersley (1990), argue the reflexivity notion is postpositivist, because it is couched “in terms of taking account of the effects of the researcher and the research strategy on the findings that have been produced.” In their view, the only difference between postpositivism and its progenitor, positivism, is its acknowledgment that the reality “out there” cannot be completely apprehended or understood, so that the researcher should become sensitive to observation effects. Marcus (1994,569) argues that this conception of reflexivity as observer bias is inadequate because it is merely a confessionalist frame for standard ethnographies; although experiential, it is not properly contextualized. Although reflexivity has been expressed here as a form of bias, it is quite different from the reflexivity-as-personal-bias associated with postpositivism. That sort of bias is consciously recognized and reported by the researcher as a kind of caveat with respect to results reached prior to or independently of the reflexive act. The Reflexive Analytical measure of bias, in contrast, is found by a formal modeling technique, not by introspection, and is fully integrated into the analysis rather than remaining a subjective or experiential overlay to the quasi-objective treatment of data. The Reflexive Analytical approach can contextualize events in a variety of ways (depending on what factors are identified in the model), so that the researcher can determine how individual cases deviate from expected outcomes because of the influence of each effect impinging on the observed situation. For example, the nature of the interpersonal chemistry that lies at the foundation of intersubjectivity in the human sciences can be elucidated. Marcus (1994, 572) prefers a reflexivity that is positional and locational. “Locational” means that the ethnographer’s approach is located with respect to alternative interpretations based on different sociopolitical agendas. This contextualization is achieved “through a keen sensitivity to the complex overlay of related, but different, accounts of almost any object of ethnographic interest” (Marcus 1994,57l)-in particular, previous ethnographic accounts of a particular cultural group or the topic’s disciplinary history. In feminist research, positioning is associated with standpoint epistemologies (e.g., Hartsock 1983; Smith 1989). Although in practice positioning is often reduced to essentialistic confessions 114

REFLEXIVE REALISM: A NEW WAY OF DOING ETHNOGRAPHY (e.g., “I am a white, Jewish, middle-class, heterosexual female”),the ideal is to describe individuated personal experience within a particular material and social situation (often as part of or leading to a Marxist social critique). Standpoint feminist research is located, but only “locally,”since it is impossible to encompass the universe of context; all knowledge is necessarily situated and partial (Marcus 1994,572). I would argue that the analytical reflexivity advocated here provides the positional and locational information associated with the qualitative approaches favored by many feminists, critical theorists, and advocates of cultural studies. Further, Marcus (1994,572) argues that Donna Haraway’s (1988) intellectual program is particularly bold and laudable because it aspires to more-than-local positioning by defining “a space of juxtapositions and unexpected associations formed by a nomadic, embedded analytic vision constantly monitoring its location and partiality of perspective in relation to others.” This seems an accurate if poetic description of the Reflexive Analytic approach as well: embedded, analytic, located, partial. Not only does the Reflexive Analytical approach capture the experimentalistethnographers’multi-vocality through the use of many observer-observedpairings, but these different partialities are themselves situated: each voice is positioned relative to the others in an overall space defined by a set of biases. This provides the Reflexive Analytical approach with an overarching frame similar to Haraway’s jwtapositioning, although not dependent on the researcher’s conscious organizing abilities. Despite all these claims, I’m sure that some will say, like Graham Watson (1987, 29), that I am (like other anthropologists) “claiming to confront reflexivity while merely managing it.” 6. Bias can also be introduced into the presentation of a study at other points; for example, during the development and implementation of a coding scheme for responses, in the formulation of a statistical model, or in the interpretation of results. I conducted all of these steps single-handedlyin the present case. However, the reliability of these steps could also be addressed by introducing variation at these points (e.g., by including interpretation of the model results by other individuals). 7. There are exceptions to this rule, however. Recently, sophisticated non-parametric multivariate statistical techniques that are applicable to ethnographic estimation problems (e.g., social network analysis-see Scott 1988; Wasserman and Faust 1993-or empirical resampling techniques such as bootstrapping) have been developed. When these techniques are used, the results are specific to the sample population. However, the sample is sometimes a close approximation to the population that interests the ethnographer and so this need not represent an argument against the use of such techniques in these cases. If the population of interest is relatively large (e.g., an ethnic group with over one thousand members), random sampling is still warranted. 8. However, only the Cultural Consensus approach has attempted to justify the claim that information from a relatively few key informants can reliably indicate characteristics of the group as a whole, based on highly restrictive assumptions concerning the inference of normative cultural values from small samples.

115

6

MEASURING THE “STRENGTH OF BELIEF”

P

revious chapters suggested there is real psychological variation in belief within cultural groups, not just variation in data due to methodological problems. How much variation is real within a given belief system can be determined by the methods used in chapter 4. And the Reflexive Analytical approach outlined in chapter 5 is able to produce reliable ethnographic analyses of particular cultural systems. But to achieve a fully scientific ethnography of cultural traits, we need to have a method for comparing the degree of real variation between belief systems or value systems, and even between cultures. This is, in effect, a way of getting external validity for the general method we are advocating by being able to apply it everywhere. We need to know: Is there more or less variation in belief among the Agta than the Zuni? Just how different are their cultural systems? A first step toward this lofty goal is to devise a comparative measure of the reliability of individual ethnographic reports-in effect, a simple index of the degree to which an ethnographer can trust that an informant’s oftentimes idiosyncratic response is their personal truth, rather than being fabricated for the occasion. I will call this measure of individual reliability the informant’s “strength of belief? One way to assess the strength of belief in a particular domain is to measure just how strong the tendency is to repeat a response, given changes in the circumstances of an interview. In effect, we need to calculate the probability a response will be repeated in multiple interviews of the same respondent. The measure I advocate is related to the difference between the random expectation of a repeated response and the observed frequency of the same response. I call this measure the repeatability bias index (RBI). The index provides a means of comparing error rates between cultural systems, and even across societies, because it is expressed in 116

MEASUFUNG THE “STRENGTH OF BELIEF”

terms of probabilities that do not rely upon information about the peculiarities of a particular belief system. In order to be comparable across cultural groups and cultural systems, the measure of “strength of belief” must account for the fact that the expectation of a repeated response is a function of the number of different responses that can be made (i.e., the number of different categories of the cultural trait being studied). The simplest way to measure the reliability of repeated responses is to compare the observed rate of repetition to that expected for purely random responses-that is, to estimate the degree of bias toward repetition of the same response. The measure I suggest is related to the dflerence between the random expectation of a repeated response and the observed frequency. Mathematically, I suggest the following as an index of repeatability bias:

where po = the observed proportion of repeated responses, pr = the random probability of a repeated response (measured as a relative frequency), and N = the number of different states of the trait being considered (a scale correction factor). This measure is simply a standardized Chi-square statistic under the expectation of random responses. This statistic measures the observed deviation from expectation as a percent of the expected frequency of response, thus allowing a comparison between systems that have different expected error rates. This measure of bias toward the repetition of responses ranges from -1 when there is no overlap between responses to 1 when there is perfect agreement (or an infinite bias to repeat a response). Larger positive values of this index indicate greater bias toward the repetition of responses, negative values indicate a bias toward the mention of different responses, and a value of zero is the degree of repetition expected when responses are random (given the number of possible response types). Note that this index provides a measure of the relative strength of situational versus non-situational factors in the determination of informant responses-that is, informant- and interviewer-related effects-without any additional information about the belief system except for the number of categories of belief, which should be known a priori. 117

CHAPTER 6

This measure has the additional advantage of requiring no additional information beyond that already available in many anthropological studies in which repeated interviews were conducted: calculation of the repeatability bias index depends only on the average rate with which informants change their reports concerning a particular belief or attitude. In lieu of conducting new studies that include more sophisticated research designs for the investigation of intra-cultural variability, this index provides a way to glean additional information from this summary measure of average intra-informant error. Use of the index can assist in the formulation of an appropriate statistical model for the future investigation of the case at hand, as well as provide a means of comparing error rates between cultural systems, and even across societies. I now want to explain just why RBI can be considered a measure of reliability by looking more closely at what reliability means and why it is important.

Improving the Reliability of Ethnographic Reports If we are to measure variation, not just central tendencies or modal values, then we must have a sense of whether that variation is real. As a result, ethnographers must be able to assess the degree of confidence to place in the data they collect in the field. In this context, an important benefit of using a Reflexive Analytical approach is that the reliability of data can be estimated.l Determining data reliability is a much-neglected, but methodologically vital first step in ethnographic research (White 1990). As Douglas White (1990, 118) notes, “The study of reliability is a necessary starting point for any attempt to provide explanations that account for variability.” The historical lack of attention to data reliability may be due to the view that observations made in the field cannot be replicated. Philosophers of science have long argued that replicability is the sine qua non of scientific investigation (see Rosenthall991 for review). However, this criterion is problematic for ethnographers because replicability in the temporally dependent world of social science becomes somewhat nonsensical. Any observation is uniquely tied to the time, place, and mental state of the observer. Harry Collins (1989, for example, dismisses the possibility of replication as a criterion of social science. It is simply impractical in cases 118

MEASURING THE “STRENGTH OF BELIEF”

where the researcher cannot control for the relevant conditions, thanks to ethical constraints on manipulating people. This is especially the case, obviously, for anthropological fieldwork in “naturalistic”conditions. Ethnographic studies cannot be replicated for the obvious reason that no ethnographer steps into the same river of events twice. This is the problem faced by those attempting to adjudicate between Mead and Freeman (discussed in chapter 1):generations had elapsed between the visits of these two ethnographers to Samoa, and obviously many things on the ground had changed in the interim. In one sense, however, this view reflects a misperception about the nature of replication: while all of the conditions of a given observation can never be duplicated, different observations nonetheless do share some aspects in common. In the case of interviews, the identity of the interviewer or the nature of the question asked may be the same. Reliability analysis makes use of what is common to several instances of a phenomenon in order to determine the effects of aspects that have changed between the different instances (Rosenthal 1991,2). Reliability concerns what is repeatable in measurement, such as different measures of an attribute, or different repetitions of the same measure of an attribute. It is a generic concept, referring to the accuracy (equivalence and stability) of measurement within a particular context of replication, including both the population studied and those engaged in doing the study. (White 1990,110)

From this perspective, reliability is a measure of how likely it is that similar conditions will give rise to the same observation. White argues that this is a function not just of the phenomenon being observed but of the effect observation itself has on the nature of what is observed. This is similar to the textualist criterion that ethnographic representation must reflect the interpersonal nature of fieldwork. Thus, reliability is taken not in the sense of temporal repeatability (i.e., an ability to return to the field site, collect new data, and find the same cultural state in place), but rather as a kind of analytic replicability: the ability of repeated examinations of the data (e.g., by an independent researcher) to come up with the same answers. High reliability therefore suggests internal validity, or conjunction with the aspect of reality examined by the research (Altheide and Johnson 1994,487), and so is to be 119

CHAPTER 6

valued.2 As White (1990) continues, reliability can either involve multiple observers of the same phenomenon, or multiple observations of similar phenomena. In the present case, then, reliability will be taken as a measure of the strength of bias or the tendency to repeat responses under semi-replicated conditions. This measure is the difference between the expectation of a repeated response (which is a function of the number of possible responses) and its observed probability-in effect, just what RBI measures. This emphasis on reliability suggests that the most that can be achieved in ethnographic research is a temporally specific assessment of current variability that strives for accuracy, along with a characterization of the processes affecting the current state of the system. For this reason, the ambitions of anthropology must become more modest: a t minimum, they will be culture-specific and time-specific. O n the positive side, such an approach will go a long way toward remedying the complaints of critical cultural anthropologists like Clifford Geertz and James Clifford.

An Example: Food-Related Belief Systems To establish the utility of the index I am proposing, I turn again to published studies on humoral beliefs (used previously in chapter 3), and compare the RBIs for those studies to that characterizing the Ituri Forest data on food taboos, which have been the subject of numerous analyses in earlier chapters. Comparison to the Ituri study of food avoidances is particularly appropriate because it also involves food-related beliefs, but has a distinct cultural origin (it is unrelated to the humoral system of early European civilization and its subsequent spread) (Murdock 1980). In addition, to a greater extent than seems the case with the humoral systems, the food avoidance system in the Ituri is dominated by culturally sanctioned taboos. It is also useful, first, to compare the value of the index in these foodrelated studies to another real-world value to establish a kind of baseline, or expected value, for the index in a relevant cultural task. For this purpose, a simple identification task with respect to an everyday object is best. Fortunately, a published report concerning just such a trait is available to provide this baseline point of comparison. Boster (1985) describes results from a study of variability in the ability of Aguaruna Jivaro 120

MEASURING THE “STRENGTH OF BELIEF“

horticulturalisdhunters of Peru to identify examples of different varieties of manioc plants, their staple crop. The basic task was to wander around a specially planted field and identify the fifteen most common varieties of manioc, with five plants per type, making a total of seventy-five identifications per informant. Boster (1985, 193) states that the intrainformant proportion of agreement in identifications from this task, which was repeated with six individuals after an interval of three months, was approximately the same as the average inter-informant proportion of agreement. This latter rate can be calculated from Boster (1985, figure 2) as approximately equal to .76. Since Boster (1985, 180) coded for 127 different varieties of manioc, the index for intra-informant repetition bias in this case would be .75.3 In the studies on the classification of the humoral qualities of foods, to which I now return, there are only two possible responses to the question asked (“hot” and “cold”).The probability of randomly repeating the same answer when there are only two possible answers is one in two, or .5. Unfortunately, few of the available studies on humoral systems provide quantitative estimates of the degree of intra-informant discrepancy. However, two such studies do exist. Foster (1979, lsl),in his long-term study of humoral beliefs in Tzintzuntzan, Mexico, reports a roughly 20% error rate in three expert informants on a 184 item list of foods, where responses were primarily bivariate (Is food X “hot” or “cold”?).Thus about 80% of answers were repetitions of previous responses in Foster’s study. Three to four years transpired between his interviews. In contrast, Mathews (1983, 841) found that the overall rate of discrepancy was 11.8% in a sample of forty Oaxacan informants when the second interview took place after a period of four months. Her list had 102 items. For these two studies, the value of the bias index ranges from 0.30 for Tzintzuntzan to 0.38 for Oaxaca, indicating a small positive bias toward repetition of the same response on subsequent interview^.^ These two values are also very close to each other, as befits studies undertaken in nearby locales with respect to the same belief system. (This closeness further suggests that there isn’t a great ethnographer influence evident on the estimation of this index of bias.) These values are lower than the baseline value calculated above for the manioc identification task, as might be expected when the task is to elicit cultural beliefs as opposed to simple identifications of real-world objects. (Responses to the identification task were simple nouns, rather than the adjective required by the humoral belief studies.) 121

CHAPTER 6

I will now compare these results from humoral belief systems to the intra-informant error rate determined from the study on food avoidances in Zaire. For this system, with thirteen different types of re~ p o n s e ,the ~ probability of a repeated response is only %3, or 7.69%, assuming no bias. However, the overall rate of discrepant responses actually observed was 11.20% of all repeated responses (with 145 items on the list, which is roughly comparable to the humoral studies above). The index of bias toward a repeated response is therefore .81 in this case, suggesting a considerably more powerful bias than in the humoral systems mentioned above.6 I should note, however, that the value of this index can change, even for the same set of data, if the coding scheme is varied. For example, with the Ituri data coded more specifically,using thirty-six different categories (rather than thirteen), the value of the index becomes lower, only S4.’ This value is still larger in size than those from the South American humoral studies, but not by such an impressive margin. Note that both of these indices are calculated from the same set of verbal responses; only the ethnographer’s interpretation of those responses has changed. This difference in the index suggests that the more specific coding scheme is perhaps splitting the cognitive domain of food taboos in a less meaningful fashion, since the index could have been roughly equal in value for both coding schemes. This last comparison does suggest a certain weakness in the index, since it introduces a subjective element: the ethnographer’s potentially somewhat arbitrary categorization of responses. This kind of weakness in RBI is consistent with the literature on coding problems that arise when making comparisons between informants. Forrest Young and Robert Young (1961), for example, found that inter-informant agreement was high on questions that didn’t require much inference by informants (such as factual ones: Is there a church here?), but low for high-inference questions (How friendly are people here?). Since it is almost always possible to code informant responses in binary fashion, this kind of coding would apparently make for the safest comparisons. Of course, if the Ituri data had been coded in binary fashion like the humoral studies (e.g., just taboo/no-taboo), the RBI measure would be even higher, and the gap between the taboo value and those of the humoral systems even larger. So the coding difficulty does not introduce a problem in terms of ranking 122

MEASURING THE “STRENGTH OF BELIEF

outcomes, only in terms of measuring the quantitative difference between these belief systems. A shorter interval between interviews does not appear to be the reason for the higher index of repeatability bias in the Ituri. The Ituri interviews were repeated after a shorter interval than Foster’s, but not Mathew’s humoral studies, which would tend to argue for higher repeatability. And the fact that Foster’s and Mathew’s studies show roughly the same bias rate, despite one set of interviews being conducted years apart and the other only a few months apart, also suggests low salience of timebetween-interviews, even though a period of years between interviews would allow learning to take place, which might have made a greater difference. It is also the case (see chapter 3 ) that the amount of time between interviews had only a marginal impact on repeatability in the Ituri case. At any rate, for true comparability, an investigator would want to ensure that not only was the bias index calculated, but that these other factors (numbers of potential responses, the average time between interviews, and so on) were also taken into account. A natural extrapolation of this kind of study would be to calculate the index more specifically for each question in an interview. For example, there is a clear pattern in the Ituri data for more dangerous beliefs to be more frequently repeated (or reliably reported) than more innocuous beliefs (see chapter 3).The directionality of the bias is also interesting: there is a higher probability of an individual “forgetting” a previously reported response than for a taboo to be suddenly “remembered” the second time around. This seems sensible enough, since individuals will probably be less attentive in an interview situation that is more familiar because it has been engaged in before. A major problem with attempting to make this extrapolation, however, is that the necessary, question-specific information is rarely reported in the ethnographic literature. The use of such a measure would therefore be highly circumscribed, whatever its intrinsic merit and interest might be.

The “Need to Believe” One curious aspect of the investigation thus far is that the data from the Ituri, although with respect to a belief system, have a higher value of repeatability bias than found in the more behavioral task of identifying 123

CHAPTER 6

manioc. This is somewhat surprising since many kinds of responses to the food edibility questions are possible, and most people would think that it would be easier to name a plant twice in the same way than to remember a complex rule about some obscure animal of the forest. There are two possible explanations for the difference in estimated bias between the Ituri study and the studies of humoral belief shown above and perforce the manioc identification study. Either there are methodological improvements in the Ituri study, or the bias to repeat responses is actually stronger in the case of food avoidances. It might be thought that my use of local assistants in the Ituri study reduced any problems associated with inter-cultural communication, characteristic of the humoral studies conducted by the anthropologists concerned. But while the discrepancy rate for responses in the Ituri is somewhat higher for those repeated interviews involving only the anthropologist-as-interviewer (see chapter 3), this factor cannot explain very much of the difference between the RBIs of the humoral and food taboo studies. I therefore believe the overwhelming proportion of the difference is due to a difference in the ontologicdepistemological status of the two kinds of food-related belief systems. Both the humoral systems and the taboo-based system of the Ituri are used by their respective cultures to explain types of illness that are believed to result from consumption of inappropriate foods. Yet it is generally agreed among researchers on humoral systems that valence beliefs exhibit a relatively high degree of variability, both between informants and within informants (see Mathews 1983 for summary). The difference in the bias index calculated here suggests that the situation- or context-based variability within informants is significantly higher for humoral systems than the taboo-based system of the Ituri. There are two possible reasons for this. One possibility is that there are simply many contexts in which the valence of a particular food changes, whereas the taboo status of a food does not change with reference to the food’s preparation for consumption, use as a medicine, or qualities as an aphrodisiac: foods that are taboo to an individual are tabooed for consumption regardless of the context. However, with respect to humoral beliefs, two comparable studies on neighboring populations have come to opposite conclusions regarding the sensitivity of food valence assignments to contextual variation (Boster and Weller 1990; Mathews 1983; both studies were reviewed above). Further research will be required to determine whether questioning informants with respect to 124

MEASURING THE “STRENGTH OF BELIEF”

specific contexts can significantly lower within-informant discrepancy rates concerning food valence assignments. Only then can it be known whether the difference documented here between repetition bias in the two kinds of food-related belief systems is due to this methodological problem. The other possibility is that the severity of the consequences following violation of humoral beliefs may be lower than for food avoidance beliefs in the Ituri. Humoral systems may not be culturally sanctioned to the same degree as the sub-Saharan African food avoidance system. My impression is that valence beliefs are not viewed as being as dangerous as food taboos when violated. For example, Mathews (1983, 829) reports that Oaxacan informants regard few foods as being dangerous. In comparison, consumption of tabooed foods in the Ituri is associated with a wide variety of severe illnesses, including rapid death. In addition, there is a normative model of cultural transmission of food avoidances in the Ituri, so that individuals know from whom they should learn their food taboos. Given the danger of even mistaken violation of these taboos, individuals tend to take what they learn to heart.8 The sample of informants in the Ituri study can also typically rely on specific memories of cultural transmission of their avoidances to improve their probability of repeating the same response. The measure of repetition bias used here might therefore be a good indication of the degree of a belief’s memorability and/or the strength of cultural sanction behind that belief. The lower rate of repeated responses for humoral systems is also consistent with the ethnographic impressions of the anthropologists who conducted the studies. For example, Valerie Hull (1986) and Laderman (1983) both report considerable flexibility in the use of humoral categories for particular foods in the societies they investigated: the valence of the same food would often fluctuate according to the kind of use assigned to the food. Given that this practical context would often be lost or remain only implicit in a formal interview setting, such multiplicity in the uses of some foods would naturally lead to a lower probability of the same individual repeating their earlier answer to a question about those foods. Let me now draw some lessons from this preliminary investigation of the utility of the RBI. Generally speaking, this analysis suggests that the rate of overall discrepancy in informant responses on questions of belief is likely to be domain-specific, and to reflect the degree of social pressure 125

CHAPTER 6

supporting the belief system itself. Humoral studies seem to show a more random pattern of responses than the culturally sanctioned food avoidance system in the Ituri, due to their lower values of response bias. Informants’ responses appear to change less significantly for food avoidance reports due to situational variability than do humoral beliefs. For more personal or attitudinally determined beliefs (i.e., ones not supported by social pressure), greater situational or contextual variability might therefore be expected. However, it should be noted that the measure of response bias is not a measure of the uniformity of correct responses. Rather, this measure estimates the degree of consistency within informant responses to a set of stimuli (not between informants, as is the case with cultural consensus modeling). This difference can readily be seen in the system of food avoidances in the Ituri population, where there is tremendous variability in the types of avoidances associated with a given animal, even within a single family, but also a high degree of consistency in what an informant says about his or her own belief, This bias measure can thus detect the existence of a strong cultural system of belief despite significant withincultural variability in beliefs, simply by measuring the degree of nonrandomness in responses. In addition, as I argued earlier, the cultural system is not about beliefs per se, but the nonrandomness in responses was the result of a normative system specifying from whom individuals learn their food taboos. Reliability thus appears to depend not so much on the methods of the interviewer as the strength of the cultural system being addressed in the questioning, although in chapter 3, I did find that discrepancy rates are higher for interviewers unfamiliar with the culture of the group being investigated (to wit, the anthropologist). I therefore conclude that the value of RBI for the Ituri exceeds that from any study of “hodcold” beliefs, as expected by the central role food taboos play in ordering Ituri society. I also find that the values for different humoral belief systems vary with the degree to which the ethnographer sees such beliefs as being normative for that group. These results suggest that RBI is a robust measure of the reliability of individual responses. In effect, it is an indirect measure of the degree of normativity of a belief system. This is not surprising, since social support enforces consistency in normative beliefs and behavior. 126

MEASURING THE “STRENGTH OF BELIEF”

Given that many different methods of data elicitation were used in these food-related studies, I also conclude that the stability of responses depends not so much on ethnographic methods as on the strength of the cultural system being addressed. This, in turn, is a function of the degree to which beliefs are socially sanctioned. As in the Ituri Forest, norms do not necessarily insist that everyone’s beliefs be the same; only that an individual’s belief is the expected one for him or her. The RBI provides a baseline against which one can compare not only the degree of cultural variability, but also the normative nature of belief in different cultural groups. This ability to assess how strongly moralized a belief system is turns out to be a useful side effect of RBI. I have thus presented a simple measure that indicates the degree to which informant responses are likely to reflect the operation of a strongly held belief system. This measure is comparable between belief systems and cultural groups. It has the additional virtue of being based strictly on the rate of intra-informant error, which is available for a number of existing anthropological studies. The value of this index was shown to be higher for a taboo-based food avoidance system than for a variety of humoral belief systems, probably because the former system is maintained by stronger social sanctions. This simple measure cannot discriminate between the various sources of variability in responses-a task that requires more sophisticated attention to the context of data collection (which the Reflexive Analytical approach provides). However, RBI has the virtue of requiring only that interviews be repeated with a sample of informants. It nevertheless remains likely that there are cultural domains where high degrees of reliability simply aren’t achievable, regardless of what data collection method is used, due either to the cognitive difficulty of the task set the informant, or cultural sensitivity, or some other peculiarity. So while the reliability method advocated here cannot be considered a panacea for all problems relating to the interpretation and translation of meaning between cultural systems, it does provide a method for the validation of primary ethnographic data. Without knowledge of the quality of one’s primary data, the utility of any analysis using those data is considerably compromised andor restricted. Comparative procedures such as that suggested here are necessary first steps toward the resolution of current debates concerning the epistemological and ontological status of our knowledge of other cultures. 127

CHAPTER 6

Notes 1. The Reflexive Analytical approach is similar in purpose to the famous “multitraitmultimethod” (hereafter MTMM) approach of Donald Campbell and Donald Fiske (1959). In both cases, the basic idea is to separate methodological from true variability in psychological measures by comparing different methods of assessment. However, the MTMM approach is concerned with relationships between different measures that are related to the unmeasured or latent variable(s) of true interest. The Reflexive Analytical approach, on the other hand, emphasizes the effects of different factors on a particular variable of substantive interest, considering this true variable to be observed but indeterminate. Other considerations make the approach advocated here an appealing alternative to MTMM. To overcome statistical limitations in the original formulation (which consists of a search for patterns in a matrix of correlations), confirmatory factor analysis has become the standard implementation of the MTMM approach in recent years. However, Davis Kenny and Deborah Kashy (1992) have shown that this multivariate approach itself has undesirable statistical properties. In particular, the complete factorial model (which includes trait-trait, trait-method, and method-method correlations) is under-identified. As a result, while restricted or reduced models can be estimated, in most cases the MTMM method does not support a detailed analysis of the sources of variation in data. H. W. Marsh (1989) reaches a similar conclusion. The Reflexive Analytical approach is roughly equivalent to the complete M T M M model, providing a sophisticated breakdown of variability due to methodological, substantive, and random factors. However, unlike MTMM, it makes use of standard, wellbehaved statistical techniques. As a result, it might be preferred over MTMM, at least in those situations where the variables of interest can be considered measurable. 2. Alex Stewart (1998) argues that reliability cannot be a goal of ethnography, because ethnography cannot claim consistency, one of the dimensions of reliability in conventional research. Consistency is the application of the same measurement and getting the same outcome. This doesn’t happen in ethnography, of course, because dynamics internal to the social or even purely ecological situation have changed since the last attempt at measurement of some phenomenon in the field. But Stewart, I suggest, misunderstands the concept of reliability-which, as I have argued, does not require a same inpudsame output relation, but rather a foundation on which to investigate the causal influence of specifiable factors on any differences that arise in measurement. 3. (.76 - .007874)/.007874 = 95.52/127=.752. 4. Tzintzuntzan: (.8- .5)/.5= .6/2=.3; Oaxaca: (.882- .5)/.5=.76/2=.38. 5. The coding includes 13, rather than 12, categories as in other analyses because here we are also counting the possibility of no avoidance. 6. (.888-.0769)/.0769= 10.547/13=.81. A potential problem with this comparison is that the majority of questions (referring to animals) were not associated with an avoidance in either interview. The category that would be equivalent to this “no avoidance” response for humoral systems, a “neutral valence” for a food, is typically not considered by researchers on humoral systems. However, even if I allow that three potential responses were possible in the humoral studies cited above (i.e., “hot,” “cold,” and “neutral))),the bias index for Foster’s study becomes 1.42/3 = .47, while for Mathews the value becomes 1.6713 = S6. Despite these adjustments, the index for the Ituri study remains larger than those for the humoral systems. 128

MEASURING THE “STRENGTH OF BELIEF”

7. With thirty-six different types of response, the probability of a repeated response is only 1/36, or 2.78%,assuming no bias. The overall rate of discrepant responses was 42.74% of all repeated responses. The index of bias toward a repeated response is therefore = .54 [(.5726- .0278)/.0278 = 19.597/36=.544]. 8. That food avoidancesin the Ituri are stored in long-term memory is suggested by the results in appendix B showing few effects of increasing amounts of elapsed time between repeated interviews on the probability that a specific answer would not be repeated. This is most likely due to the dangerousness of forgetting such beliefs.

129

7

TOWARD A REFLEXIVE ETHNOGRAPHIC SCIENCE

Once we have agreed that cultural variation is primary, the paramount challenge is one of elucidating with greater clarity and sophistication the character of these variations. -Kenneth Gergen, The Saturated SeZ$ Dilemmas of Identity in Contemporary Lfe

T

here has recently been considerable controversy in cultural anthropology concerning the status of ethnography as science. Textual deconstructionists argue that the ethnography, as a written document, represents a form of knowledge that is unknowably transmuted by the process of cross-cultural translation of source materials, as well as by the power relations between the anthropologist and his or her informants. These debates about the authenticity of reports do not exist in other sciences, and only arise because of the general unreliability of ethnographic data. This unreliability is due to a variety of standard ethnographic practices, such as participant observation and the practice of using only very small numbers of people as informants to investigate a particular domain of interest. Some researchers have despaired of any solution to the problem of representing foreign cultures in the categories and style of presentation traditional to Western ethnography (e.g., Tyler 1986). Rather than the dispassionate scientist, reporting from behind the scenes with an allknowing, third-person voice, we have instead the heroic, socially engaged author and passionate critic of the status quo, or even just a frustrated artist seeking to dispense aesthetic experiences about exotic locales. From the ashes left by Geertz’s (1988) critique of both traditional and this reformist ethnography,a new species of cultural anthropology must be created. I suggest that new models for both the purpose and method of cultural anthropology can provide what at least some see as a desideratum: 130

TOWARD A REFLEXIVE ETHNOGRAPHIC SCIENCE

a study of cultural belief that cannot be relegated to an art form, but rather can stand together with scientific investigations of other kinds of phenomena. Through an examination of misconceived assumptions underlying the historical debate, I have shown in this book that the death of scientific cultural anthropology has been prematurely announced. I agree with recent critics (e.g., Clifford and Marcus 1986) that it is necessary for any analysis of ethnographic data to acknowledge explicitly the nature of the data on which it is based. Inferences of the anthropologist do not have the same ontological status as informant responses to questions in the field, nor do verbal reports of behavior have the same ontological status as observed behavior. These differences must be accounted for when constructing ethnographic reports. In other words, we need to import reflexivity into ethnographic practice. Wendy Ashmore (1989,234) calls reflexivity the “monster: the abyss, the specter, the infinite regress,” due to the problem of constantly trying to contextualize contexts. Ashmore (1989, 88) urges us to “celebrate the monster.” My goal, instead, is to tame it. In this book, I have developed an approach to the analysis of primary data based on the concept of ethnographic data collection as a situated event involving two participants (the observer and observed) engaged in a special form of social interaction. This view suggests that any reflexive method of analysis must consider the influence of the observer and circumstantial aspects of the social situation itself, as influences on what informants do or say during observations. Generally speaking, people vary in their responses to questions or when observed in social situations. Typically, some of this variability is due to the identity and characteristics of the data collector. This proportion of the observed variation can be estimated by comparing the behavior of a single informant to repeated tests of a particular elicitation process using different observers. Some variation in informant behavior is also due to differences in what informants themselves believe about or do in the world. This is variation that correlates with specified characteristics of the informants themselves, such as features of their life history and social roles. Another part of the observed variation in informant behavior is due to specified contextual factors such as environmental conditions. This proportion of variation in data can be estimated by comparing the behavior of the same individuals under different test conditions, holding the observer constant. Finally, the variation that is left over after controlling for all of these effects can be interpreted as having been 131

CHAPTER 7

caused by informant-specific inputs, such as their true beliefs. Ethnographic research based on identifj4ng these various influences on ethnographic data can be called “reflexively analytical” in nature. A virtue of such an approach is its ability to tease apart these various influences on informant behaviors, and to isolate what belongs to the informant, the observer, and the situation, in each observed event. The research approach proposed in this book should bring considerable illumination to what Sperber (1985,64) and Scott Atran (1990, ix) consider the great question in anthropology: the relationship between human nature and cultural variability, or more particularly, how the study of cultural variation can produce a better understanding of universal mental structures (Sperber 1985,88). The basic quest is to link anthropology more explicitly to its foundation in psychology. Anthropology, as currently practiced, is “a puzzle. ... [For example,] most introductory courses and texts begin by proclaiming the psychic unity of humankind, only to proceed straight to the study of cultural variations” (Atran 1990, k).There is no connection of culture to cognition. As Boyer (1994) points out, the problem has been the ethnographic tradition of describing cultural groups using ideal knowledge systems. Anthropological portrayals of cultural belief systems do not describe thoughts that occur to actual people; they describe thoughts that people might entertain, in the anthropologist’s view, if they wanted to make sense of what they actually do and say. . . . [Nevertheless, anthropologists treat] such constructs as direct, literal descriptions of people’s mental representations,which of course leads to rather extravagant interpretations. . . . [For example,] ritual statements, which people take as counterintuitive and which demand attention precisely because of their counterintuitive quality, are thus described as though they formed the basis of people’s ordinary apprehension of natural and social phenomena. (Boyer 1994,Sl) Thus, “collective representations,” “worldviews,” and “[primitive] modes of thought’’ have to be considered as scholastic concoctions rather than as descriptions of psychological realities. Only individuals can have knowledge because only individuals have minds. And what people actually know does not necessarily exhibit the qualities of idealized systems, such as logical consistency or completeness. To construct such representations (e.g., in ethnography) involves filling in what a number of informants report with implications that may not depict what any collection of individ132

TOWARD A REFLEXIVE ETHNOGRAPHIC SCIENCE

uals actually knows about their culture. Indeed, considerable effort has been expended in symbolic anthropology explicating sophisticated representations of culture that exist only in the anthropologist’s mind. This was acceptable because the anthropologist was assumed to be describing ways of life that characterize cultural groups as wholes-no single individual could be expected to hold such a corpus of knowledge in memory. However, if explanations of cultural beliefs are restricted to the level of groups, then it could be argued that environment is the sole determinant of cultural variability: different groups simply live in distinct circumstances that influence what is culturally marked. One need not even consider cultural history as a causal factor. Once it is admitted that intra-cultural variation exists, however, then other types of explanatory factors-such as cognitive structure and social differentiation-must be taken into account, since environment and cultural history can be assumed to be constant for all members of a group. In particular, once the study of cultural belief is grounded in the psychology of individuals-a position forced upon the researcher by a recognition of the computability and transmissibility constraints on learning culture-the connection between cognition and culture becomes obvious: culture is composed of socially learned mental representations. I suggest that if a belief is learned from others, then it is cultural; if it is invented or inferred from individual experience, it is not-at least until it is imparted to others (Boyd and Richerson 1985, 33; see also Swartz 1982,316; 1991,7).’ A realist approach such as Reflexive Analysis is thus concerned with how culture is learned and, once learned, transmitted between individuals (Aunger 2000; Hewlett and Cavalli-Sforza 1986) or between groups (Barth 1987; Soltis, Boyd, and Richerson, 1995; Guglielmino et al. 1995). By virtue of being learned socially, such knowledge is cognitively distinct and hence relatively easy to recognize: only socially acquired beliefs can be what Sperber (1985, 51) calls “semipropositional.” Due to incomplete information about causal antecedents and consequences, semi-propositional beliefs are consistent with a variety of interpretations, rather than a single, precise referent. Thus, Claudia Strauss and Naomi Quinn (1994,293) are wrong to argue that “cultural schemas differ not at all from other schemas except in being shared”; they can be inferentially limited in ways that individually learned schemas are not. Thus cultural knowledge can be distinguished from other sorts of knowledge by its source in the social world, and cognitively by its lack of inferential robustness. 133

CHAPTER 7

Further, what makes human culture unique is its ability to accumulate knowledge through generations (Tomasello, Kruger, and Ratner 1993). That function is crucially dependent on the mode of information acquisition. It is the transmission of a belief, not its frequency within a population, or its status as a norm, that makes it c u l t u d 2 This conclusion simply reflects the situation on the ground. No one is an expert in all aspects of social knowledge; even in relatively simple societies, there is social role differentiation (Roberts 1951). Nevertheless, individuals are able to communicate with one another, coordinate their social behaviors, and reproduce their cultural beliefs and values through succeeding generations. Uniformity of cultural belief is not required for a society to function. Significant consensus on certain rules of behavior is required: language sharing (for communication); concepts of justice and social contractual obligations-behavioral codes going under the rubric of “morality.”I would argue these are systems of meta-rules, or rules for setting rules. However, in any society that exhibits social role variation, a society does not, and indeed cannot, impose uniformity of substantive belief. In the Ituri Forest, we have seen that this is manifest as considerable variability in what individuals believe is edible for them, but also in widespread recognition of the power relations between individuals, and in particular the authority of parents to transmit taboos to them. It is therefore vital that cultural anthropology begin to deal with cultural variation. But while I agree with Kenneth Gergen (1989) that the objective of cultural anthropology should be to gain an understanding of the mental schemas of others, I disagree with him that this requires establishing an intersubjectivity that is an emergent ontological product of the social interaction itself, somehow above and beyond the awareness of the participants themselves. Rather, I suggest that cultural anthropology can reliably infer neither individual mental states from public manifestations of mentality (such as spoken phrases or behavior) nor any such metaphysical intersubjective state as Gergen proposes. With the Reflexive Analytical approach, an ethnographer can no longer report a particular individual’s belief or behavior as “being X.” Instead, the observer must recognize that he or she has only elicited a datum that represents an expression of belief X or behavior that means X. The informant in question can only be said to hold this belief or intend this meaning with some probability Y , Y itself being some function of the methods used and the social situation in which the datum was generated. Thus, the Reflexive Analytical ap134

TOWARD A REFLEXIVE ETHNOGRAPHIC SCIENCE

proach is concerned with patterns in data, with characterizing variation in a population-unlike idealism, which seeks to find the one truepoint, abstracted from that pattern, that can be used to represent the whole. So Reflexive Analysis shares the weakness with idealistic strategies like cultural consensus analysis (Romney, Weller, and Batchelder 1986) of not being able to deal directly with informants as individuals. They can only appear in the ethnographic report as the source of anecdotal material. However, this is simply an unavoidable constraint of scientific work, which requires repeatability; individual data elicitations cannot be repeated. Through reflexive analysis, the loss of certainty about individuals is ironically coupled with increased certainty about the population-level pattern. The virtue of this tradeoff, therefore, is that ethnographic authority is reestablished. Further, a much richer and more varied picture of mental and social life can be presented to readers. This should have benefits in terms of credibility-people know that not everyone is the same and that human social life is incredibly complex. Attempting to simplify this complexity too much is, in the end, ineffective in these days of hypersensitivity to representational issues. My concern with population-level patterns of cultural sharedness does not imply that each cultural trait is universally held: variation in belief and behavior is endemic within cultural groups. Luckily, the analyses presented in this book conclusively show that describing variation is a robust strategy for ethnographic representation-more robust than any attempt to typologize beliefs in cultural groups. Analyses with the goal of variation-based representation are significantly less sensitive to the unavoidable deviations arising from social data collection and analysis, and should therefore be preferred. However, on the down side, this intensive approach can only be carried out with respect to some central focus of study; any ethnography must also rely on generalizations as background information covering other domains of life. For example, the use of repeated formal interviews cannot be carried over to every aspect of ethnographic investigation. So the reflexive analytical solution is necessarily limited. Any realism achieved will always be set within a typological context, and any ethnography must be a combination of both reflexive and standard descriptive elements. But does the adoption of a realist approach mean we can never do principled aggregation or cross-cultural comparison? Certainly this would be an unappealing conclusion and a major strike against realism if true. 135

CHAPTER 7

But fortunately we are no longer constrained to the traditional kind of cross-cultural work that has characterized cultural anthropology until recently: the bivariate cross-tabulation of societies (e.g., Ford 1967). Instead, we can do quantitative comparisons of structures with techniques such as loglinear modeling for cross-tabulated data (Agresti 1990), quadratic assignment for similarity matrices (Hubert 1987), or various scaling methods for categorical data (Weller and Romney 1990). This means each group can be represented not by a single value but by a uni- or multivariate distribution of values. Such comparisons should thus preserve much of the unique character of each group while being compared with others. Of course, there will always be situations in which one wishes to characterize groups using single values-for example, when data on distributions are not available, or when norms with respect to some domain of behavior are to be discussed. In such cases, it should simply be made clear what claims are being made about sharedness or the applicability of such characterizations to particular individuals. Thus, Max Weber’s (1971) claim that idealization is forced on the researcher due to the myriad manifestations of cultural traits in individual minds no longer has power: this variation can now be treated analytically. The great benefit of idealization claimed by Weber-that it permits a succinct and persuasive representation-also no longer holds in the face of the postmodern critique of such representations (e.g., Clifford 1988; Clifford and Marcus 1986). Traditional practice privileges the outsiderethnographer’s perspective, to the exclusion of native voices, which presumably are more legitimate or truthful. Most recent critics of ethnography, recognizing there is a great deal of intra-cultural variability in beliefs and values, have argued that new ethnographies must recognize this variability explicitly by incorporating multiple voices and viewpoints within the document itself. While the inclusion of relatively undigested ethnographic data in ethnographic reports is useful, it puts a considerable burden of interpretation on the readers of that ethnography. Perhaps more importantly, it also limits its usefulness, since this practice makes it difficult to compare ethnographies that are produced using idiosyncratic organization and experimental presentation styles. The textualist concern with ethnographic representation was part of the reason that their critique appeared to lead to an analytical impasse. The implicit goal of both traditional and textualist ethnography, to depict cultures, requires that a culture be a self-consistent, integrated whole ca136

TOWARD A REFLEXIVE ETHNOGRAPHIC SCIENCE

pable of being presented in some unidimensional form. But as Barth (1993,4) argues, “[Wle must break loose from our root metaphor of society as a system of articulated parts. The image is too simple, and it misleads,” because society is not an organic body with definite boundaries and a particular nature. Rather, it is the result of multifarious agents’performance of specific roles, each with unique competencies and different strategies. From these considerations,Barth (1993,4) concludes, “[Tlhe image of processes serves us better than that of a structure or a closed system.” Handwerker (2002, 110), recognizing that cultural groups represent a moving target for analysis, argues more precisely that cultural systems evolve. This processual orientation, I argue, must be the starting position for any ethnography aiming at scientific status. (I would only counsel Handwerker and Barth that reflexivity in practice remains a prerequisite for anyone recognizing the need to characterize cultural variationvariation is, after all, the substance of an evolutionary process.) How then would a scientific ethnographic analysis be conducted?The changes a Reflexive Analytical approach would require in ethnography are similar in nature to the changes wrought by the Darwinian revolution in biology. At least according to Mayr (1982), Darwin’s contribution consisted most importantly in the abandonment of the Platonic notions of metaphysical ideals and the imperfect nature of ontological entities in favor of an anti-metaphysical postulation where categories are abstract and individuality is ontologically real and valuable. A hndamental reorientation of thought w ill be required for the metaphysical position of cultural anthropology to come into line with other sciences. If cultural anthropologists would also consider that variability in cultural traits is the thing to be explained by ethnographic investigations, the “multiple voices” lauded by interpretive anthropologists will be truly heard. Whereas an idealistic approach seeks to find a set of points, or a structure among a single set of points (a cognitive/social mind), to stand in for the variation in belief within a group (and which therefore constitutes a symbolic representation), a realistic approach seeks to characterize who knows what and why-that is, to determine the social distribution of belief and its causes. Representational issues ensue when the “who” is no longer individuals, but categories of individuals (e.g., social role-based classes). There are also cognitive structural issues (which an idealistic approach ignores in its abstraction to a group mind): How are various beliefs related to one another by that individual and what kinds of individuals have a particular cognitive 137

CHAPTER 7

architecture for the given cultural domain? In this sense, no single structure is considered sufficient to represent the group. One general realist approach to the study of cultural belief systems is “cultural epidemiology,” outlined in Sperber (1996). This constitutes a fairly fundamental reorientation of the cultural anthropological project. Whereas the traditional ethnographic enterprise has been interpretive (e.g., to depict cultures in a static, idealized fashion), the quest of cultural epidemiology is explanatory: to uncover how social and cognitive mechanisms work over time to produce the distribution of cultural beliefs both within and between cultural groups. Its fundamental premise is that culture consists of meaningful units of information that are duplicated during transmission between individual minds. Transmission occurs either through imitation of the behaviors inspired by beliefs or by the communication of signals related to belief content. Beliefs (mental representations) are then transformed into observable phenomena such as rituals or expressed opinions (public representations),just as a genotype determines its respective phenotype. The explanatory goals of cultural epidemiology are to track the life history of these representations as they metamorphose from one form to the other, to understand the psychology of choice among competing beliefs, and to uncover the forces that influence social access to these representations.Together, these factors determine the distribution of beliefs across individuals. Since both mental and public representations have material manifestations (e.g., as sound waves in the case of speech-as-belief-expression, and brain states in the case of mental representations), cultural epidemiology is M y materialistic (Sperber 1996, 26). Implementing a program of cultural epidemiology would require clean data about the pattern of variation in a population-in other words, it would require that the ethnographer use an approach like Reflexive Analysis to clean the information on intra- or inter-cultural variation in the first place. The similarity of cultural epidemiology to biological evolution is not coincidental. Both cognition and cultural transmission can be studied from a Darwinian viewpoint. Decision making among alternative beliefs is the cultural analog to natural selection since it leads to the differential replication of beliefs within groups (Boyd and Richerson 1985),while the origins of these cognitive biases can be explained as the result of the evolution of the brain as an information-processing device (Tooby and Cosmides 1992). Cultural epidemiology, dthough stretching across both 138

TOWARD A REFLEXIVE ETHNOGRAPHIC SCIENCE

psychological and social scientific levels of explanation, is united under a single theoretical umbrella. “A central part of a theory of natural selection-functional adaptation-is millennia old, universal, and easily grasped by young pre-school children, whereas natural selection [as the differential reproduction of genetic variants in a population] seems to have emerged only when Darwin and Wallace abandoned strongly held ideas of species having essences” (Keil 1995, 266). Cultural anthropology remains stuck in a pre-Darwinian state, with a functionalisdstructuralist underpinning. But the same liberation must now occur for culture as has already transpired for other sciences. Just as biology as a science separated from folk biology (Atran 1990), the study of culture must abandon the folk idea that cultural groups have essences. The recent round of self-reflection about ethnographic practice has had the virtue of increasing awareness not only about the rhetorical but also about the ethical nature of ethnographic research. An important point made by recent ethnographic critiques is that some individuals tend to be excluded from active participation in culture formation (Carspecken and Apple 1992). Concern with consensus can thus obscure underlying inequalities in social power. Johnson and Grifith (1996), in discussing their finding of underlying variation within an American consensus, quote Roger Keesing’s (1987,166) argument that consensus values may be shared (at least in surface observance) even though they sustain the interests of some and work against the interests of others. We must . . . dig beneath surface consensualityto seek counterideologiesand cultural expression of subaltern struggle. The overlay of consensuality, viewed uncritically,can make an anthropology of meaning insidious as well as politically naive.

This perspective suggests that the interesting question is not whether consensus exists but, who makes consensus and how do social elites influence mass opinion? Such questions are difficult to answer authoritativelyexcept through some kind of repeatable, transparent procedure. This is, of course, just what the Reflexive Analytical approach provides. A number of critical ethnographers (e.g., Fabian 1983,1991; Rosaldo 1989; Said 1978,1989) are also concerned with the intrinsic differences in power between ethnographers and their subjects. Fabian (1991, 193-94), for example, believes that “hanging the walls f dlwith reflexive mirrors 139

CHAPTER 7

may brighten the place, but offers no way out [of the ethnographer’s dilemma of being] . . . stuck in the dirt of politics or the mire of epistemological quandaries.” Some react to this problem by arguing that anthropology must become “emancipatory”or “participatory”by humanizing the conditions of life in the cultural group being studied (Burawoy et al. 1992; Fals-Borda and Rahman 1991; Huizer 1979; Scholte 1974; Stull and Schensull987; Torbert 1991; van Willigen, Rylko-Bauer, and McE1roy 1989;White 1991;Wulff and Fiske 1987).This can involve returning the knowledge gained through research to the community from which it was derived (rather than merely enhancing the Western academic’s reputation) or, more radical, making research itself a form of action designed to better conditions in the community. Chris Argyris, Robert Putnam, and Diana Smith (1989) and William Torbert (1991) emphasize the need for the animatorhesearcher to acquire the necessary self-reflexive, interpersonal, and sociopolitical skills before attempting to lead others into the development of novel social organizational forms. However, “it is impossible to create a research process that erases the contradictions (in power and consciousness) between researcher and researched” (Acker, Barry, and Esseveld 1991, 150). Nevertheless, most liberationists (e.g., Fabian 1991, 195;Wolf 1992,6) do not advocate that Westerners stop conducting fieldwork. What is the conscientious ethnographer to do? In fact, this dilemma cannot be minimized because the politicaleconomic effects suggested by these critical ethnographers do exist. We saw earlier that informants in the Democratic Republic of the Congo bias their responses in a number of specific ways when the interrogator is a powerful interloper into their social circles rather than a local individual. However, I would argue that merely by providing evidence of the political/ epistemological factors intrinsic to the ethnographic situation, the Reflexive Analytical approach represents a partial solution to the liberationist dilemma. Thus, to find out what effect being an expatriate representative of a colonialist power has on what informants say or do in one’s presence, one can rigorously compare their behavior with persons of their own culture under similar interviewing circumstances. This does not represent a complete solution, of course, because it remains to do something about the imbalance of power in ethnographic research. As a first step, selection of research questions can certainly be tied to issues decided by the native community rather than by Western intellectual fashion. Further, the results above suggest that the best way to get 140

TOWARD A REFLEXIVE ETHNOGRAPHIC SCIENCE

high-quality ethnographic data is to turn data collection over to those who are enculturated in the cultural group of interest. Of course, the research will probably not be done without the direction and monies of the ethnographer-a point recognized by feminist ethnographers, who argue that since the researcher controls the terms of interaction and the framing of the study, a power differential exists between ethnographer and informant even if women are writing about women. As a result, they argue that one must adopt in one’s research a political agenda explicitly directed at overcoming the forces of oppression (Mascia-Lees, Sharpe, and Cohen 1993, 246). Perhaps an intimate collaboration between ethnographers and local individuals, from research design and execution to the interpretation of results, would provide the best, if still imperfect, s ~ l u t i o nEven .~ so, concern with social power is absolutely necessary for a more complete understanding of the situational and historical context of ethnographic fieldwork. However, the existence of power differentials even at the interpersonal level does not preclude a scientific analysis of social phenomena; rather, it precludes exclusive reliance on researchers’ personal sentiments and observations, which must necessarily be biased by their positions in social networks. The Reflexive Analytical approach does not dehumanize the subjects of ethnographic research, as some critics argue is true of reductionistic scientific methods (e.g., Tyler 1991). O n the contrary, because of its situationalism, the Reflexive Analytical approach views individuals as strategizing agents. It does not gloss over differences in motivation, belief, behavior, or life experience but admits this variability directly into the analysis, thereby granting a considerable degree of personhood to those living in other cultures. Since moral systems are sometimes judged by their tolerance of the deviant, I argue that this treatment of variation in the two-step approach is humane. I therefore conclude that scientific analysis and evaluation is not a valueless, morally odious, or impossible goal for ethnography. In fact, I maintain that formal methods provide a rigorous means of addressing many of the issues considered important by textualist anthropologists. In general, as noted by Clifford Behrens (1990, 325-26), statisticians have been extremely creative in their development of methods to deal with types of problems previously treated in a purely interpretive manner: “Take for instance the impact of multidimensional scaling on the measurement and representation of complex cultural constructs. W h o in the 40’s or 50’s 141

CHAPTER 7

might have thought it possible to derive a statistical representation of emotions or kin relations, concepts of illness, or a culture’s food classification? . . . [New techniques] have allowed us to quantify what were once thought of as strictly qualitative phenomena.” The continuing development of formal analytical methods suggests that the approach outlined here can serve as a general framework for conducting replicable ethnographic research on the full range of traditional anthropological topics, from social interaction patterns to marriage rules and religious belief^.^ Let me conclude by noting that several important implications of the analyses here have not been previously mentioned. First, the debates regarding reflexivity in the social sciences have overemphasized epistemological and purely methodological concerns. Assumptions about the nature of reality (ontology) made by researchers have been shown here to have more significant consequences (a claim repeatedly made by the new structuralists as well [e.g., Outhwaite 1987, Layder 1990, Giddens 1991, Bhaskar 19791). For example, despite an objectivist epistemology, microeconomics has made considerable progress in describing individual decisions regarding resource allocation, indicating that having the correct epistemology is less important in implementing a theory than having the right ontological categories. In particular, the choice of ontology has been shown to determine an approach‘s notion of causality, units of analysis, logical form, temporal framework, and quality assessment criteria. Sperber (1985) and Boyer (1994) have emphasized that granting ontological reality to inappropriately characterized concepts (i.e., those with an academic but not a natural history) can lead research programs to become moribund: a science that does not “carve nature at its joints” will be more debilitated than one that simply makes relatively large observation errors. Powerful explanations deal with “natural kinds,” entities that have causal force in the world. Unfortunately, there is no scientific method for discovering what categories of things in the world constitute natural kinds. Progressive research programs, it seems, can only be discerned by their results. In any case, the battle over epistemology has been won: it is now widely agreed that positivism-or at least its objectivist epistemology-is dead (Rosenberg 1988; Bohman 1991; Outhwaite 1987). However, debates about ontology continue to rage. For example, an important current theoretical controversy in sociology is the so-called “agency/structure debate” (see Archer 1988; Bourdieu 1977,1984; Giddens 1979,1984). This 142

TOWARD A REFLEXIVE ETHNOGRAPHIC SCIENCE

is the latest incarnation of the ancient rivalry between partisans of intentionality and those who argue that human action is significantly constrained by social institutions. Perhaps it is time for ethnographers to follow the example of their sociological cousins. We should turn our attention from solipsistic concerns with texts to substantive questions about what variation in cultural practices can tell us about the human condition. I therefore believe that perhaps the most important benefit of the Reflexive Analytical approach is that it forces the ethnographer to focus on intra-cultural variation. A powerful objection to idealism is that it can too easily lead to a denial of the metaphysical and ethical preeminence of the individual. As Weber notes, the great temptation is to treat ideals as real; indeed, to “do violence to reality in order to prove the real validity of the construct (1971, 507).”5This is because idealism is always normative (at least implicitly), as suggested by the use of such language as ucorrect”or “true” belief (e.g., Romney, Weller, and Batchelder 1986; Romney et al. 1996,4704). The notion of informant “competence,”also associated with idealistic methods, implies a hierarchy of value-in particular, a gradually increasing degree of approximation to the ideal informant, who represents what everyone in the group should be striving for, because competence is correlated with intelligence, social status, reliability, and so on. However, the devaluation of individuals through comparison to a fabricated ideal is contrary to a democratic humanism. Since nationalism and xenophobia rely upon the objectification and normative idealization of one’s own group in comparison to others, I believe it is a moral imperative that individuals be considered real and their minds unique. Each should be treated with respect and valued for his or her diversity of experience and opinion. Indeed, the survival of humankind as a species probably depends upon the maintenance of, and development of tolerance for, both intra- and intercultural diversity.

Notes 1. The definition of culture as similarity can cause CCA researchers to make peculiar statements. For example, Boster (1991,223), in surveying ethnofaunal classification studies, argues that “culturally diverse groups of informants can converge on a single consensus; they can agree (share cuZture [emphasis in original]) without the benefit of social information transmission. It is ironic that the cultural consensus model may work best when . . , individuals agree by virtue of their independent insights into the task.” In fact, in this case, the cross-cultural similarity is due to universal cognitive mechanisms for the

143

CHAPTER 7 perception of animate objects in nature, as Boster himself has argued (Boster and D’Andrade 1989). Thus, in effect, Boster here suggests that CCA is most appropriate for domains that are significantly genetic in origin! Defining culture as socially learned information would guarantee that culture remained a social scientific concept rather than a biological one. 2. To be fair, there have been recent shifts among anthropologists away from an idealist conception of culture (see Borofsky 1994). Indeed, D’Andrade (1995,216) has come to the conclusion that there are two kinds of cultural domains: consensual, sanctioned domains created by “the need to communicate effectively and share expertise,” and lessshared knowledge systems marking subgroups as somehow distinct. Now “the issue is not ‘how shared is culture,’ but rather how to understand both distributed and high consensus aspects of cultural knowledge” (D’Andrade 1995, 216). Douglas Caulkins is even using CCA to explore cultural diversity, rather than consensus (see, e.g., Caulkins and Hyatt 1999;Trosset and Caulkins 2001). 3. Various forms of such collaboration have been utilized by a variety of ethnographers (e.g., Abu-Lughod 1992; Behar 1993; Bernard and Pedraza 1989; Chapman 1992; Crapanzano 1980; Davison and Women of Mutira 1989;D y e r 1982). However, the work of each participandauthor in the ethnographic report, although often contextualized inside framing devices, remains fragmentary and thus is not validated by an overarching analytical framework. 4. As Russ Bernard points out to me, this effectively means that there is no crisis in ethnography at all for positivists, who feel they can simply go about the business of doing ethnography in standard fashion using the latest scientific methods. 5. A clear example of this comes from the CCA literature. Romney et al. (1996,4704), in discussing high-consensus results concerning kinship terminology, claim that “the cultural [i.e., consensus] definition [of a domain by CCA and scaling techniques] is a better estimate of what is in the mind of the subject than an estimate of a cognitive representation based on the subject’s own responses. This is because of the vastly increased reliability of aggregate measures compared with single measures.” Thus, a group-level representation of the individual should be substituted for the individual’s one, based on reduced error in measuring the group-level construct. The informant’s differences from that construct are implicitly considered to be due strictly to methodological causes. This is because the informant’s responses are just a “test,” whereas the group-level statistical construction represents the “truth (Romney et al. 1996,4701). As Romney et al. (1996,4704) note, this use of “single pictures based on aggregate data” is common practice in psychological research. But as Boyer (1994) has argued, it is a fundamental fallacy to assume that beliefs or values characteristic of a group actually exist in the minds of individual members of that group.

144

APPENDIX A

THE CASE STUDY FOOD TABOOS IN THE ITURI

T

his appendix includes information about the collection, characterization, and analysis of the ethnographic data on which all of the empirical analyses in the book are based. The study concerns foodrelated beliefs in a multiethnic population living in the Ituri Forest of the Democratic Republic of Congo. I begin with some background information about the people under study.

The Study Population Ethnic groups in the Ituri can be divided into several categories, based on whether they are pygmy foragers or slash-and-burn horticulturalists of either Sudanic or Bantu origin. The farmer villages consist of wattle-anddaub houses, and range in size from fourteen to one hundred individuals, consisting of a small number of extended families divided into shallow patrilineages. These groups (Lese as Sudanic horticulturalists; Budu as their Bantu-speaking neighbors) are polygamous, but practice mostly serial monogamy, with both sexes often taking a number of spouses. Sister exchange between clans is the ideal marriage model. Post-marital residence is typically virilocal. Their gardens are typically less than one hectare in size; primary crops are manioc, rice, corn, and peanuts. The foragers, in contrast, live in mobile camps of less than thirty people, which are close to horticulturalist villages for about half the year; during the other half, they live farther in the forest to take advantage of seasonal forest produce. The Efe (Sudanic-speakingforagers) and Batswa (Bantu-speaking compatriots) subsist through gathering and archerybased hunting, although a significant proportion of captured meat is traded to the Lese for garden produce (for more background on these 145

APPENDIX A

groups, see Bailey 1991; Bailey and DeVore 1989; Grinker 1994). Foragers hunt the wild animals of the forest with bow-and-arrow and exchange some of the resulting meat with the horticulturalists, who give them domesticated plants from their gardens. Often, there is no effective road transportation through the area, or access to an active market economy, so the population is quite isolated, both culturally and socially. Education is minimal, due to only sporadic, missionary-based schooling. As a result, most social information is transmitted during face-to-face interactions.

Types ofTaboo Nutrition is at least seasonally inadequate in the Ituri, where individuals annually go through a hunger period that can result in significant weight loss (Bailey et al. 1993). Meat in particular is quite scarce even though a wide variety of animals live in the Ituri. Nevertheless, individuals in the Ituri cannot eat many animals due to food avoidances. The reasons for these food avoidances range from personal attitudes to culturally sanctioned taboos. The basic distinction is between taboos and other kinds of avoidances. Taboos are rules whose violation is followed automatically by some (supernatural) punishment; avoidances are not socially sanctioned and have less severe consequences. (Taboos are thus distinguished from the more general class of avoidances, which includes purely personal or experiential beliefs about the edibility of foods.) Taboos are typically complicated rules that specify the conditions under which meat should be rejected, and the dangerous consequences that befall the person who violates the taboo (e.g., contraction of a terrible illness). Individuals usually learn these beliefs through verbal transmission from others. Many taboos surround pregnancy and childbirth. An individual may have to follow taboos against more than sixty edible animals during the course of his or her lifetime. Although violation of these taboos is believed to produce consequences ranging from a quick death to a variety of debilitating diseases, a system of ethnomedical treatments called dawa, typically involving consumption of specific plants together with the tabooed meat, can be used in some cases to prevent the onset of these consequences. Taboos can be divided into categories, ranging from beliefs involving witchcraft to more banal restrictions, such as “Men don’t eat siobo [house 146

THE CASE STUDY FOOD TABOOS IN THE ITURI

mouse] because it is too small.” “Family-based”taboos, a major category of taboos in the Ituri, are beliefs that are quite dangerous to violate since the consequences include the death of the consumer, hidher child or other relatives, or the violation of a relationship between the clan and supernatural entities. Some of these taboos are sex-based. Women, for example, typically must forego carnivores because their meat is considered too strong for women’s weaker physiology to withstand. Another major category of taboo surrounds pregnancy and tends to involve the homeopathic principle of contagious similarity between an item consumed by the mother and anomalous characteristics acquired by the fetus, discovered when the mother delivers. For example, during pregnancy a woman may have to refuse a certain type of forest antelope because its skin is “red,” and will cause her to bleed profusely during childbirth. A final major category concerns avoidances of foods that are not supported by social sanctions, but that reflect personal experiences with, or beliefs about, a particular food. These are often related to reactions of disgust. Others are purely idiosyncratic. For example, some individuals believe storks and egrets physically resemble the Italian nuns who go about the area in white habits. Considerable variation exists both within and between the different cultural groups in the Ituri concerning which animals are to be rejected under particular conditions. In fact, beliefs within each such grouping have been shown to be meaningfully different (Aunger 1994b, 1996). The present study exhibits the following characteristics: The sample of cultural beliefs was derived from structured interviews with nearly everyone in the relevant population-not just “expert”informants, or individuals of a particular age or sex class, but a 95% complete sample of males and females over the age of ten, with a total sample size of over 449 individuals. Data were collected in three separate locations having somewhat different compositions by ethnic group, but within a single cultural area. The research consisted of an in-depth investigation into a single cultural system or domain, since the dynamics of each cultural subsystem are likely to be different. 147

APPENDIX A

Data were collected explicitly with a view to investigating within-cultural variation and cultural transmission in order to describe the likely long-term evolution of the cultural system.

A significant proportion of the sample was interviewed twice using the same methodology (structured interview), with varying lengths of time between interviews. Native data collectors conducted interviews using identical methods (for comparison with data collected by me, a Western anthropologist), with an overlap of informants in my sample. Genealogical relationships between individuals in the sample are known, as well as other background information (e.g., degree of acculturation) to ascertain how these variables and life experiences correlate with the adoption of these cultural beliefs. Cognitive tests were also given to a subset of the sample in domains relevant to food avoidances to assist in determining the units of cultural variability. The practice of these beliefs-that is, information about individuals refusing opportunities to eat relevant foods-was determined by recording meat consumption (as reported during thrice-weekly household interviews) in eighteen horticulturalist villages for the period of a year; this allows cultural beliefs to be connected to actual behavior. Some data were collected in related tribal groups for a regional perspective on variability in cultural beliefs. These characteristics permit a wide variety of tests to be performed on the resulting data, particularly with respect to establishing their validity and reliability through isolating a variety of contextual factors influencing subject responses.

Interviewing Procedures Since my overarching purpose was to investigate within-cultural variability in belief, I determined to interview as large a proportion of the indi148

THE CASE STUDY FOOD TABOOS IN THE ITURI

viduals over the age of ten in the local population as possible. Since I conducted interviews in KiNgwana, a local variant of Swahili, the only real requirements of an informant were an ability to speak KiNgwana moderately well and a willingness to talk privately with a male anthropologist. My assistants interviewed many of those who did not fill one or the other of these conditions. A total of 449 different individuals, constituting an approximately 95% complete sample of individuals over the age of ten along a 16kilometer stretch of road, as well as several subsidiary samples from other locations for a regional perspective, were collected by myself or one of three local assistants, in the form of structured interviews during 1989-1990. This represents a relatively unbiased intensive sample of individuals living in geographic proximity. Each ethnic sample constitutes a high-proportion sample of individuals more than ten years of age from geographically distinct clusters of clans, and should thus be representative of these cultural groupings. The numbers of informants by ethnic group, sex, and age category are reported in table A.1. I began interviewing at the Ituri Project research station, an area dominated by Lese-Dese. The methods outlined here resulted from experience during a pilot study conducted in this area during 1987. During my first one hundred interviews, I found over two hundred foods to Table A. I.

Composition of lturi Sample*

Age Class Ethnic GrouplSex

9to I6

17to25

26to44

44+

Total

Sundanic female male

26 46

22 23

49 53

31 34

I28 I56

female male

3 8

6 14

12 19

8 12

29 53

female male

I -I

7

13

2

23

12 -

12 -

10 -

35 -

female

30 (7. I%) 55 (I3.0%)

74 (I7.5%) 84 (I9.8%)

41 (9.7%) 56 ( I 3.2%)

I80 (42.5%) 244 (57.5%)

Bantu

Forager

Total

male

35 (8.3%) 49 (I I.6%)

*Percentages are with respect to entire sample (N = 449).

149

APPENDIX A

be widely considered edible by the primarily Lese informants. Of these, I selected 140, representing all the major orders of animals. This list, comprising only a portion of the animals considered edible by the local people, included all of the mammals, most of the named varieties of fish and birds, and the most nutritionally significant animals of other groups, organized into emically legitimate groups in order to aid informant memory and provide contextual cues. To this list, I added five other items, either culturally significant plants or domesticated animals. It was this shorter list that I used for all later interviews: all results reported here depend only on responses with respect to this set of 145 foods. As I proceeded, I began to be aware of several problems with the initial research design. First, I came to believe there were language-based difficulties in the ability of some individuals, especially children, to recognize my pronunciation of Lese animal names. In addition, some individuals, for whatever reason (some, I believe, from shyness about their inability to speak KiNgwana), were unwilling to be interviewed by me. In order to ensure the representativeness of my sample as well as to get a better idea about the nature of the information I was gathering, I determined to employ several literate local individuals able to perform the same basic interview protocol in their native tongues. AU interviews conducted by my three assistants, K, M, and N, used the shorter list discussed ab0ve.l K, at the time, was a thirty-year-old Lese man (the Lese are a Sudanic horticulturalist group) who primarily interviewed Lese, but also some Efe (Sudanic-speaking foragers), in the Lese language; M, a twenty-seven-year-old Budu man (the Budu are a Bantuspeaking horticulturalist group), who interviewed Budu informants in Budu; and N, a twenty-nine-year-old Lese man (half-brother to K), who specialized in interviews of Efe, using the Lese language. K, M, and N were paid twice the informant’s fee for each interview they completed. In the end, the anthropologist interviewed primarily Lese informants, but also a few Efe, always in Swahili, which is a second language for both me and the local individuals. Thus, beliefs concerning the edibility of 145 different animal foods were elicited during formal interviews that a local assistant or I conducted privately with informants either in Swahili or their mother tongue. These questions include all animals that are consumed on a regular basis (i.e., at least once within a year).2Thus, in the present context, 150

THE CASE STUDY FOOD TABOOS IN THE ITUN

the cultural domain is assumed to be “beliefs about the edibility of culturally salient animal foods.” Within this set, each individual proved to have several animal-specific taboos that are different from anyone else in the population-even after controlling for methodological sources of variation in reported taboos (e.g., who conducted the interview with an informant-see Aunger 1994a). The food avoidance interview data thus consist of a sequence of responses to questions concerning any avoidances the informant might have with respect to particular wild animals. Up to three different avoidances, listed in the temporal order they were elicited, were recorded for each animal. (Several avoidances of a single animal are possible because the effective periods of certain taboos may not overlap, for example, during puberty ceremonies or during pregnancy; as a result, both may be applied to the same animal.) Interviews within an ethnic group were conducted by a single interviewer, with the exception of Sudanic horticulturalists, where two were used. All questions were posed in the same fashion, changing only the animal referred to (i.e., Can you eat X? If not, why not? Further probes clarified circumstances under which the food might be refused). During the interview, responses were recorded as phrases describing the constraints on consumption mentioned by the informant. Later, responses were converted into codes using a categorization scheme that groups certain kinds of response to reduce the dimensions of variation for purposes of analysis. Since my interest was in the elicitation of individual variability in beliefs, the interviews had to be structured in order to maximize the degree of comparability between interviews. To put the informant at ease, all interviews began with a series of questions that allowed them to talk about themselves. This set of background questions provided useful information concerning the informant’s genealogy, life history, and other matters. More specifically, these data include: the date of the interview; an identification code for the interviewer; names of the ethnic group and clan to which the informant reported affiliation; the informant’s year of birth (see Bailey 1991, 34-35 for a description of age estimation methods), sex, and number of years in school. In order to elicit as much of an informant’s avoidance-related knowledge of different animal foods as possible, I resolved to organize the interview around two complementary objects of questioning. Since food avoidances in the Ituri almost strictly concern animal foods, during the 151

APPENDIX A

first and longer part of the interview I prompted the informant with names of animals. By using animal names as prompts, the intent was to elicit as many consumption-related beliefs and practices involving that animal as possible. After the entire list of 145 animal names had been completed, the second part of the interview would begin, during which the informant would be prompted with a series of questions centered on a particular kind of avoidance rather than an animal name. For example, the informant would be asked, “Are there any animals you can’t eat because they are your njou [embodiment of ancestor]?” This change in perspective to avoidancespecific questioning, would hrther spur the recall of otherwise hard-toremember avoidances. Because I was interested in consistency of interview contextualization, I attempted to ensure isolation of the informant, to minimize situational variation by ensuring the identity of location across interviews, and to minimize distractions and disruptions. However, this required that my informants come to me, and to make an appointment to conduct the interview, sometimes several days in advance. In order to garner interviews with members of several different ethnic groups, I moved my own residence three times. My interviews therefore took place in three different locations separated by up to 60 kilometers, but each time outside in a housing compound, not appreciably different from the informant’s own housing arrangements, considered to belong to me. The interview was tape-recorded and transcribed simultaneously on the spot. A quid pro quo was given at the end of the interview, a salary equivalent to eight hours of manual labor at a nearby coffee plantation. I found through playing with the amount given in recompense that this was the minimum amount that would ensure a demographically representative sample. In order to investigate inter-interviewer reliability, I also instituted a program of re-interviewing informants using varying combinations of interviewers. This sample proved vital in isolating a number of effects biasing responses, as described in chapter 3 and appendix B.

Coding of Responses There have been a number of steps during which interpretation of the interview record, once produced, has taken place. Responses were originally handwritten by the interviewer. After leaving the field, with the written 152

THE CASE STUDY FOOD TABOOS IN THE ITURI

record of the informant’s response in hand, several additional steps of transcription and interpretatiodclassification then occurred to get the information into the coded computer form in which they have been analyzed. The first step was to code these written records into a computerized database. There are two interactive steps to this coding process: development of the coding system itself and the subsequent use of that system to summarize the salient meanings of interview responses. The rationale for the development of the coding system will be described at length elsewhere. The next step in interpretation consists of reduction of the high dimensionality of the encoded data into transformed data that can be more readily examined for patterns. I attempted to include as much of the information in the response as possible into a categorical system. From the detailed list of over six hundred different reasons for refusing foods, I again reduced the number of variant avoidances, based on my a priori expectations of where the meaningful variability lies (the cognitive units). In this book, two different systems of interview response categorization are used. The first, less specific scheme allows for thirteen different types of response, while a more detailed categorization includes thirty-six classifications. A catalog of the thirteen- and thirty-six-valued classification schemes can be found in table A.2. The thirteen-valued scheme is used in empirical analyses throughout this book unless otherwise noted. All interview coding and the categorization scheme were my responsibility. Coding typologies reflect my best sense of the cognitive and cultural distinctiveness among these beliefs: they vary by the periods during which they are effective (e.g., pregnancy versus adolescence), the kinds of consequence they invoke when violated (e.g., nausea versus dangerous cultural illnesses such as acquiring an “elephant nose” from eating elephant trunk), or the kind of implicit authority from which their value is derived (e.g., transmission from a kinsperson versus the intrinsic psychological appeal of the homeopathic principle [((youare what you eat”]). Table A.3 lists the 145 animals and the modal response made to the question about the edibility of that animal for each of the four primary ethnic groups (the two horticulturalist and two forager groups living in the Ituri). It also includes information about what proportion of each ethnic group reported that modal value, or had any sort of avoidance against that animal. 153

Table A.2.

Coding Schemes for lturi Food Avoidance Data

13-Valued Scheme

36-Valued Scheme

OK-to-eat

no avoidance positive belief non-affect dismissal “dirty” animal “irritative” animal anomalous animal uncertain animal unknown family authority sib-based restriction tareta asili clanlanimal relationship adolescent ceremonial birth ceremonial contact infection child non-homeopathic consumer non-homeopathic child-as-consumer non-homeopathic child homeopathic consumer homeopathic child-as-consumer homeopathic eke juvenile animal pregnancy-related sexual attraction place/marriage residential allegiance general populace sex-based restriction non-food magicalleconomic lrnbara baraza competitive carnivory non-consumption restriction

Attitudinal

Uncertain Family

Ceremonial Non-homeopathic

Homeopathic

Pregnancy-Infancy

Marriage Personal General

Barazallmbara Other

Code

I 2 3 4 5 6 9 U

a b t

@ f C

n g

k I m h i

i e 0

q d

P Y z V

! r X

0 # S

50 50

27 22 53 71 30 31 40 64 35 69 57 61

e OK OK

OK OK

OK OK OK

Okapi Akbu Akbu-bou

Ei

Meba Ndato Apayaka Balike Tiko

Mukadiu Akbutu

OK OK OK OK OK

g h

50 39 61 53 44 22 39 42 47 50

67 56 58 44 39

32 28 31 66 70

OK

e

Bosmans potto Crested mangabey Blue monkey LHoesti monkey Black-cheeked monkey Green monkey Abyssinian colobus Red colobus Angolan colobus Hamlyns/Owl-faced monkey De Brazzas monkey G rey-cheeked mangabey Anubis baboon Chimpanzee [mythological monkey] Giant forest hog Bush pig Wart hog Okapi Sitatunga Sitatunga (black)

AtidAbende Angara Saba NgimdNguta Bengi

Chabira Bururu Aboi Moko Mundukbu

58 61 33 50

32 34 67 49

OK e OK OK

Ennlish Name

Lese Name

% NOR Zero

%of Total

Model Value

Sudanic

Modal Response by Ethnic Group*

Table A.3.

OK OK 6 OK OK OK OK OK OK OK

OK

U

OK OK OK

U

e

OK e OK OK

Modal Value

87 75 46 75 62 76 73 81 93 92

74 73 22 25 42 25 25 19 22 19 14 14

14 28

28 25 28 25 19

36 22 17 17

68 38 88 85 52 44 33 76 83

% NonZero

%of Total

Bantu

~~

OK OK OK e h OK OK OK OK OK

OK OK

e OK OK

OK

e

OK e OK OK

Modal Value

98 95 89 66 57 88 71 88 89 70

70 71

41 75 50 96 96

84

82 36 98

~

Total

%of

Efe

~

~~~

14 17

It

14 14

II

14 22

II

6

17 17

22 6 6

II

14

17 19 6 14

% Now Zero

OK OK OK OK OK OK OK OK OK OK

OK

U

e OK q OK OK

OK

e OK

OK

Modal Value

~

96 59 52 85 89 85 93 96 93

loo

70 89

41 44 33 100 100

100 44 100 100

%of Total

Tswa

%

8 6 6

II

3 6 14 28 6 6

8

II

17 II 3 3

II

3 II 3 3

NonZero

(continued)

~~~

Spotted hyena Bongo Forest buffalo (red) Forest buffalo (black) Blue duiker Yellow-backed duiker Bay duiker Peter’s (or Redflanked) duiker Gabon duiker Black-fronted duiker Water chevrotain Bates pygmy antelope Tree dassie Cape (or Rock) dassie Aardvark Elephant Giant genet Spotted African civet African civet (“black) Small-spotted genet African linsang Unknown Viverrid sp. Black-footed mongoose

Aunga Soli Kekonupi-koba Tupi-gangi Medi Tochi Iti Daka

APOPO Yama Kaka Arofeillngbo Uku Banzo Chamu-kao Chamu-kutu Pandima BerekeKimbu Ndera Anzanza

Tau Munzu Ambaka

Endish Nome

OK OK OK OK OK OK OK OK OK

U

h

t

q

OK OK

OK

OK

OK

0

OK e e OK

Model Volue

44 44 42 44 47 53 69 50 64 58 53 47 53 53 50

35 41 30 25 25 13 43 24 39 31 27 19 30 31

40

53 58 58 39 39 53 36 58

Zero

% Non-

23 38 23 62 45 31 55 19

%of Totol

Sudonic

Modal Response by Ethnic Group* (continued)

Lese Nome

Table A.3.

14 28 17 25 28 22 36 28 28 33 33 28 31 33 17

88 81 82 88 32

84

OK OK OK OK

OK OK OK OK OK OK OK r OK

X

V

OK

OK

59 49 62 67 78 49 46 23 89

28 33 28 19 17 28 14 36

25 38 82 89 83 54 87 66

OK i OK OK OK OK

Zero

%of

% Non-

Tom1

Modof Volue

Bontu

OK OK

X

X

OK OK OK OK OK

OK OK OK

OK

0

OK

OK OK

0

0

OK e OK OK

Modol Volue

II

45 48 38 82 73 61 77 46 39 46 43 43 52 68 57 80 82 77 70 46 46 88 70

17 II 19 25 25 28 22 25 17 28 19 25 17 22

31 19 25 14 14 17 II 17

Totol

%of

% NonZero

Efe

OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK

OK OK

V

9 OK OK OK OK

Modal Volue

90 9 0 78 82 56 96 96 85 93 93 93 78 70 85 100

90

loo

41 44 89 100 74 56

Tot01

%of

Tswo

II 14 8 6 6 8 8 8 8 II 8 6 3

6 6

II 25 8 3 6 II 3 6

% NonZero

Anbisin bisi Katinjo Katinji Mbali Munjol Techakisornbi

Bekisi Asiresire

Boro-kboro Techakbutu Okbiditodi Wepinga-pinga uta-puvu-puvu Angangawei Kuru-kuru Vindi-vindi Oku Kate Fere lkule Abeke Akbedu-akbedu Kau Aka-kau Aka-medi Pinga Akora/Mungbu Akoramau

Cusirnanse Unknown Viverrid sp. Unknown Viverrid sp. Unknown Viverrid sp. Egyptian mongoose Unknown Viverrid sp. Honey badgerlRatel Clawless otter Tree pangolin Ground pangolin Brush-tailed porcupine Crested porcupine Giant elephant shrew Giant otter shrew Leopard Golden cat (striped) Golden cat (“red“) Giant forest squirrel African ground squirrel Cuvier’s fire-footed squirrel Palm squirrel Lecontes four-striped tree squirrel Lesser galago Forest rat Swamp rat Unknown mouse African dormouse

6 6 OK OK OK

OK 3

d OK OK

I I

X

OK OK OK OK OK OK

a

OK OK OK OK OK OK $

19 43 21 44 24

81 43

32 29 35 36 35 30 16 20 37 39 61 40 46 22 23 39 27 29 60 59

56 50 67 61 58

36 44

47 39 44 44 44 44 44 47 50 56 47 50 56 56 58 53 64 58 56 47

6 6 OK OK 3

OK 3

OK OK OK OK

X

X

P OK OK

OK OK OK OK

U

OK OK OK OK OK OK

31 60 26 38 35

73 56

54 83 92 93 92 90 31 57 85 88 93 32 49 66 57 44 81 80 82 68

25 28 36 28 28

19 17

II 14 14 II 8 22 22 17 17 17 22 33 28 42 33 19 25 19 19

19

OK 6 OK OK OK

OK OK

OK OK OK OK OK

X

OK OK OK OK OK OK OK OK OK OK OK OK OK OK

55 64 68 89 62

96 61

88 88 61 88 46 63 55 77 73 80 89 75 80 82 48 54 55 70 91 I00

17 17 19 14 19

6 14

17 17 22 17 42 28 28 22 22 28 14 14 14 19 25 31 28 22 8 3

OK OK

U

OK 6

OK OK

OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK

93 93 70 93 93

I00 89

100 96 I00 I00 93 I00 67 85 100 100 I00 52 67 89 70 78 93 93 I00 89

itinued)

6 8 14 6 6

3 8

3 6 3 3 6 3 19 14 3 3 3 II 22 II II 17 8 8 3 II

Unknown mouse House mouse Giant pouched rat Cane rat Zambeze grunter Heterobranchus longfilis Unknown Chrysichthys species Unknown Clarias species Squeaker Channallabes opus Electric catfish Unknown klonglanis species Unknown Citharinus species Unknown Cichlidae species Unknown Cichlidae species Lobeo fakcipinnis Unknown Citharinidae species Crowned hawk-eagle

Bunge Siobo Apulu Taro Bukana Kebi-tupi Mbichil Kebi-kosa Kodi Mborokoto Kebioaua Asarnba Epikba

lnjo

Ongo Apongo-pongo

Makburukutu

Abu

Atopi

English Name

50 42 33 33 42 44

73 74 64 65 58 80 69 66 78 78 53

OK OK 6

OK

OK

OK

OK OK

OK

OK OK

34 52 26 64 53 36 62

3 4 OK OK OK OK OK 44 44 39 53 42

44 58 47

53 44 69 50

%of Total

% NonZero

Modal Value

Sudanic

Modal Response by Ethnic Group* (continued)

Lese Nome

Table A.3.

OK

OK OK

OK

OK

OK

OK OK OK OK OK

3 4 OK OK OK OK OK

Modal Value

87

95 98

96

96

90

93 95 50 93 70

54 54 64 88 71 83 93

%of Total

0antu

22

93

OK OK 6

OK

89 63

OK

II

II

96

OK

95

98

71

84

93 93 63

46 63 73 93 86 77 95

%of Total

8

OK

OK OK 6 OK OK

14 14 28 II 28 17

OK 4 OK OK OK OK OK

Modal Value

22 19 31 19 22 22 17

% NonZero

Efe

14

14 14

6

II

6

14 II 8 17 8

14 17 17 14 14 19 II

% Now Zero

OK

OK OK

OK

OK

OK

OK OK OK OK

OK

OK OK OK OK OK OK OK

Modal Value

~~

100

3

3

3 100

loo

3

loo

3

3

loo

100

6 3 8 6 8

3

6 6 8 3 6 19

% NonZero

96 100 85 96 90

78 93 67 100 93 70 100

%of Total

Tswa

West African little sparrowhawk Kayarnbombo f'ygmy rail Bata Geese Ndulungenge Tiger bittern AnzanzilMasoeur Cattle egret Apidongi Great white egret SipilBayoya Openbill stork Alira Starlings Mvo-rnvo/Pongu Black-casqued hornbill Akoko Woodpeckers Siakobo Black-headed oriole Ndornbi Forest francolin Bingbingbi Black guineafowl Arnba Scaly francolin Ngiti Village weaver Akoru Blue turaco Ndoko Green turakos Keli-kofa Robin chat Aku Green parrot Ndirokbo Nightjars Akalingbongbo Palm swift Auku Owls Eku Old World fruit bats Aropi Flying squirrels Aferunoki Land tortoises Gbesa Kbekbenda Water turtles Matebe Unknown frog species (giant)

Sikbikuko

64 50 28 50 47 47 47 33 56 47 58 42 47 28 42 44 56 28 50 39 47 50 36 61 44 33

37 39 76 44 43 28 39 34 54 44 43 54 68 94 56 81 43 68 32 56 30 40 61 43 69 76

OK OK

OK 6

Z

OK

Z

OK OK

Z

h OK OK OK OK OK OK OK

2

6 OK

Z

Z

OK

r

OK

58

50

X

OK 6

OK OK

Z

3

Z

OK OK OK OK OK OK OK OK OK OK OK

Z

6 OK

r OK OK OK

X

X

45 44

90 88 77 34 72 70 50 29 37 52 42

84

34 61 83 78 51 29 72 33 61 31 67 74

59

31 36

25 19 17 19 22 33 22 28 36 33 25 25 17 14 19 22 25 22 25 28 33 31 28 28

36

OK OK

OK OK

Z

6

Z

OK OK OK OK OK OK OK OK

r r h OK OK

r

OK OK

r OK

X

X

95 50

52 57 54 91 50 39 52 46 36 52 64 98 75 I00 84 95 84 95 48 59 46 61 55 75

59

II 17

II 17 II 14 33 22 19 25 19 17 17 6 II 3 19 II 17 II 14 17 19 25 14 17

17

OK OK

OK I OK OK

Z

OK r OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK OK

1

96 85

II

89 44 52 96 89 67 82 26 85 78 93 100 96 100 100 100 74 96 89 41 82 41 74 67

(continued)

6 II

II 14 6 8 II II 19 8 II 8 3 6 3 3 3 II 6 6 19 14 17 14 14

14

67

Unknown frog species (leaping) Unknown frog species Lizards and gekkos Monitor Nile crocodile African python Gabon viper Unknown crab species Unknown crab species softshell crab species freshwater shrimp Unknown snail species Unknown snail species Unknown snail species Unknown snail species Rice locust KatydidlCaddisfly

Engbe

Mafiti Akongbo Abepfa orau Usa Kodo Kovo Koumakaki NatolKue-kue Njaka Lingba Matudu Makbou Angbatc ha Kiterere Batchakongu

English Name

OK d OK 6 OK OK 4 3 OK OK OK OK 6 3

2

2

2

Modal Value 2

50 53 42 42 39 53 53 28 47 33 42 47 47 33 47 33 50

Total

64 36 51 62 62 60 23 87 56 42 45 19 45 45 57 49 49 X

4 OK OK OK OK 4 OK 3 OK OK OK OK

2

OK OK

Modal Value

%of

% NanZen,

Sudanic

Modal Response by Ethnic Group* (continued)

Lese Name

Table A.3.

67 70 59 40 89 68 93 57 27 32 37 62 74 89 68 35

37

%of Total

Bantu

25 31 22 25 19 22 14 22 28 31 31 36 33 25 14 31

33

% NanZera

OK OK OK OK OK f

OK

OK

OK OK OK OK OK

2

OK

2

2

Modal Value

63 41 70 64 77 75 98 96 79 95 50 80 54 89 79 50

64

%of Total

Efe

19 22 17 14 25 22 6 6 17 8 25 28 17 19 14 II

17

% NonZero

OK OK OK OK OK OK OK OK OK OK OK OK OK

2

OK OK

2

Modal Value

96 70

loo

96 100 100 89

loo

100 100

loo

89 93 37 85 96

41

%of Total

Tswa

6 8 19 8 6 3 3 3 3 6 3 3 8 3 6 14

25

% NonZero

Unknown Morcusenius species

Unknown weevil (plantain) Unknown beetle (banana) Termites Unknown beetle larvae Unknown beetle larvae Goat Sheep Honey Mushrooms Nuts Shamba plants

28 25 14 6 3 14 14 8 8 17

71 62 90 99 I00 93 83 98 98 82

OK OK OK OK OK OK OK OK OK OK

50 56 31 25 19 44 47 8 3 28

21 25 74 96 96 67 74 99 100 92

2

OK

OK OK OK OK OK OK

OK OK

19

89

OK

56

28

28

51

a

OK

47

30

3

OK

OK OK OK OK OK OK OK OK OK

OK

3

98

I00

71 100

so

36 41 89 98 100

77

48

6

33 39 II 6 3 28 14 3 3

22

22

OK

OK

OK OK OK

OK OK OK OK OK

OK

OK

~

~

~~~

63

63 93 96 100 100 41 41 100 100

96

93

NOTETable values utilize the coding system listed in tableA.1. Column values for each ethnic group list the modal response from informants in that ethnic group, the proportion of all informants in that ethnic group making the modal response, and the proportion of informants reporting some form of avoidance for that food item, respectively.Al1 modal values equal to the number one (I) have been changed to “OK’ (i.e..“OK-to-eat“) to distinguish them typographically from values of “el” (I).

Etchu Anjutoku Tangbflongoso h e m e [Mburi] Kondoro meli [bioka] [rnbego ya poli] [miyani ya shamba] Boku

Apachirichiri

Apangulele

8

II 8 6 3 3 17 19 3 3

6

6

APPENDIX A

Notes 1. The names of individuals as well as the exact location of the Ituri Project research station are, by policy of the Ituri Project (and standard ethnographic practice), not identified, in order to provide a measure of protection to those who participate in our research. 2. To determine which animals were most salient, I did extensive standard testing, First, animal names were free listed by numerous knowledgeable informants, who then also responded to animal pictures in guidebooks for the area. At my request, animals killed locally were also brought to me, at which point I collected ad libitum identifications from those present at the time. From this information, I determined the complete set of names for animals in a local language, KiLese, with careful translation to other languages by informants fluent in both. From this list of approximately three hundred animal names, and after questioning over fifty informants with regard to the complete list, a select list of 145 was created, which included all names for major animals and the others about which there was little confusion regarding the referent. All of the animals killed by over four hundred local individuals during the period of a year are but a subset of this reduced list, which therefore should be the set of animals figuring prominently in everyday experience, as well as those serving most of the cultural and biological functions associated with animals in Ituri society.

162

APPENDIX B

REFLEXIVITY IS NECESSARY

T

his appendix presents an empirical study that illustrates the virtues of the data collection situation approach to ethnographic data collection and analysis, in support of the analysis in chapter 3. For the purposes of this appendix, I restrict my attention primarily to a subset of the Ituri data described in appendix A. This subset of interviews derives from sixty-five Lese, Budu, and Efe individuals who were interviewed twice concerning their food avoidances; two of these individuals were interviewed a third time, making a total of 134 interviews. (Table B.l describes the demographic composition of those who participated in the repeated interviews.) Different combinations of interviewers were used with respect to each informant. In most cases, one of the pair of interviewers belonged to Table B. I.

DemographicComposition of Informants in Repeat Interviews

Ethnic Group/

Mole

Age Class ~~

~

Fernole

Lese
TOTAL

Totol

~

45 (69%)

7 5 4 9

2 10 3 5

0 3 2 2

0 0 0

0 5 3 2 -

0 2 0 0 -

42 (65%)

23 (35%)

8 (12%)

I

12 (19%)

163

65 ( I 00%)

APPENDIX B Table 6.2.

Paired Interviews by Ethnic Group

Ethnic Group

AIA

AIK

KIK

Lese

21

Efe

23 I I -

0 -0

Total

25

21

2 0 0 2

Budu

AIM

KIN

AIN

I

0 0

0

7

-0

-4

7 -

8

4

7

0

NOTETwo of the sets of paired interviews were performed on the same individuals, so that although there are sim-seven total sets of paired interviews,only sixty-five different individuals were involved. A is me, the anthropologist; K is the Lese interviewer of horticulturalists; M is the Budu interviewer of Budu horticulturalisu; and N is the Lese interviewer of Efe foragers.

the same cultural group as the informant, while I was the other (see table B.2). The same method of eliciting responses to a highly structured questionnaire was used in all interviews, but with varying amounts of time elapsed since the first interview (the mean number of elapsed days was 109, with a maximum of 394 and minimum of 2 days). Based on these data, I develop a multivariate model of the data collection situation. This model is then estimated using logistic regression to determine what influences can bias the likelihood of repeating the same response to a question about a food’s edibility. Although repeat sampling has been used for small samples of individuals in ethnographic research (e.g., Foster 1979), and similar statistical analyses of interview data have been performed in the sociological survey literature (e.g., Blanc and Croft 1992), to my knowledge such a comprehensive model has not previously been applied to a traditional anthropological population. This procedure allows me to partition the likelihood that an informant makes different responses to the same question to several classes of effects: those due to the interviewer, those due to interactions between informant and interviewer characteristics, and those due to aspects of informant cognitive variation (or real variation in belief).

The Statistical Model The most important factor that influences how an informant responds to interview questions should be what an informant believes. However, an array of other factors also influence interview responses. The goal of this analysis is to distinguish variability in responses due to informant knowledge from the effects of a wide variety of exogenous factors, including other aspects of informant cognition. 164

REFLEXIVITY IS NECESSARY

How can a statistical model accomplish this task? It is relatively easy to separate informant-based variability from that linked directly either to the identity of particular interviewers or random situational variability. However, it is a more difficult task to separate informant-based variability into that due to knowledge differences and that due to differences in recall ability or strategic reasoning biases caused by the public nature of the interview. This division of informant-based variability depends on interpretation of the pattern of results from the statistical models. Next, I discuss how informant-based effects can be statistically isolated from the other sources of variability. The goal is to uncover the “minimally contextualized” response. This is a response that reflects only the informant’s own cognitive operations (i.e., does not reflect the presence of the interviewer, or the randomizing influence of situational distractions). Such a response is what an informant would tell him- or herself when sitting quietly alone in reaction to a spontaneous, internal motivation-a udaydreamed” thought representing the informant’s private, nonstrategic reflection. Since the ethnographer is typically attempting to assess informant beliefs reliably and accurately, variability in interview responses that reflects the influence of an informant’s recall ability, as well as interviewer and situational factors, will be called “exogenous” or methodological. Only variability in responses that is due strictly to informant knowledge will be considered here to be “endogenous,” because such variability legitimately reflects differences in belief among informants. The proportion of response variability that is exogenous will be of particular interest, since higher proportions of exogenous variability suggest that informant responses during ethnographic interviews are relatively unreliable. O n the other hand, the demonstration that a proportion of variability in responses to a particular question is due strictly to individual knowledge will indicate the existence of intra-population variation in beliefs. In this way, the estimated model will enable me to address the question of whether there is “real” variability in beliefs in the study population or whether all the observed variability in informant responses can be attributed simply to methodological factors. The resulting, presumably hopeful, message is that, despite methodological problems, informant beliefs remain an important influence on responses to interview questions. 165

APPENDIX B

Discrepant Responses in Interviews To measure the effects of the different factors that influence interview responses requires a measure of what differs (i.e., a unit of analysis that can vary). In the present case, interview responses concern an individual’s beliefs about the reasons and times during which consumption of a particular animal should be avoided. These responses were classified into a number of different categories. I then compared responses to the same question by the same informant on two different occasions. These paired responses are considered discrepant when they are classified as belonging to different categories of response.’ The objective is to determine the factors that significantly correlate with the probability that an informant will change his or her report of avoidance type X with respect to animal Y between the first and second interviews. However, I believe there are two distinct cognitive operations that result in discrepant responses. These are distinguished in the analysis, because comparing two sets of results should help to pin down the underlying effect each methodological factor has on what informants say, The first type of discrepancy between responses arises when an informant on one occasion “forgets” a relevant avoidance with respect to the animal in question, or did not make an appropriate connection between their knowledge of an animal and any restrictions on its consumption (e.g., due to confusion concerning the animal name as a stimulus). The first measure is therefore defined as a report of “no avoidance” coupled with an avoidance report in the other interview. The other possibility is that an informant made a “mistake”in one of their two responses to interview questions, and reported an avoidance on one occasion that was substantively different from the avoidance they reported on the other occasion. These mistakes constitute the second type of discrepancy (see the statistical postscript for other details concerning the discrepancy measures).’ The data on Ituri food avoidances used here consist of 8,484individual responses to interview questions (sixty-seven repeated interviews times the 140 different questions of each interview, minus those observations missing values for some variables). The data were used to determine the probabilities of making discrepant reports associated with various factors. Two models were formulated, one for each kind of discrepancy: forgetfulness and mistakes. Discrepancies in responses, coded as 1if responses were different or 0 if the same, are used as the dependent variable. Each of these 166

REFLEXIVITY IS NECESSARY

models includes the same set of explanatory variables, as well as two-way interactions involving these variables. I divided the explanatory variables into four sets, corresponding to the four classes of effects identified by the Reflexive Analytical approach: (1)situational factors; (2) the interviewer; (3) various types of interaction between the interviewer and informant; Table 8.3.

The Multivariate Model

Efea

Measure

Interview-Based: Situational

the unexplained variation in the probability of a discrepancy

Question-Based: Animal Familiarity Animal Group

ElapsedTime Cognitive Load Interviewer-Based: Experience

Dissimilarity

Major Cultural Difference

Minor Cultural Difference Informant-Based Ethnic Group

Sex Age Schooling

dummy variable, coded as I if the informant reports having seen the animal (alive o r dead); otherwise coded as 0 seven classes, defined according t o named emic categories; classes are roughly equivalent t o the primates, viverrids, bovidlsuids, rodents, birds, fish, and “other” animals; the last class is used as a reference (statistically redundant) category number of days intervening between the first and second interviews of an informant measured as the total number of avoidances expressed by the informant during his or her interview number of an informant’s first interview in chronological order for the pair of interviewers who conducted the interviews degree of dissimilarity in cultural background between the pair of interviewers who conducted an informant’ interviews; coded as I if the two interviewers were the same person; coded as 2 if the two interviewers came from the same ethnic group; coded as 3 if the two interviewers came from different ethnic groups; considered t o be ordinally scaled number of times the informant faced the anthropologist as interviewer; also equal t o the number of interviews conducted in Swahili with an informant; considered to be ordinally scaled number of times a forager informant faced a horticulturalist as interviewer; considered t o be ordinally scaled categorical variable with three values: Lese or Budu horticulturalist. or Efe forager; Bantu is used as the reference category dummy variable: coded 0 for female, I for male measured in years years in school

NOTE See text for details concerning variable definitions

167

APPENDIX B

and (4) the informant’s private beliefs. Table B.3 provides a summary listing of the explanatory variables included in the model, how they are measured, and some details concerning variable values. The rest of this appendix consists of a detailed discussion of the results from the statistical models of discrepant interview responses. Since this kind of data is rare in ethnography, it is analyzed in depth. I believe this depth of discussion is informative, because despite the particularity of its collection, the messages that can be derived from it are probably universal in the sense that the kinds of problems it highlights are likely to be associated with ethnographic data collection procedures of any kind. In particular, I discuss each of the four classes of effects identified by the data collection situation approach in turn, beginning with the situational, interviewer-based, and informant-interviewer interaction effects. I argue that the randomizing influence of the unique situation of each interview on what informants say is considerable; that interviewers have individual styles of recording informant responses; and that the pattern of results from the informant-interviewer interaction effects shows the strategic nature of informant reasoning when confronted with the public situation of the interview. I conclude by discussing the variability in responses deriving from characteristics of the informants themselves. I argue that several of these variables reflect the operation of differences in recall ability, one of the three aspects of human cognition involved in making the decision of how to respond to interview questions, while others reflect reasoning about the social situation itself. Only one set of variables suggests differences in knowledge or belief with respect to individual questions. I call these “SOcial roles” because they divide the population into subgroups by ethnic group, sex, age, and educational level. This process of eliminating the influence of all possible exogenous factors allows me to determine that there are non-methodological differences between informants in their beliefs or knowledge concerning the edibility of particular foods. The best-fitting models were estimated using backward stepwise model selection criteria and the SPSS logistic regression pr~cedure.~ Loosely speaking, logistic regression determines the probability that a particular event (i.e., a question as stimulus) will result in one or the other possible outcome (i.e., a discrepant or equivalent response), based on the relative frequency of the two outcomes in the total number of responses given by informants. Since each discrepancy arises from comparing two specific interviews-which involved a particular informant and combina168

REFLEXIVITY IS NECESSARY

tion of interviewers, dates, locations, and so forth-the procedure can also determine those aspects of the interview situation that correlate with these events. Depending on the relative strength of the correlation between a particular aspect of the interview and discrepant responses, the model assigns a statistical significance to that aspect. Like any correlation, the measure of statistical significance is also a measure of the relative proportion of variability in discrepant responses that is related to that aspect of the interview situation. The implicit suggestion of such a complicated model is that many factors have different roles in changing informant responses to particular questions. The point of using such a model is to assess the independent effect of each factor on the probability of a discrepant response. Multivariate models such as that specified here do this by identifying the estimated change in the probability of a discrepancy arising from a unit change in the value of each factor while controlling for the influence of the other factors specified in the model. Because of this feature, the logistic regression procedure can isolate the independent effect of each of the four kinds of factors identified by the data collection situation approach (discussed in chapter 3). First, purely situational factors such as a difference in the weather between the two different days on which the informant’s interviews were conducted may affect an informant’s mood, and hence vary responses. Ideally, these contextual sources of variation manifest themselves as random noise when considered from the viewpoint of a large sample of repeated responses. Situational variability can therefore be measured as the proportion of total variability that remains unexplained by the multivariate statistical model of discrepancies, since this variability is not systematicallyrelated to any of the explanatory variables. However, the variability that is unexplained is a function of the model chosen. Since it is typically desirable to explain as much variation as possible (in order to minimize the proportion of variability accredited to this leftover category), a rather complex multivariate model will usually have to be estimated. Next, by implicit or explicit direction of the interview, the interviewer also plays some role in determining what informants say. In the anthropological context, there is the additional complication that the interviewer and the informant typically come from different cultural groups, and must find some common language for their discourse. Moreover, the anthropologist typically does not have the degree of acquaintance with local cultural 169

APPENDIX B

practices characteristic of someone who has grown up in the cultural group under study. These factors suggest that cultural differences between the interviewer and the informant may significantly influence the kinds of responses informants will make. Specifically, an interviewer’s biases in attending to and recording a response are likely to be unrepresentative of the informant’s beliefs. Still, a particular interviewer’s style of recording responses should be statistically consistent and appear as a bias in the types of coding used by that interviewer when compared to other interviewers. This bias is an estimate of the second influence on interview responses identified by the Reflexive Analytical approach, namely the interviewer. If single informants are interviewed repeatedly by different interviewers, these biases in coding should appear as particular patterns of discrepancy in an informant’s recorded responses to paired questions. These patterns can be detected by estimating the model specified above. Interactions between the informant and interviewerthe third factor-can arise either as a result of interpersonal effects (e.g., personality clashes, sexual politics, cultural background differences), or through the medium of their exchange (e.g., misunderstandings due to language use). Regardless of the channel through which such interactions occur, they can be isolated as statistical correlations between specific characteristics of informants and their interviewers. In particular, such correlations suggest that certain types of informants bias their responses in a specific fashion when faced with different kinds of interviewers. Their responses can therefore be interpreted to reflect strategic decisions concerning what they will reveal about their beliefs to such interviewers. There are also factors that tend to be associated with informants that might influence the probability of a repeated response, but which do not represent substantive differences in belief between informants about a particular topic. For this reason, the models include several methodological control factors to take account of these other possibilities. The first of these factors is designed to control for the fact that interview data consist of responses to a set of different questions. It would not be remarkable to observe that informants report being able to eat a species of pangolin that lives in trees but not the one that lives on the ground, for instance. What would be of interest is if some informants report being able to eat the ground-dwelling pangolin, while others claim to avoid it. The question then becomes whether this represents real variability in belief about the ground-dwelling pangolin. However, logistic regression is an 170

REFLEXIVITY IS NECESSARY

iterative, nonlinear fitting procedure, adjusting the estimated relationship between the probability of a discrepancy and any given factor according to the pattern of discrepancies across all responses. Unless variables that reflect the characteristics of the different questions are included in the model, the results will confound intra-question with inter-question variation in the likelihood of a discrepancy. Variation in the salience of the question as a means of eliciting responses has intrinsic interest in some contexts (e.g., educational testing). Such variation also makes for a more robust dataset from which to estimate the model. In the present case, what is of interest is whether informants have different beliefs about the same topic. The model thus includes variables designed to adjust for the salience of the question itself in order to get at this substantive question. It is also possible that some informants are more prone to make discrepant reports simply because they are less able to recall their beliefs. This variation in the ability to recall what is stored in memory is a methodologically confounding effect. I assume that the probability of making a mistake or forgetting a previously remembered avoidance increases as the total number of avoidances stored in memory increases. In order to control for any potential variation in the ability to recall particular avoidances, the statistical model includes a variable measuring the total number of avoidances reported by an informant. Finally, there is a third problem: informants may learn something that changes their belief between the two interviews, making their reports on each occasion substantively different but still true. In order to control for the possibility of such changes in an informant’s belief, the model also incorporates a variable that measures the time interval between an informant’s two interviews, on the assumption that individual learning is more likely to occur as more time elapses between interviews. This attributes some proportion of any difference between reports as due to the possibility that they are in fact two true but different beliefs. Inter-informant variation in belief is thus isolated from any intra-informant variation that may be present in the data. What remains after controlling for question-dependent effects, changes in the base rate at which different informants are able to recall their beliefs, and independent learning between interviews, should be question-specific variation in the likelihood of a discrepancy. We may assume that such variation in the probability of repeating responses is due to different informants having beliefs of different salience about the 171

APPENDIX B

topic of specific questions. I will therefore argue that variability in the likelihood of a discrepancy that attaches to characteristics of informants themselves (e.g., their ethnic group and sex), once all other possible sources have been controlled for, is indirect evidence for the existence of meaningful variation in belief in the population. Such variability is assumed to be due either to differences in access to the appropriate cultural beliefs (e.g., from social restrictions on interpersonal contact) or to the development of non-normative beliefs from prior experience with particular foods. In either case, this variability will be considered independent of any methodological influence, and hence representative of the informants themselves, the last of the four sources of variation considered by the model. Table B.4 presents the results from the multivariate statistical estimation process concerning forgetfulness, while table B.5 presents the best-fitting model of mistakes. Informants forgot previous responses on about one of five occasions (22.4%). Mistakes were made in another 17.9% of repeated responses. These occurrences of discrepant responses are correlated with a variety of characteristics of both the informant and the interviewer. By making the unlikely assumption that all the specified effects have an independent effect on the probability of a discrepancy (i.e., all the explanatory variables are statistically independent), the absolute values of R can be summed to measure the relative proportion of variability in the probability of a discrepancy due to the various effects included in the model (see table B.6 for the results of these calculation~).~ Using this rough method, the variability in forgetfulness explained by the statistical model can be attributed as follows: 17% to interviewerbased effects; 16% to interviewer-informant interactions; and 67% to informant-based effects (situational factors contribute to the unexplained proportion of variability). However, effects due to the characteristics of informants themselves only account for about 19% of explained variability. Similarly, the explained variability in mistake making can be divided into the following classes: 12% to interviewer-based effects; 10% to interviewer-informant interactions; and fully 79% due to informant-based effects. However, non-methodological informant-based variability is again relatively small: only 16%.5 When the methodological control factors related to differences in informant responses are isolated (as in table B.6), each of the three non172

Table 8.4.

Determinantsof “Forgetfulness”

Variable MethodologicalControl Facors: Animal Familiarity Animal Group Primates Viverrids BovidslSuids Rodents Birds Fish Cognitive Load Animal FamiliarityWith Cognitive Load Animal Group With Animal Familiarity Primates Viverrids BovidslSuids Rodents Birds Fish Animal Group With Cognitive Load Primates Viverrids BovidslSuids Rodents Birds Fish Interviewer-Efect Experience

B

Wold

Sig of Wold

-1.1621 25.9559 -.6942 I.7863 I.0937 -.2956 - ,4902 -.2413 -.0180 .O I64

I 3.290I

.OW3

-.0357

,0002 I .3222 16.2257 6.6603 .22I6 .9I75 .I976 19.0774 16.0094

.0397 ,2502 .000I .0099 .6378 .3381 .6566

.040I ,0229 .0000

10.2524

.I 144

.oooo

- .OO47

.OM4 .0084 .475I 3.82I0 2.2322 5.7398

.9842 .9269 ,4906 .0506 .I352 .O I66 .0659

.O I65

.I200 .7I25 -.3147 - s449 50.4I69 .0026 -.0159 - .0044 -.oOOl

.oooo

.oooo

.omI

.6424 28.7927 2.5837

,4228

.OOI5

.9692 S278

.oooo .I080

.oom .oooo - .0439

.0398

.4995 5.9673 2.9853 ,744I .6125 .7856 ,9822 1.0166

.moo

.9953

.OOOO

I.O I66

.moo - ,005 I - .0206

1.1275 2.0392 ,7300 s799

.oooo

I.0026

-.0550 - .008I

.9842 .9956 .9999 ,9983 I.0203

.0143

.oooo .0000 .0554

-0017 .020I

.3987 29.I588

- 1.0363

I 1.3867

,0007

-.0326

- ,0625

.3777

.9394

I I .2399

.5388 .OM8

.moo

.2064

,0323

I.2292

4.8658 I.6363 .0203

.0274 .2008 .O I84

.0180 .OOOO

2.1 152 3.8274

.6442 2.I555 3.01 I I 1.0728 5.6434 1.5105

,4222 .I421 .0827 .3003 .O I75 .2191

.moo

I.0989

-.ow2 -.0107 .0000 ,0203

.8803 .8649 .8955 1.2607 ,8783

.oooo

Inter-lnterviewer-Efefeas:

Dissimilarity Experience With Dissimilarity

.oom

.3I28

Informant-lnterviewer Interaction Effects: .7492 Major Cultural Difference Minor Cultural Difference I .3422 Dissimilarity With 14.9945 Animal Group ,0943 Primates -.I275 Viverrids -.I452 BovidslSuids -.I 104 Rodents .23I7 Birds Fish -.I297

.oom

,354

(continued)

APPENDIX B Table B.4.

Determinants of "Forgetfulness" (continued)

Variable Major Cultural Difference With Sex Major Cultural Difference With Animal Group Primates Viverrids BovidslSuids Rodents Birds Fish Major Cultural Difference With Experience Minor Cultural Difference With Sex Minor Cultural Difference With Cognitive Load Informant Characteristics: Sex Schooling Constant

8

Wald

-.5105

Sig of Wald

12.8781

.OW3

19.69I3

.0031

,0295

.3045 -.2913 - .222 I -.0362 .2a4 I - .4050 .4023

2.3759 4.8137 2.9034 .0462 2.9985 6.1879 10.7247

I

R

hP@)

-.0351

.6002

.I232 ,0282 .0884 .8297 .0833 .0129 .OOl I

,0065 -.0178 -.0101 .O I06 -.0218 .03 14

I .3560 .7473 .8008 .9644 1.3286 .6670 I .4953

4.2742

.0387

-.0160

.3566

.o I20

8.7027

.0032

.0275

1.0120

.9257 -.043 I -.8982

16.8596 12.3339 1.4551

.0000 ,0004

,0410 -.0342

2.5235 .9578

- I.03 I

.0000

.2277

Model:

-2 Log Likelihood Model Chi-square Goodness of Fit

Chi-square 8692.430 342. I58 8484.079

df 8439 30 8439

Significance

.oooo

.moo .OOOO

NOTEThe Wald statistic is the square of the ratio of the coefficient value (B) t o its standard error: R is a measure of the strength of relationship between the covariate and the dependent variable; Exp(B) is the change in the odds (not log odds) of the event measured by the dependent variable occurring due t o a unit change in the covariate.The categorical variables (Ethnic Group and Animal Group) were converted to indicator variables, each of which was weighted according to its difference from the mean value for all the categories of that variable.

situational classes of influence on responses explain about a third of the observed variability (averaging the figures for forgetfulness and mistake making). In addition, the overall fit of the statistical models is not particularly good (as shown by the significant -2 log likelihood statistic), indicating there is significant situational variability in interview responses as well. Thus, factors other than what the informant believes appear to dominate what an informant says on a particular occasion. Since ethnographers derive most of their information from interview-like responses (even if the dialogue between the anthropologist and the informant is very informal), and many ethnographers take informants at their word from a single interview concerning a particular topic, these results suggests that 174

Table 8.5.

Determinants of “Mistakes”

B

Wold

- I.2985

Sig of Wold

R

.OOOO .OOO7 .2376 .oOOI .3325 .I043 .0963 .3 I94

-.I750 .038 I

.2729

- .4894 - I.2522 -.3160 ,6930 .4932 .SO34 .0191 -.MI7 I1.1015

238.7772 23.2 II 7 I.3945 15.3589 .9392 2.6386 2.7652 .99 I6 42.9927 I.5857 .0853

.moo

.oooo

.0728

.2079 .OOOo

.oOOO

.6 I30 .2859 .7290 I.9998 I.6376 I.6543 1.0193 .9983

-.07W .2580 .0402 - .4852 .353 I -.3479

.I072 2.9254 .0624 3.5542 3.9237 2. I002

.7433 .0872 .8028 .0594 .0476 .1473

.moo

32.304 I 2. I085 16.9095 5.0 I20 3. I480 .6672 5.8959 5.0090

.0000

,0062 .O I43 -.0076 .0072 - .0024 -.0126 4.3 IE-05

Interviewer Efea: Experience

.0116

Inter- Interviewer Efea: Dissimilarity

.3950

Variable MethodologicalControl Faaon: Animal Familiarity Animal Group Primates Viverrids BovidslSuids Rodents Birds Fish Cognitive Load ElapsedTime Animal Group With Animal Familiarity Primates Viverrids BovidslSuids Rodents Birds Fish Animal Group With Cognitive Load Primates Viverrids BovidslSuids Rodents Birds Fish ElapsedTime With Cognitive Load

Informant-Interviewer Interaction Efects: .3527 Major Cultural Difference Experience With Age -.OOO5 Dissimilarity With Age -003 I Major Cultural Difference With Animal Group -.0371 Primates .I068 Viverrids BovidslSuids .4767 Rodents -.3949 Birds - .0700 Fish -.I209

-MI6

.m .009 I .0099

.oooo

.O I09 .OOOO -.0142 .O I58 -.0036

.9324 I.2943 1.0410 .6 I56 I.4235 .7062

.0252 .0760 .4 I40 .O I52 .0252

.05 I 2 .0037 .0439 -.0197 .o I22 .om0 - .0224 .O I97

I.0062 1.0144 .9924 I.0073 .9976 .9875 I

1.0108

.3 147

.m

1.01 17

28.4928

.OOOO

.0585

I.4844

IS. I282 3.2307 2.6047 20.4497

.OOOI ,0723 .I065 .0023

.0412 -.0126 - .0088 .033 I

I.4229 .9995 .9969

.06 I3 .8544 13.7803 8.04 I8 .4388 .4049

.8045 .3553 .0002 .W6 .SO77 .5246

.moo .moo

.9636 1.1 127 1.6107 .6738 .9324 .886 I

.I465 .m

.0390 -.0280 .om0

.oooo

.oooo

{continued)

APPENDIX B Table B.5.

Determinants of “Mistakes” (continued)

Variable Informant Choracteristics: Ethnic Group Lese Forager Age Ethnic Group With Age Lese Forager Constant

Wald

Sig of Wald

R

17.8227 14. I969 6.0387 5.3 I82 7.51 18 5.5785 3.9943 124.7973

.ooo I .om2 .OI40 .021 I .0234 .OI82 .0457

,0423 .0397 -.0229 .0207 .02I 3 -.0215 ,0161

B

.5926 -.4816 .OI49 -.OlOO ,0115 -3.5862

ExblB)

1.8088 .6178 1.0151 ,9900 1.01 16

.oooo

Model:

-2 Log Likelihood Model Chi-square Goodness of Fit

Chi-Square 7653.477 1089.443 8496. I I5

df 8445 23 8445

Significance

.oooo .oooo .ow0

NOTEThe Wald statistic is the square of the ratio of the coefficient value (B) t o its standard error; R is a measure of the strength of relationship between the covariate and the dependent variable; Exp(B) is the change in the odds (not log odds) of the event measured by the dependent variable occurring due t o a unit change in the covariate.The categorical variables (Ethnic Group and Animal Group) were converted t o indicator variables, each of which was weighted according to its difference from the mean value for all the categories of that variable.

the great majority of interview-based ethnographic studies are probably methodologically problematic. The methodological problems indicated by these results warrant further discussion. The discussion has four parts, reflecting the four types of factors affecting informant responses: (1) situational or contextual factors, (2) interviewer-based factors, (3) informant-interviewer interaction effects, and (4)informant-based effects. Within this last category are some effects considered to be meaningful because they reflect variability in what informants believe about the subject of the question. While the quantitative aspects of the results discussed are specific to one population, the pattern of results provides ample evidence that a wide variety of factors influence informants’ answers to questions put to them in an ethnographic context. This discussion will make clear that many types of effects not typically considered in the anthropological literature are significant determinants of discrepancies in informant responses. In addition, while the method of data collection used here was not standard for cultural anthropology, the subject of the interviews certainly is: food avoidances in a traditional anthropological population. This suggests that

176

Table B.6.

TYP of

Determining Discrepancies'

Em

Forgetfulness: UnexplainedVariation: Situational Factors Methodological Control Factors: Question-Dependent Response Bias Cognitive Load ElapsedTime Totals: Interviewer Effects: Experience Inter-Interviewer Dissimilarity Informant-Interviewer Interaction Effects: Major Cultural Difference Minor Cultural Difference Informant Effects: Informant Characteristics Totals: Mistakes: UnexplainedVariation: Situational Factors Methodological Control Factors: Question-Dependent Response Bias Cognitive Load ElapsedTime Totals: Interviewer Effects: Experience Inter-Interviewer Dissimilarity Informant-Interviewer Interaction Effects: Major Cultural Difference Minor Cultural Difference Informant Effects: Informant Characteristics Totals:

% of Explained Variationb

% of Voriotion Within CIass

??? (significant)

??? (significant)

28.158.0 20.442.0

0.00.0 48.5 IOO.0 I I.923.2 4.79. I

12.223.7 4.07.8 18.636.2 5 1.4 IOO.0

??? (significant)

??? (significant)

42.9

68.3

18.2 I.7 62.8

29.0 2.7

100.0

I.o 10.5

2.9 28.2

9.7 0.0

25.9

16.0

43.0 100.0

37.2

0.0

aThe values reported here depend on values of R reported in tables 6.4 and 6.5, with the assumption that all effects are statistically independent the figures reflect the somewhat arbitrary assignment of one-half the variation due t o interactions between variables t o each of the variables involved all figures are rounded upward; figures in italics are totals for that class of effect bThis column does not count situational variability, which is unexplained and unmeasured, although a statistically significant residual factor.

APPENDIX B

the variety of effects found to be significant in the present case is likely to be characteristic of many ethnographic situations.

Interview-Based Effects: SituationalVariation Random variability arises from circumstantial vagaries of the interview situation itself. Using the Reflexive Analytical approach in combination with a multivariate statistical model results in a rather specific concept of the situational context of an interview. Contextual variation represents the influence of factors that are not specified in the statistical model. The relatively poor overall fit (shown by the significant log-likelihood statistic) of the models of forgetfulness and mistake making indicates that purely situational effects (such as the mood of one or other of the interactants, or social distractions in proximity to the interview location) are indeed significant. Although they bias responses in a particular way, such effects cannot be rigorously identified because they are highly specific to a given interview. With respect to that interview, a particular situational factor can significantly bias a response. Placing great weight on any single response to build up a picture of some aspect of a culture is obviously problematic when this response reflects such spurious situational factors. However, when considered over many interviews, situational factors, taken together, appear to have a random effect on the propensity to make discrepant responses, and so do not affect the relative significance of the effects actually specified in the model.6 The proportion of variability explained by the forgetfulness model is somewhat less than that explained by the model of mistakes. Therefore, situational variability is a relatively larger determinant of forgetfid responses than of mistakes. This suggests that forgetfulness is more a function of the insufficiency of the stimulus (question), or of a distracting effect on one occasion that was not present on the other. This interpretation is also supported by comparing the pattern, significance, and number of effects in the best-fitting models for forgetfulness and mistake making. The relatively few informant characteristics (especially the lack of an effect of informant’s age), and the relative preponderance of interviewerrelated interactions (especially an effect due to Interviewer Experience at interviewing), suggest a more important role for interviewer-informant interactions in the determination of forgetful than mistaken observations (shown in table B.6 in terms of R as well). 178

REFLEXIVITY IS NECESSARY

The model of mistake making, on the other hand, suggests the opposite, with considerably fewer interviewer effects but a greater number of interactions involving informant characteristics. Interviewer Experience does not decrease the prevalence of mistakes as much as forgetfulness discrepancies, suggesting there is some base rate of mistake making due to informant cognition. Mistakes probably occur at least partly from confusion in recall due to similar avoidances of different animals. If so, mistakes arise when reasoning is applied to knowledge that is disorganized in some fashion. This is suggested by the high degree of significance of the variables introduced into the model to account for differences in the total number of food avoidance an informant must remember (Cognitive Load), and the cognitive salience of the different questions used to elicit responses (Animal Group and Animal Familiarity). This analysis suggests that “forgetfulness” and “mistake making’’ are different kinds of discrepancies.

InterviewepBased Effects The first group of effects specified in the models of discrepant responses (not just the unexplained variability attributed to situational factors) describe biases in responses due to the interviewer. I first discuss the effect of Interviewer Experience on the probability that an informant will respond differently on separate occasions. But since two interviewers are involved in the production of discrepancies, each having his or her own personal methods of eliciting and recording informant responses, it is also important to isolate the inter-interviewer effects on responses that arise from particular combinations of interviewers.

Interviewer Experience

The statistical model includes an interviewer’s degree of experience in dealing with informants, measured as the number of interviews that a particular pair of interviewers had previously completed (in chronological sequence, measured from the date of the first interview of that informant). As interviewer experience has been shown in sociological survey research to significantly affect the reliability of responses (Brenner 1982), we may expect that increased interviewer experience will decrease the likelihood of a discrepancy because of the interviewer’s increasing ability to elicit responses representative of an informant’s beliefs. 179

APPENDIX B

The results from the multivariate statistical models indicate that Interviewer Experience indeed figures as a significant influence in reducing both the likelihood of forgetful as well as mistaken responses (see table B.6). However, there is a considerable difference in the significance of Interviewer Experience on the two different kinds of discrepancies: 12% of explained variability in forgethlness is attributable to Interviewer Experience, but only 1%of variability to mistakes. That the effect of experience is considerably larger on forgetfulness shows that experience primarily tends to reduce the number of occasions on which one interviewer records an avoidance while the other one does not. This probably reflects an increased assertiveness among interviewers with greater experience. Informants are generally reluctant to divulge their personal beliefs, if only because it is simpler and faster to answer, “I have no avoidance against that animal.” Experienced interviewers, to the degree that they are unwilling to accept this facile answer, will probe for responses, thus becoming more likely to elicit a response that includes types of avoidances informants are more reluctant to acknowledge. Interviews conducted more assertively should therefore report more avoidances, even for the same informant. Indeed, probing by interviewers is commonly emphasized as an important aspect of interviewing method in the social survey literature, where it has been found to increase the reliability of responses (Dijkstra and van der Zouwen 1982). Interviewers’influence was bound to a bewildering variety of interaction effects, so that whether the interviewer’s language or cultural background dominates in the explanation of the observed effects remains obscure. The question is whether the difference lies in the language I used in this case or is due to the difference in personality and/or cultural background between the informants and me (this difference is not shared by local interviewers). If language is determined to be the primary reason for the difference in interviewing outcomes, salvation for me can come simply through learning the maternal tongue. If, however, cultural differences are shown to be the primary cause of these systematic differences in responses given to me, two causes could be responsible: a difference in interviewer mission or my innocence of local cultural norms, both for discourse and/or the domain of food avoidances (implying that I, unlike local interviewers, couldn’t tell when I was being duped). If my performance is shown to have improved with experience, then increased experience of the cultural group being studied will be suggested as the remedy 180

REFLEXIVITY IS NECESSARY

for the observed ethnographic recording bias. Depending on which of these possibilities proves more significant, different courses of action will be argued more likely to save the ethnographic enterprise. So far, it appears that a difference between my modus operandi and that of the other interviewers is central to the explanation of the effects I have been discussing. I will therefore turn my attention to the possibility that my interviewing style was fundamentally different from that of the other interviewers, with the suggestion that this single factor accounts for all of the observed inter-interviewer, as well as interviewer-informant effects. I will also suggest that the proposed difference in interviewing style reflects a different perception of the purpose or mission of the study between my assistants and myself. First, it was my impression from difficulties during the training and supervision of my assistants that they did not fully understand the objective of my study: to investigate within-cultural variation. My assistants, like most anthropologists, seemed to believe that idiosyncrasies of individual knowledge were not worthy of serious study. I believe that my assistants tended to discount any perceived abnormalities or eccentricities among their informants, and emphasized what they knew to be correct, based on their own perception of the food avoidance system. They were often surprised to learn that their informants were unsure, contradictory, or ignorant of what they should know about important, culturally sanctioned food avoidances. Contrary to my mission in conducting interviews, they may have had their own self-appointed mission: to enforce orthodoxy. I therefore tested the hypothesis that differences between interviewer missions might result in systematic biases in reporting style. I assume that a systematic bias in interviewing style will manifest itself as significant differences in the total number of avoidances reported andor in the number of different types of avoidances recorded for comparable interviews. For example, my interest in variability should result in my paying greater attention to the vagaries of informant responses than my assistants. My interviews should therefore exhibit a greater diversity (number of different types) of responses. Similarly, when informants are more assertively questioned, they are more likely to produce a response including an avoidance than to make the easiest answer of “I have no restriction on consumption of that animal.” In addition, informants are more likely to be assertively questioned by an interviewer whose objective is to elicit greater variation 181

APPENDIX B

in beliefs. An increase in the total number of avoidances recorded in interviews conducted by an interviewer more interested in documenting variability should therefore also be noticeable. I therefore calculated the total number of avoidances in each duplicated interview, as well as the diversity of codes recorded by the interviewer, measured simply as the number of different types of avoidances appearing in a given interview (using a conservative thirteen-valued coding scheme). However, contra expectation, the average number of avoidances in the interview I conducted is no higher than the interview with the same informant conducted by one of my assistant^.^ The diversity of avoidance types is also no larger on average in those interviews that I conducted.8 Similar results hold even when the relationship is examined in a much larger sample of interviews. Either the Sudanic interviewer, K, or I conducted 278 independent interviews on individuals from the Sudanic ethnic group (not the same sample from which the replicated interviews were taken). If we control for variability due to the sex and age of the informants in this larger sample, there is again no difference between my interviews and those conducted by K in the total number of avoidances recorded during an interview, nor their diversity.’ I conclude that there is no systematic difference due to interviewing style discernable between interviewers, based on the two measures of style used here. However, there is considerable anecdotal information about variability in the degree of cooperativeness among informants from the personal impressions of the different interviewers. Each of them noted in their reports for at least one informant their belief that an informant was uncooperative. Most typical is the belief that the informant was simply trying to rush through the process as quickly as possible in order to get the payment reward at the end, and therefore replying in the most parsimonious fashion possible. The most facile response is of course to say that there is no avoidance against that food. This is a form of deception, since the context of the interview sets up the expectation that the interviewer wants information about what the informant truly believes, and has purchased the informant’s good will with money. I saw this happen myself. For this reason, with greater experience, the number of avoidances reported by an informant should increase. A difference in the assertiveness of the pair of interviewers during questioning of informants (mentioned above) is most notable as differ182

REFLEXIVITYIS NECESSARY

ences in the probability of eliciting an avoidance from individuals at either end of the age range: children won't produce an avoidance response without considerable prodding and coddling, while the elderly tend not to remember many avoidances that are no longer current for them unless actively reminded of that possibility; so this effect is measured as an interaction between the pair of interviewers and the age of the respondent. I investigated the possibility of differences in assertiveness by looking for an order effect within each set of interviews conducted by an interviewer to see if those interviews conducted later exhibited greater numbers of reported avoidances. (This was done after controlling for the nuisance variation due to demographic characteristics of the informants.) I therefore used a dummy regression model of the number of avoidances in all of the interviews conducted by an interviewer (not just the duplicate interviews). Two-way interaction effects due to demographic characteristics of the sample of informants were also included in order to eliminate any such effects. I also controlled for an informant's degree of education." I found that, as expected, there is a significant increase in the numbers of avoidances recorded in those interviews that occur later in the sequence (t = 2.98; p = .003). Indeed, interviewing practice was the statistically most significant determinant of the number of avoidances reported by an informant in this model. With practice, it appears there was an increase in the degree of assertiveness by some interviewers that to some extent decreased the reports of no avoidance once the interviewer became more experienced. Informants became less able to lie about their avoidances as their audience became more practiced and therefore more critical. I showed above that the likelihood an informant would report different avoidances also decreased as the experience of the interviewer increased. Whether this is due to an increase in skill of the interviewers or a decrease in the ability of uncooperative informants to lie cannot be discriminated. However, in some sense the interview is negotiated and, if we assume that informants interviewed later in sequence are not artificially boosting the number of avoidances they actually believe in as they are pressed by more assertive interviewers for more complete responses to each question, then the lower number of avoidances reported by informants getting less experienced interviewers must be taken as a violation of the implicit contract to report all their avoidances (all informants were paid for their participation in the study): earlier informants are taking shortcuts in order to reach the goal without paying their dues. 183

APPENDIX B

Even if there were no overall discernable pattern of difference in the use of categories of response between interviewers, there still remains the possibility that my interviewing practice was more closely approaching that of the other interviewers with greater experience. I showed above that replicated interviews conducted by more experienced interviewers (in the sense of having already performed a greater number of interviews) include fewer episodes of forgetfdness or mistake making by informants, suggesting a training or practice effect for interviewers. I argued this training effect might be related to an increasing assertiveness with experienceletting fewer informants get by with the easy answer of “I have no avoidance.” However, the results also clearly indicated that, despite increased experience, I still had not reduced my proportion of discrepancies to the level of the other interviewers. However the sample of interviews from which this conclusion was drawn was quite small (67 total interviews). I will therefore turn my attention again to the larger sample of interviews conducted by the same set of interviewers from the larger project of which the replicated interviews are a part. Using this sample, I will attempt to estimate the “rate of approach” of my interviews to the discrepancy pattern of the interviewers who had the benefit of being members of the local cultural group. An examination of the pattern of interviewing style in this larger sample might provide an indication of whether my salvation lies in greater experience of the culture. Otherwise, I will be forced to finally conclude that someone from outside the cultural group can never effectively elicit beliefs “like a native” (Goodenough‘s famous ideal for the ethnographer). Presumably, with increased experience I would exhibit greater changes in the measures of interviewing style (the number and distribution of avoidance reports) than the other interviewers because I started with disadvantages: lack of knowledge of appropriate responses and perhaps language difficulties. I used multivariate regression to estimate the effect of Interviewer Experience on the two measures of interviewing style, after controlling for background demographic variation due to ethnic group (dummy variables for Sudanic versus Bantu, and forager versus horticulturalist), sex, and age, Experience was measured as previously: the number of interviews previously conducted by that interviewer. For all the interviews I conducted (N = 272; the sample coming from the complete set of singleton interviews on informants), both measures of interviewing style 184

REFLEXIVITY IS NECESSARY

increased with experience.ll For my assistants (N = 177), on the other hand, both measures significantly decreased with experience.l2 I conclude that the effects of experience observed in the interviews conducted by my assistants reflects the reduced degree of supervision I was able to provide during the latter period of their work (having moved to another area), so that their later interviews became more perfunctory. The opposite trend observed in my interviews is probably due to increased assertiveness as my experience increased. I believe the results presented in this section present a consistent picture. Before I explain this principle, let me recapitulate the main points concerning interviewer effects shown thus far: Discrepancy rates decrease as interviewers become more experienced, but the improvement in interviewing technique is less evident in interviews I conducted. There is no overall difference in interviewing style among the set of interviewers in replicated interviews, nor in the larger sample of interviews on Sudanic informants. Systematic changes in style as a function of Interviewer Experience are evident in the larger sample, but there is an opposite trend between other interviewers and me.

All of these results argue that interviews I conducted are significantly different in a number of respects from those conducted by interviewers who were also members of the local groups. The hypothesized different missions in this case produced no systematic biases in the reporting of responses, since all interviewers seem to be coding the same number and types of avoidances for each informant. Rather my intention to document variation in this system (and probably greater propensity to become bored by a methodical repetition of the same formal interview format) produced comparatively high discrepancy rates between replicated interviews, while the dogmatic and methodical approach of my assistants resulted in greater uniformity in their duplications of interviews. This difference between my mission and the uniformitarian one of my assistants could explain how the same interviewer repeating the interview on a single informant could have higher rates than different interviewers: the latter group of interviewers were more methodical in their interviewing 185

APPENDIX B

technique (no stylistic differences). Thus my differing mission resulted in the reduced benefit of experience with respect to discrepancies (the first result above). The second result, no stylistic differences between interviewers, signifies that, in the set of interviews conducted by any interviewer, no significant differences could be found in the use of the different avoidance codes (number or different types). This is a measure of between-interview comparability among interviewers, while difference in mission explains the within-replicated interview pattern. Thus the combination of mission and style reconciles the first result above with the second one. However, the third result also suggests that significant variability in the elicitation of responses (not the recording of responses) derives from inter-interviewer variation in assertiveness with experience. Assertiveness is an aspect of questioning technique and thus results in significant changes in reporting style, since different kinds of responses are elicited from the same informant with different degrees of prodding. For the third result to be reconciled with the first, it must be the case that assertiveness does not result in more consistent responses, and thus does not reduce the frequency of discrepancies. Is this a reasonable conclusion? I suggest that it is, because prodding need not lead to the same result, but can follow alternative pathways, depending on the situation. Thus a more assertive interviewer, such as the anthropologist concerned to document variability, can indeed produce a greater proportion of discrepant responses from the same informant, despite greater experience at detecting culturally inappropriate responses. It still might seem that I have not excluded the possibility that language is central to the difficulties informants appear to experience when I interviewed them. If the change in style with experience is due to my improved language competency, the implication is that language proficiency results in increased numbers and a wider variety of reports of avoidances from informants. However, if the problem was difficulties in inter-personal communication due to language competence, the probability of discrepant responses should decrease with experience more significantly for me than for the other interviewers, who conducted their interviews in the maternal tongue shared by both interview participants. However, this was not observed in the analysis, where the discrepancy rate was shown to improve fess rapidly for interviews I conducted than for the local interviewers. 186

REFLEXIVITY IS NECESSARY

There are other reasons to discount language as an explanation for the pattern of results shown in this chapter. First, when I began doing the interviews in this study, I already had one year of experience in the study area. By the end of the study, I had been on-site for twenty-one months. In addition, my proficiency in Swahili was as good as that of many of my informants. Since I concentrated most of my attention on the food avoidance system, my knowledge of that system probably exceeds what could be expected of an ethnographer that did not specialize in a similar fashion. Second, the decrease in the measures of interviewing style exhibited by the local interviewers with experience cannot be explained by a reduction in their language proficiency. I therefore argue that inter-cultural communication problems cannot account for the pattern of results shown above, and is unlikely to contribute significantly to the higher rate of discrepancies observed in the set of interviews I conducted. I believe both my and the other interviewers’ temporal patterns in style result not from language problems, but from changes in the degree of assertiveness that each interviewer was using over time. After all, Bernard (1988,211), summarizing the extensive literature on interviewing style, says, “the key to successful interviewing is learning how to probe effectively.”

Inter-Interviewer Dissimilarity

Disparities between the personal characteristics of different interviewers may influence the kinds of responses that they record. They may have individualized interviewing styles that result in biases in the types of responses they report. When such biases are combined, as they are by pairing responses from two different interviews of the same informant, discrepancies may reflect these biases in recording style rather than any substantive difference in what the informant said. Inter-Interviewer Dissimilarity is a measure that combines aspects of personality as well as cultural differences among the set of interviewers who conducted the interviews in the present study.13 Due to the research design, a significant effect of Inter-Interviewer Dissimilarity primarily indicates that the kinds of responses I recorded are different from those recorded by the other interviewers. This allows me to determine whether my interviewing style was substantially different from that of the local interviewers. The statistical models indicate that Inter-Interviewer Dissimilarity is a somewhat more significant factor for mistakes than forgetfulness (11% 187

APPENDIX B

of explained variability for mistakes; 5% for forgetfulness). But note that Inter-Interviewer Dissimilarity and Interviewer Experience together constitute between 10% to 15% of variability in both models. Whereas the Experience effect is relatively large and that of Dissimilarity small for forgetfulness, with respect to mistakes, the relative significance of these two factors is reversed. It is possible that the results from both models reflect greater assertiveness in interviewing style with increasing experience. Forgetful discrepancies probably decreased over the course of the study as all of the interviewers became more experienced, and hence more assertive in eliciting avoidances (i.e., allowed fewer facile “no avoidance” reports to constitute a forgetful discrepancy). However, as all of the interviewers became more assertive, their stylistic differences also diminished, leading to more similar recording procedures. Thus, a greater proportion of the variability due to interviewer style was allocated to Experience in the model of forgetfulness, and to Dissimilarity in the case of mistake making. Forgetful discrepancies are reduced less by Interviewer Experience when I was one of the interviewers. It therefore appears that the combination of me with another interviewer is more likely to lead to two different avoidance reports than to a report of an avoidance coupled with a report of no avoidance. This pattern suggests that the local interviewers became proportionally more assertive with experience than I did. In addition, the combination of me with young informants produced more variable responses than interviews with informants of similar ages conducted by the local interviewers. This suggests inordinate communication or interpersonal difficulties between me and more naive informants.

Informant-Based Effects In this section, I discuss variables that measure knowledge and recall ability variation among informants. Several of the variables identified in the models-an informant’s Ethnic Group, Sex, Age, and level of Schoolingwill be grouped together in the following discussion as “social roles,’’ since each identifies a way in which people are classified socially. The other sources of variation I have identified as representing variability due strictly to informant characteristics are: “Question-Specific Response Bias” (including Animal Familiarity and Animal Group variables), Cognitive Load, and Elapsed Time effects (each of these variables is defined in table B.3). 188

REFLEXIVITY IS NECESSARY

I argue that only social role variation is related to knowledge differences between individuals in this population, while these other informant-based effects reflect various aspects of recall ability. I will first deal with the informant-based variability that is due to differences in recall ability. While such effects have intrinsic interest fiom a psychological perspective, they are included here for a methodological reason: to isolate recall effects from knowledge effects. There are three ways in which informant performance during interviews can influence the probability of a discrepancy but which are unlikely to reflect substantive differences in belief about the same topic. First, interviews included questions about different topics (i.e., avoidances concerning different animals). As a result, the estimated probability of a discrepancy is calculated across responses to different questions. Any question-dependent variation must be isolated before any questionspecific variation in this population can be uncovered. Second, the informants’ ability to recall their avoidance of a specific animal may be affected by the total number of avoidances they have to remember. This can influence the base rate probability of a discrepancy. Third, there may be learning of new avoidance-related information between interviews that substantively affects what informants believe about a certain topic. This intra-informant variability in belief about a certain topic must be distinguished from inter-informant variability about that topic. Since the types of avoidance associated with particular animals differ and particular kinds of food avoidances have greater cognitive salience than others, responses with respect to some questions have a higher probability of being repeated. For example, some types of food taboo, when violated by consumption of the restricted animal, are supposed to cause the perpetrator to die soon thereafter. Beliefs that carry such dangerous implications should be more difficult to forget than more personal avoidances such as a belief that consumption of an animal, if it is too rotten, causes stomachache. Question-Specific Response Bias

In this section, I discuss a variable included to account for the fact that a number of different questions are used simultaneouslyto estimate the models of discrepancy. Before I can examine variability in belief with respect to individual questions, I must control for the variability in the probability of a 189

APPENDIX B

discrepancy that arises because each question in the interview involved a different topic. Inclusion of variables specific to individual questions is therefore intended to control for the salience of the question. In the next two sections, I discuss variables that control for methodological differences in the salience of responses. Typically, anthropological studies lack explicit recognition of the varying ability of different questions to elicit reliable responses, although the literature in cross-cultural psychology (also concerned with cognitive ability testing) certainly testifies to the importance of such effects, even though they are more difficult to measure in the cross-cultural context (Laboratory for Comparative Human Cognition 1982). Cultural Consensus theory explicitly assumes such effects are insignificant (although Batchelder and Romney 1989,239, argue that the model is robust against violations of the assumption of homogenous question difficulty). However, this ignorance of question-specific response effects leads to considerable interpretive diffi~u1ties.l~ Psychometricians typically use multiple questions to determine the single variable of interest to them: informant cognitive ability. The response to a single question is obviously insufficient to measure abilities like intelligence. O n the other hand, in the ethnographic interview, each question is typically designed to elicit a different independent variable (e.g., a belief). In this case, there is a one-to-one correspondence between the question and the variable to be measured, rather than many-to-one, as in the psychometric case. This causes a difference in the way variability in the responses to particular questions is interpreted. From the psychometric point of view, a question that elicits greater variability in response among informants is viewed as being more difficult (because there is a single correct answer). From the ethnographic point of view (where the answer key is unknown), variability in responses among informants indicates one of several things, depending on the viewpoint of the ethnographer. If the ethnographer believes in the legitimacy of intra-cultural variation, different responses to a single question are interpreted as variation in what different individuals believe with respect to the topic of the question. If, on the other hand, the ethnographer believes cultural processes act at the group level, such differences will be interpreted as spurious contextual or methodological variability. This confusion has been at the center of much of the dispute about the existence of intra-cultural variability in belief (Boster and Weller 1990,172). 190

REFLEXIVITY IS NECESSARY

When each question in ethnographic interviews is used to elicit a different variable, then question difficulty cannot be used to determine question-specific reliability. Effectively, a single question is being asked to measure two things: the informant’s knowledge related to the topic of that question, as well as their ability to respond effectively under a particular set of circumstance^.^^ At least two measurements are required to disentangle these two effects: the stimulus (question) must be repeated with that informant, using some means to isolate whether the reliability of the response is due to that informant’s knowledge or to their ability to recall that knowledge. The degree to which a question consistently elicits the same response from an informant under a variety of conditions (i.e., its salience to that informant) is a measure of the informant’s ability to reliably recall the information contained in the response. Question-specific salience is thus a reasonable measure of the impact of cognitive ability on response to that question. It is also a relative measure, like a question’s difficulty. However, salience compares variability of response within a question (i.e., among trials), rather than between questions (from a single trial). Repeated trials can also determine what knowledge an informant has with respect to the topic of the question. For example, if an individual repeatedly produces non-normative responses to a question in various situations, it can be concluded that this is not purely spurious variation in belief. Thus, determining question salience validates data because it provides the researcher with information about the quality of responses. If variability in responses remains after consideration of interviewer-based and contextual effects, and this variability correlates with characteristics of informants (such as social roles), that variability can be unequivocally assigned to differences in knowledge and attitudes about the subject of that question.16 Thus, perhaps the major methodological question introduced by variability in primary data-whether there is substantive variation in cultural beliefs among informants-can be answered by repeated questioning of individual informants, since it allows the determination of question-specific reliability. Similar modifications to other data collection regimes can alleviate observation-specific reliability problems for those kinds of data as well (e.g., controlled variation in experimental conditions for behavioral observation). The two variables designed to account for Question-Dependent Response Bias are Animal Group and Animal Familiarity. The first, which simply identifies an animal’s ethnofunal group, reflects the different kinds 191

APPENDIX B

of roles animals play in the culture (i.e., their cognitive markedness). Animal Familiarity, on the other hand, deals with the perceptual salience of each animal (i.e., whether the informant has ever seen the animal). Both variables are measures that control for the different probabilities of an answer being repeated for questions of different salience to informants. As expected, these measures of a question’s salience prove to be the most important determinants of the probability that informants will make both mistaken and forgetful responses (28% of explained variability for forgetfulness; 43% for mistakes). Even though tests of cognitive performance have not been undertaken on the individuals in this study, information concerning an informant’s recall abilities can nevertheless be indirectly measured by their performance during food avoidance interviews themselves. Interviews in this study involved questioning informants with regard to their avoidances of specific animals. Different animal names might have different degrees of salience, or abilities to prompt informants to recall any avoidances they might have with respect to that animal. Different abilities to recall avoidances should appear statistically as different probabilities of repeating (i.e., of not making discrepant) responses. This probability, which partly reflects the different cognitive characteristics the animal name has for informants, I w i l l call the Question-Specific Response Bias. Salience is an amalgam of the degree of emotional attachment and/or certainty attached to a belief, as well as other features indicating its cognitive importance, all of which should lead to different degrees of recall. I hypothesize that there are two dimensions to this saliency. The first measures the degree of familiarity of the animal to the informant. For example, more familiar animals (i.e., those caught more frequently), present individuals more often with the need to recall any conditions on their edibility. I therefore include in the model a measure of the familiarity of an animal (a binary-coded variable indicating whether or not the informant reported having seen the animal alive or dead). The other dimension of stimulus saliency I hypothesize reflects less an animal’s perceptual familiarity to the informant than its generalized cognitive “markedness.”For example, more dangerous animals (e.g.,those that prey on humans), and/or more anomalous animals, may also be more “marked’’(scnsu Saussure 1974). In addition, the kinds of cultural uses to which different types of animals are put might vary (e.g., the place of the animal in folktales or the use of certain animal products in ethnomedical 192

REFLEXIVITY IS NECESSARY

treatments). This more generalized salience is measured by membership of the animal in a linguistically named (marked) category in the ethnofaunal taxonomic system.l’ Where Animal Familiarity, the first measure of salience or question-specific bias, is a measure of an informant’s knowledge or experience of an animal, Animal Group effects are a hnction of how that knowledge is organized (i.e., the hierarchical structure of knowledge). Controlling for variability in question-specific saliency has been the primary concern of psychometricians, especially those measuring intellectual performance in an educational context (e.g., standardized testing for college entrance examinations-see issues of the Journal f o r Educational Measurement or the Journal for Educational Statistics for examples). Researchers in this area call this effect “item bias,” due to an informant’s reduced ability to formulate the correct response to test items of greater difficulty (measured simply as the proportion of informants who get the question right; see e.g., Hambleton et al. 1991).lS Psychometricians assume that questions in a test probe a single cognitive faculty in each respondent. Variability in the probability of a correct response is a function of the question’s difficulty, which is unknown a priori (but assumed equal for all respondents). The overall ability of a particular respondent to answer the set of questions correctly is her or his measure of intelligence. Psychometricians must control for variability in the difficulty of specific questions when calculating this summary measure of a respondent’s cognitive ability. Cross-cultural psychological studies in general have also been stymied by the fact that differences between cultural groups in the average level of performance on cognitive tests appear to depend primarily on the cultural relevance or familiarity of tasks in the cognitive test, rather than any systematic difference in cognitive ability between groups (Laboratory for Comparative Human Cognition 1982, 687). In particular, some studies suggest that stimuli that are more familiar to informants elicit valid responses more reliably (e.g., Cole et al. 1971; Okonji 1971). These two literatures suggest it is imperative to include Question-Specific Response Bias effects in any study of variability in cognitive performance. However, my approach is somewhat different from that of the psychometricians and cross-cultural psychologists, since I am more interested in the difficulty of the question than in measuring an informant’s generalized cognitive ability. Nevertheless, as in these other types of studies, 193

APPENDIX B

both the difficulty of the question and the cognitive ability of the informant are unknown a priori. Since the question in the present case didn’t vary (Do you have any avoidances of animal X? was consistently used), the salience of the question cannot be due to the phraseology of the question itself. Rather, differences in the difficulty of responding to this question must be the cause of any variability in the reliability of responses. Note that difficulty in the present case has a different meaning from that used by psychometricians and cross-cultural psychologists: it measures the ability of a question to elicit replicable responses (i.e., no discrepancy) on two occasions, rather than a correct response on a single occasion. This difference reflects the fact that, unlike questions in standardized cognitive tests, there is no correct answer to questions regarding personal beliefs.19 The results show very strikingly the reason why psychometricians and cross-cultural psychologists pay such close attention to this type of bias in the measurement of informant cognitive performance: Question-Specific Response Bias is the most significant category of effect in both models of discrepancies. By far the single most statistically significant relationship between any covariate and either of the two measures of discrepancy is between Animal Familiarity and mistakes (29%of all explained variability): having seen an animal very significantly reduces the probability that two different avoidances will be mentioned with respect to that animal. In addition, the largest changes in the odds of a discrepancy occurring (measured by Exp [B]) tend to be associated with the categories of Animal Group (see tables B.4 and B.5). Thus, by either criterion of significance, Question-Specific Response Bias effects are the strongest exhibited in these models. The results also show that these two measures interact significantly. For example, the probability of making a mistake with respect to viverrids (mongooses and genets) and birds is much less benefited by familiarity with the species in question than for other animal groups, probably because such animals are difficult to discriminate visually, due to their very similar morphotypes. Thus, the perceptual familiarity of an animal appears to improve an informant’s ability to repeat their previous responses, but only when the species in that group are relatively distinct morphologically, and associated with relatively few, culturally important avoidances. In sum, the results clearly indicate that the probability of discrepant responses is significantly affected by the salience of the question to the informant. Note that these variables do not measure the salience of the an194

REFLEXIVITYIS NECESSARY

swer, but of the question: the likelihood that an informant will repeat an answer depends not only on what kind of answer they gave the first time, but on how consistently the question prompts similar response-making cognitive processes such as recall and reasoning abilities (i.e., on the question’s degree of difficulty). The inclusion of question-specific effects is thus a way of determining the degree of trust to place in interview data, since Question-Specific Response Bias is a measure of the relative degree of reliability in interview responses to specific questions (White 1990). The present results indicate that the general lack of consideration for question-specific effects in the cultural anthropological literature leads ethnographers to mistakenly believe all of their data is equally reliable.20

Cognitive Load: Recall Variation

Whereas Animal Group and Animal Familiarity are included in the statistical model to control for variability in salience between the questions used to elicit responses, another variable, Cognitive Load, controls for methodological differences in the probability of discrepant reports between informants. It measures the number of avoidances that an informant must store in memory. As Cognitive Load increases, I assume the probability of being able to recall any particular avoidance decreases. The results suggest this is exactly what happens: informants with relatively high Cognitive Loads are more likely to report two different avoidances with respect to a given question (i.e., more likely to make a mistake). In addition, those with larger stores of avoidances are less prone to make “no avoidance” reports (i-e., less likely to forget a previously reported avoidance). Cognitive Load accounts for about 19% of explained variability in discrepant responses. Indeed, as table B.6 shows, Cognitive Load and Question-Dependent Response Bias together account for about half of all explained variability in the probability of a discrepancy. Another aspect of an informants’ recall ability might be indirectly measured by their ability to search through the knowledge they have stored in memory. In particular, I assume that if informants have a greater store of avoidances to remember, the likelihood they will get confused by a particular stimulus and report an incorrect avoidance will increase. Therefore, the more avoidances individuals have to remember-i.e., as their Cognitive Load increases-the higher their probability of reporting discrepancies will be, ceteris paribw21 195

APPENDIX B

The results from the multivariate statistical models indicate that Cognitive Load is one of the most significant variables afTecting both forgetfulness and mistake making (see table B.6). Indeed, one of the most striking results of the analysis is that Cognitive Load has a significant independent effect, but in opposite directions for the two different dependent variables.22As an informant’s Cognitive Load increases, the likelihood different reports of avoidances would be made on two different occasions increases (18% of variability), but the probability of forgetting to mention an avoidance decreases (20%).Cognitive Load therefore seems to act like a pool or store of knowledge of avoidances from which the informant draws responses randomly. From this perspective, an increase in Cognitive Load reduces the probability that an informant will report “no avoidance”since he or she will have a larger store of avoidances from which to choose a response. However, the more avoidances individuals have, the more likely it is they won’t mention the same avoidance as a response on two separate occasions: the larger their pool of responses, the more likely they are to draw out two different types of response. More precise insight into the difficulties of recall comes from the results involving interactions with Cognitive Load. As an informant’s total number of food avoidances increases, the problem of locating the appropriate response also seems to increase. For example, given a particular amount of time between interviews, a higher Cognitive Load increases the probability of a mistake. In addition, as the total number of avoidances an individual must remember increases, the familiarity of an animal becomes less effective in reducing their forgetfulness. A similar kind of mental confusion can explain the pattern of interaction between Animal Group and Cognitive Load. For example, when an informant’s total number of avoidances increases, they become less likely to forget that there is an avoidance associated with a particular viverrid (genet or mongoose), but more likely to mistake the type of avoidance associated with that species. How can this opposition be explained? There are two kinds of avoidances associated with viverrids. Both types of avoidance tend to be associated with the entire group (the species in this group are very similar in appearance, so that applying a single avoidance rule to all the species in this category is one way of getting around the necessity of telling the animals apart). However, the ability to remember which of the two avoidance types applies to this group is reduced by having to remember a larger number of other avoidances. This combination of effects 196

REFLEXIVITY IS NECESSARY

with respect to viverrids for forgetfulness and mistake making again suggests a degree of mental confusion. In sum, Cognitive Load appears to reflect the increasing difficulty of searching through a larger and larger number of avoidances stored in long-term memory for the best response to a given question. The ability to recall the appropriate response (i.e., locate a particular avoidance) seems particularly sensitive to vagaries in the organization of an individual’s knowledge base-not just with respect to food avoidances themselves, but other kinds of knowledge as well.

Elapsed Time Effects: Memory, Learning and Experience

The third variable introduced into the statistical model representing an aspect of recall ability is the length of time between the informant’s first and second interviews. I expect that longer lengths of time between interviews will cause the memory of previous interview responses to fade. As a result, the propensity among informants to make discrepant responses should increase as the elapsed time between interviews increases. One of the benefits of the model design used in this study is that it allows an explicit treatment of the cognitive processes that vary with the length of time between interviews. In the cognitive psychological literature, elapsed time effects are widely used to measure aspects of cognition, ranging from intelligence to categorization tasks (for a review see Bower and Clapper 1994). The results in the present case show that as Elapsed Time increased, the probability of a different kind of avoidance being mentioned in the two trials also increased (2%of explained variability). However, contrary to expectation, Elapsed Time did not affect the probability that a previously reported avoidance would be forgotten at the next opportunity. In general, temporal effects are small in comparison to other kinds of effects. The elapsed time between interviews confounds several somewhat different factors: (1) short-term versus long-term memory effects; (2) changes in interviewer performance due to experience interviewing other informants during the interim; and (3) informant learning, either from individual experience with foods or the cultural transmission of food taboos, that occurred during the interim between interviews. A difficulty in determining whether informant beliefs are different within a cultural group is that a single informant’s beliefs may change over 197

APPENDIX B

time. In the present case, this is most likely to result when informants learn new food avoidances (the third case above). Informants’ behavior was not monitored between interviews in order to know what sorts of food-related experiences might have affected their responses. Nevertheless, one way to check on the possibility of informant learning is to compare the proportion of “no avoidance”reports in second interviews to the proportion during first interviews. If the learning of new avoidances during the interim between interviews is insignificant, then “no avoidance” reports should occur with equal frequency during the first and second interviews. A significant informant learning effect, on the other hand, should appear as a bias toward forgetful responses being made during the first interview. A cross-tabulation comparing “no avoidance”reports to the frequency of reports of an avoidance (regardless of type) shows that “no avoidance”reports are made with much higher frequency during the first interview than the second (Phi coefficient = .0413; p = .OOS). This suggests that informant learning may be responsible for a significant proportion of observations that I have called forgetful. However, a greater proportion of avoidance reports in second interviews is also consistent with the second possibility noted above: an improved ability among interviewers with increased experience (conducting the later interview of an informant) to elicit avoidances from informants. Interviewer Experience is a much less significant factor of mistakes (where the Elapsed Time effect appears) than of forgetfulness. Nevertheless, it is possible that interviewer learning, rather than informant learning, can account for at least part of the observed increase in different avoidance reports during second interviews with informants. However, the fact that the Elapsed Time effect was significant only in mistake making indicates that in their second interviews, informants were not reporting newly learned avoidances instead of the “no avoidance” report from their first interview. Rather, it appears that interviewers were being more assertive during later interviews and probing for new types of avoidance reports, reflecting their greater experience at eliciting avoidances. Finally, on average, less than four months transpired between an informant’s first and second interviews. It seems unlikely that learning of new avoidances in such a short period of time can account for much of the significantly reduced fraction of “no avoidance” reports. Familiarity with food items typically doesn’t change that rapidly. I will therefore assume that informant learning between interviews is insignificant; instead, 198

REFLEXIVITYIS NECESSARY

it is more likely that interviewer experience or learning is responsible for the observed effect. The third possibility is that the observed Elapsed Time effects reflect short-term or working memory effects (which plague repeated measures analyses). A short-term memory effect requires that the informants remember their responses to particular questions from the previous interview, and respond differently during their second interview as a result (e.g., by repeating the previous response rather than searching long-term memory afresh). However, it is known that after about 30 seconds, information leaves working memory, and is either lost or stored in long-term memory. Forgetting thereafter occurs at a much slower rate over minutes, days, and even years (Potter 1990,20). Since the mean time between interviews is about four months, any short-term memory effect that could extend over this length of time should be relatively weak. Certainly, the relatively small effect of Elapsed Time suggests this is the case. It is also possible that the proportion of interviews occurring close together in time is sufficient to produce this result. This short-term memory effect can therefore contribute to the Elapsed Time effect, along with an effect due to Interviewer Experience, as suggested above. However, the results clearly indicate that these Elapsed Time effects-whether due to Interviewer Experience, a short-term memory effect, or both-are relatively unimportant. Perhaps it is not surprising that the ability to recall food avoidances is not particularly affected by the passage of time. Story-telling specialists in oral societies such as those living in the Ituri have exhibited surprising feats of memorization (Neisser and Hyman 1999).The inability to rely on information storage devices such as books in oral societies appears to promote individuals’ abilities to store necessary cultural facts in long-term memory. For example, the ability to recall aspects of stories is better among the Setswana of Botswana than rural Americans, who are presumably less oral (Dube 1977). Cultural beliefs that have dangerous social andor personal consequences may also be supported by social pressures that provide additional inducements to memorability. Many informants in the Ituri can recall specific events of transmission of food avoidances from a culturally prescribed authority figure, usually at the first acquaintance with the animal involved in the avoidance. There are indications from cognitive psychology that a special form of memory is involved in the remembering of such autobiographical events (Rubin 1986). This form of 199

APPENDIX B

memory may increase recall ability, stabilizing interview responses that invoke autobiographical memories (Bradburn et al. 1979). Thus food avoidances appear to be effectively stored in long-term memory, so that the passage of time does not significantly affect the cognitive processes that lie behind discrepancies in such cultural beliefs. The overall lack of significance of Elapsed Time effects relative to others corroborates the results of Foster (1979, 18l), who found that even very long periods of time between interviews (up to fourteen years) had no appreciable effect on rates of within-informant discrepancy among rural Mexican informants with respect to humoral (“hodcold”) beliefs related to food. It therefore appears that ethnographic studies need not in general explicitly account for the time between interviews when estimating discrepancy rates for strongly held cultural beliefs (although a very long time between interviews would probably introduce a confound of individual learning). Even when replicated interviews are not used, this analysis suggests that variability in responses with respect to cultural beliefs is unlikely to strongly reflect purely temporal or ontogenetic variation (as also suggested by the low significance of Age in this analysis). Other sources of variation in responses are much more important. However, methodologically, repeating interviews with informants is crucial to the isolation of situational effects on responses in interview data.

Social Roles: Knowledge Variation

The types of experiences individuals undergo over their life history should affect the kinds of knowledge that they acquire. An important determinant of the kinds of experiences individuals will have is the social roles that they play. The roles that are widely used in the anthropological literature are ethnic group, sex, age, and education. Because these roles are easily identified as morphological characteristics (except the last), they are significant cues for a wide variety of social processes. Each of these roles has therefore formed the foundation of an extensive literature in the social sciences (e.g., on ethnicity, gender, ontogenetic processes, and acculturation). The first indication that there might be variability in informant knowledge is that different social role effects appear in the best-fitting models of forgethlness and mistakes. Ethnic group (accounting for 8.9% 200

REFLEXIVITYIS NECESSARY

of explained variability in discrepant responses)23and Age (7.1%)are significant predictors of mistake making, while Sex (12.3%) is a significant predictor of forgetfulness. Efe foragers are significantly less likely to make mistakes than Sudanic horticulturalists, while Bantu horticulturalist~~~ exhibit a rate between these other two groups. Men, regardless of Ethnic Group, exhibit a significantly higher probability of forgetting a previously mentioned avoidance than do women. This suggests that women, when they report an avoidance on one occasion, are more consistent in reporting an avoidance on the other. An informant’s age also has an effect on their ability to repeat particular types of answers. Younger individuals, regardless of Ethnic Group or Sex, make more mistakes than older individual^.^' However, Lese show greater reductions in the probability of making mistakes as they get older than do foragers, even after accounting for ethnic differences in the tendency to make mistakes. The number of years an informant has spent in school also affects the probability they will make discrepant responses in interviews: informants with more education are less likely to forget a previously mentioned avoidance the second time they are asked the same question. I suggested above that social roles limit the kinds of knowledge to which an individual can gain access. Other work (e.g., Garro 1986; Kempton 1981) also suggests that what individuals know is a function of their social role (typically considered as “expertise” in some domain of cultural knowledge). For example, Jivaro horticulturalist women are better at identifylng plants than men, who spend less time gardening (Boster 1985). However, there have also been a variety of arguments in the literature suggesting that social roles in fact lead to differences in other aspects of cognition, such as reasoning ability. For example, some anthropologists argue that forager reasoning or “mentality”is different from that of horticulturalists. Peter Gardner (1991) argues that foragers have evolved a unique psychological syndrome, reflecting their reliance on hunting and gathering for subsistence as well as the relatively small, unstratified, and flexible social groupings that can be supported by such a lifestyle. This “individualautonomy syndrome”is characterized by an egalitarian social ethic, coupled with socialization for individual decision making and self-reliance, as well as considerable interpersonal cognitive variability (Gardner 1991,549). In the present study, however, the lack of any interaction between Ethnic 201

APPENDIX B

Group and other cognitive measures (discussed below) argues against the significance of mentality or reasoning differences between ethnic groups in the determination of discrepant interview responses. The likelihood that an Efe forager will make different avoidance reports is lower overall than among horticulturalists, but also changes less with Age. This is perhaps because foragers have greater knowledge-both of animals and the avoidances associated with those animals-at earlier ages. This may arise because of their greater dependence on the forest for subsistence (Efe trade meat to horticulturalists in return for garden produce, as well as eating it themselves-see Bailey and DeVore 1989). It may also be that foragers pass on their knowledge of animal lore (including cultural beliefs involving those animals) to their children at an earlier age, as an important part of subsistence training and as a reflection of the greater cognitive salience that animals have for foragers. It therefore seems these Age effects more likely represent differences between the two subsistence groups in their knowledge of food avoidances at a given age than differences in reasoning, with this knowledge difference reflecting the greater salience of animals in the everyday life of foragers compared to horticulturalists. Age has also been argued to affect general features of cognition rather than simply the acquisition of increased knowledge with experience. Researchers have argued that in the course of cognitive development, children undergo a qualitative shift in reasoning, from a style more reliant on instances, definitions, and perceptual features to one featuring more elaborate theories about the way the world works (e.g., Gopnik and Wellman 1994). This change reflects the reorganization of knowledge into more complex hierarchical structures. However, the sample of individuals in this study are all older than the age at which this hypothesized transition in reasoning style is supposed to occur (the youngest informant is nine years of age, while the subjects of these child development studies are typically less than six). It seems more likely that the observed age effect can be interpreted as older informants having greater (and longer) familiarity with the system of food avoidance beliefs and, therefore, a greater ability to respond consistently. Does a person’s sex correlate with differences in cognition?There are several studies in the cross-cultural psychology literature that suggest there is relatively little psychological differentiation between the sexes in 202

REFLEXIVITY IS NECESSARY

traditional anthropological populations.26 For example, Van Leeuwen (1978) suggests, based on cross-cultural results of cognitive style (i.e., performance on standardized embedded figures tests), that the relatively small differences in cognitive performance between the sexes in foraging societies reflects the lack of diversity in social roles (everyone spends most of their time in foraging-related activities). Similarly, Barry, Child, and Bacon (1959) found in a cross-cultural study of child-rearing practices that, in societies that accumulate little food (i.e., subsistence-level groups), there are fewer differences in the socialization of girls and boys than for more intensively productive societies. The results in this study indicate that women are less likely than men to forget an avoidance, but equally likely to make a mistake. Since forgetfulness requires that informants reply that they have no avoidance against a particular animal, this pattern of results may simply reflect the fact that women in these societies have more avoidances on average than men (Aunger unpublished data). However, any effect from differences in the likelihood to report avoidances due to having a larger number of them is controlled for by the variable Cognitive Load discussed below. Hence, other explanations for Sex-based differences must be sought. Significantly, Sex interacts with Cultural Differences 1and 2, which are measures of informant-interviewer interaction (also discussed below). As I argued above, Sex appears in the model of forgetfulness probably because women react differently to male interviewers than men. Neither this result, nor the cross-cultural literature in general, suggests that there are sex differences in reasoning or recall abilities in this population. However, Aunger (unpublished data) shows that men and women within ethnic groups have differential knowledge of food avoidances because they report different numbers and types of food avoidances. Schooling is another case where the existing literature suggests that the educational experience produces a change in reasoning rather than knowledge. Schooling has consistently been found in the cross-cultural literature to exert a significant influence on Western-devised psychological tests of cognitive ability (Irvine and Berry 1988,28).In Aunger (1996), I show that schooling increases an individual’s knowledge of the food avoidance system: schooled individuals report a larger number and wider variety of avoidances than do the less educated. These results suggest that schooling improves the ability of individuals to recall previous answers to 203

APPENDIX B

questions. However, in the best-fitting models of discrepancies, Schooling was the only variable to appear just once: Schooling did reduce an informant’s tendency to forget avoidances, but was unrelated to their propensity to make mistakes. Is this pattern of results consistent with a significant effect of schooling on reasoning? I believe the key to the unexpected lack of significance of Schooling is that schooling only reduces the proportion of “no avoidance” responses (i.e., reduces “forgetfulness”), reflecting the greater knowledge of food avoidances exhibited by educated individuals (Aunger 1996). However, schooling does not reduce the likelihood that an individual will report one avoidance on one occasion, and another avoidance at the second opportunity (i.e., Schooling does not significantly reduce mistake making). Mistake making should be more likely to reflect variation in informant cognition, since it requires consistency in responses, and perhaps an improved ability to recall knowledge. Since mistake making is unaffected by schooling, schooling does not appear to affect any aspect of cognition that can be trained. Schooling is also unrelated to any measure of informant cognition in the multivariate model. Schooling therefore only appears to increase the number of possible responses an educated individual can make to questions about food avoidances. This is likely to reflect the wider personal experience of educated individuals, who have come under the influence of new sources of knowledge in the context of the formal education process itself (Aunger 1996). This surprising conclusion suggests that, at least in the Ituri, the effect of schooling on cognition is similar to that of other social roles, exposing those in school to specific types of knowledge not accessed by others without that experience. However, schooling does not appear to train the more educated mind to manipulate that knowledge in any observable way.27 Thus schooling-as the arguments above suggest for each of the other social roles-primarily influences informant knowledge, rather than other features of cognition. A large proportion of the probability of a discrepancy is accounted for by variables in the model other than those representing social roles. Nevertheless substantive variation in what informants believe with respect to the questions they were asked remains, and is associated with social roles. I therefore conclude there is meaningfd variability in food avoidances in this population. 204

REFLEXIVITY IS NECESSARY

Interviewelr Informant Interactions: Variation in Strategic Behavior A potentially significant determinant of informant responses is the degree to which the language used in an interview is familiar to both the informant and the interviewer. Another is the degree of difference in cultural background between the interviewer and the informant. Unfortunately, the type of language used in the present set of interviews overlaps significantly with the cultural group of interviewer.The familiarity to the informant of the language used in the interview, as well as the effect of the disparity in ethnic background between the informants and their interviewers, are thus confounded to some degree in the statistical model. Both are therefore considered as sources of miscommunication that cannot be readily distinguished in the present case. As a result, I have resorted to using two variables to indicate differences in cultural background between the informant and the interviewer.The first variable, Major Cultural Difference, largely measures variability associated with my use of Swahili. The second measure, Minor Cultural Difference, is related to the lesser cultural difference between a horticulturalist interviewer and forager informants.28 In general, Major Cultural Difference proved to be considerably more important than Minor Cultural Difference. Since the former measure largely reflects the effects of the anthropologist, the present result once again suggests that the anthropologist-as-interviewer functions differently than the local interviewers. For example, in both models of discrepancy, there was an interaction between Major Cultural Difference and Animal Group (a methodological control factor discussed below): as the number of times I interviewed an informant increased, the greater was the variability in responses associated with specific types of question (i.e., those involving animals from particular ethnofaunal groupings). When I was paired with myself as an interviewer, the number of differences between responses was greatest. This indicates that informants were mentioning a wider variety of avoidances in interviews conducted by me. It appears that informants were less constrained to say the same thing on repeated interviews with me than when they were interviewed by local interviewers. It may be that they felt I was less likely to detect implausible answers, with my lower degree of familiarity with the local culture.

205

APPENDIX B

An interaction effect between the Sex of the informants and some interviewer characteristic appears several times in the models. Although male informants in general are more susceptible to forgetfulness than women, their propensity to make mistakes was reduced, compared to women, when dealing with interviewers from different ethnic backgrounds than themselves. This is true whether the difference in background is that between a forager male and horticulturalist interviewer, or any male informant interviewed by the anthropologist. As the familiarity of the interviewer and/or his language decreased, the strictly male set of interviewers apparently had greater difficulty eliciting avoidances from female informants, who were more prone to make a response of “no avoidance” than male informants. This suggests that women become more restrained when faced with a relatively less familiar situation, and tend not to mention an avoidance that they would otherwise report to an interviewer with whom they share a greater degree of similarity in background. Since women in the Ituri tend to have less competence in secondary languages than men, it is possible that they had greater difficulty in comprehending questions put to them in a less familiar language than did men. When women did not understand a question or, alternatively, were more intimidated by male interviewers (especially when those men come from different ethnic groups than themselves), they had a greater tendency to make the easiest response: “I have no problem in consuming that animal.”29 Like women, younger informants appear to become less voluble when faced with disparate sets of interviewers, again probably due to uncertainty and/or language difficulties. As they become more experienced, however, interviewers become better able to deal with any reluctance or confusion, probably by learning to probe without frightening the younger informant. I also figure prominently in the significant interactions involving purely interviewer-based effects. For example, Interviewer Experience interacts with Inter-Interviewer Dissimilarity and Major Cultural Difference variables, both of which largely reflect variability due to the anthropologist-as-interviewer. These interactions suggest that my performance is different from the other interviewers, even after considerable experience. My performance over time does not become as similar to that of the other interviewers as the other interviewers’ performance becomes similar to each other’s. In some sense, then, my interviews continue to be different from those of the other interviewers, despite my having con206

REFLEXIVITY IS NECESSARY

ducted a great many more interviews than any of the other interviewers by the end of the study. This might be because I began interviewingwith less knowledge of the local culture and language. In sum, the results on interviewer effects suggest that I either had a different interviewing style or was being told substantively different avoidances by informants. Unfortunately, it is difficult to distinguish which of these two explanations is more likely from these results.3o The set of results involving interactions between interviewers and informants suggest that young and female informants were reticent in responding to their interviewers. Combined with the earlier results on Interviewer Experience and Inter-Interviewer Dissimilarity, this pattern indicates that informant responses seem to reflect strategic reasoning about how best to respond to the questions put by the interviewer. The suggestion that informants are circumspect in their responses can be generalized in a way that highlights the kind of detailed and interpretive statements that can be supported by this kind of empirical investigation. For example, a fundamental criticism of interview data is that what an informant says is constrained by the sociopolitical differentials between interviewer and informant cultures (e.g., Bourdieu 1991; Marcus and Fischer 1986; Hymes 1969). An obvious example is when the society being studied by the anthropologist has been colonized by the society to which the anthropologist belongs. Such bias is argued to color indelibly, but invisibly, all responses that informants make (e.g., Clifford 1983). However, since the present study includes comparable interviews, one conducted by an interviewer of the informant’s cultural group and the other by an interviewer from a Western culture, the response bias due to cultural domination can be estimated. There are two levels or types of possible domination that can be investigated in the present case. The first is between Westerners and nonWesterners, or between me and members of the local groups. In this comparison there is not only economic but also ideological domination, since Western influence in northeastern Democratic Republic of the Congo has been not only through colonialism, but also through Christian evangelization. This difference in cultural background is measured by the variable Major Cultural Difference. The second level of domination is through the difference in social status between horticulturalists and foragers. These social relationships also have a variety of components (Grinker 1990, 1994). However, these groups share a more recent ethnic 207

APPENDK B

history and even intermarry. The difference in cultural background between horticulturalists and foragers is measured by the variable Minor Cultural Difference. If the Westerdnon-Western power differential can be presumed to be larger than that between horticulturalists and foragers, then the greater involvement of Major Cultural Difference than Minor Cultural Difference variables in the models for both measures of discrepancy reflects the two degrees of cultural domination between interviewers and informants in the data. Another pair of interaction effects also exhibits the same pattern: women are more forgetful the greater the difference in sociopolitical power between their own cultural group and that of the male interviewer (i.e., the interaction of Major Cultural Difference with Sex is more significant than Minor Cultural Difference with Sex). This suggests an interaction exists between the sociopolitical domination of a cultural group with social domination along the lines of sex that is typical of a patriarchal society (the societies in the West, as well as in the Ituri, can be considered patriarchal, since they are patrilineal and virilocal). Admittedly, there are other effects that correlate with the degree of cultural domination (e.g., the interviewers lack of familiarity with the local language and/or the cultural norms of the group being studied). But cultural domination is probably closely intertwined with these other aspects, each of which contributes to the observed pattern. These kinds of conclusions about ethnographic practice have traditionally been made on a purely speculative basis. It is one of the benefits of reflexive analysis that these kinds of conclusions can emerge from an independent analysis, rather than spring, full-blown, from the head of the ethnographer.

Notes 1. This measure of discrepancy is a function of how many categories of avoidance are admitted into the analysis. Since a central concern of this study is the measurement of meaningful variation in informant beliefs, a measure is required that reflects changes in mental models rather than context-dependent aspects of belief due to slight changes in phrasing of the question or some trend in the interview discourse at the point of questioning. Inclusion of too much detail in the coding scheme (i.e., using a l l three hundredplus reasons for rejecting foods mentioned during interviews) would probably increase the degree of situational randomness in the model, as well as making the interpretation more difficult. A balance between precision and meaningful variability being necessary, I used a coding scheme with thirty-six different types of avoidances that, in my estimation, best reflects the salient aspects of cognitive variation in food avoidance types. In addition, since

208

REFLEXIVITY IS NECESSARY there were three possible avoidances associated with each prompt of an animal name, I compared the most salient (first-mentioned) avoidance during one interview to all three of the codes from the other interview to ensure that no code from one interview matched the most salient code from the other before counting the observation as discrepant. 2. Sudman and Bradburn (1974) distinguish two similar types of memory error in their review of response bias in sociological survey data. The first type is equivalent to what I have called forgetfulness: an informant simply doesn’t recall the fact in question. The other kind they call telescoping, because events that actually occurred some time ago are recalled as having happened more recently. This is a temporal rather than category error as in the case of mistake making. Nevertheless, the two factors distinguished by these researchers seem to be related to those I have named. 3. The SPSS Logistic Regression procedure was used to estimate the multivariate model developed for the present study. I have followed the traditional procedure for multivariate linear modeling with categorical covariates of estimating only hierarchical models (Fox 1984,343). This requires that effects of a given order include the effects involving those variables at all lower orders. I have therefore estimated a first-order or main-effects model (including only the independent effects of each variable), and then proceeded to investigate second-order interactions between variables that were determined during the first step to have a statistically significant effect on the dependent variable. As a result, to determine the best-fitting model for each dependent variable, two estimation steps were used. First, the best-fitting main-effects model was estimated using a backward stepwise procedure based on changes in the likelihood ratio statistic. Simulation results have shown that the recovery of interpretable models is most efficient when the probability criteria for admittance and removal of variables to and from the model are set between .15 and .20 (Hosmer and Lemeshow 1989,108). I therefore set the maximum chi-square probability value for admission of a variable to the model at .20, while the maximum value for a variable to remain in the model was set to .15. In almost all cases, the actual probabilities that variable coefficients were different from zero fell considerably below these critical values. In the second step of the estimation procedure, two-way interaction effects involving the variables remaining in the main-effects model were tested using the same backward stepwise procedure and criteria for variable selection. I excluded from the analysis, however, those interaction terms that made no sense (e.g., Major Cultural Difference by InterInterviewer Dissimilarity, since both variables are ordinal, and already measure interactions of other variables). I also excluded those that introduced redundancies into the design matrix (e.g., the third measure of Cultural Difference). Higher-way interactions were also excluded from the model due to the difficulty of their interpretation and because higher-order effects are highly dependent on characteristics of the sample. There is another concern regarding the measurement and calculation of the dependent variables (the measures of discrepancy). In the majority of observations no avoidance was reported on either occasion. Two different kinds of effects would be estimated depending on whether such observations were admitted to the analysis. There is no way to know, even after asking an informant twice, which animals have no avoidance for that informant since they may have forgotten it on both occasions. Every question thus has a positive probability of being responded to with mention of an avoidance. There is thus no a priori or a posteriori justification for excluding particular questions from the analysis. In order to know what effects result both in reports of an avoidance when none really exist, as well as cases when informants report no avoidance even though they have such an avoidance, all observations must be admitted to the estimation procedure, as was done here. 209

APPENDIX B

4. The variable R can be used to estimate the degree of influence particular effects have on the likelihood that the informant will forget or make a mistake (R is listed in tables B.4 and B.5). Negative values of R indicate that as the value of the effect increases, the likelihood of a forgetful or mistaken response decreases. Small absolute values of R indicate that the effect has little influence on the probability of a discrepancy (in this sense R is analogous to the partial correlation coefficient of multiple linear regression). Since neither model explains a large proportion of the variation in discrepancies, most effects accordingly show low values of the R statistic. This indicates that no single effect has a large influence on the likelihood of making a discrepant response. 5. I do not report the total proportion of variability in the likelihood of the dependent variable accounted for by this procedure because the assumption of statistical independence of all explanatory variables is reasonable only for the determination of relative contributions, not absolute values. 6. The log-likelihood chi-square test value (called -2 log likelihood) shows both forgetfulness and mistake-making models are significantly different from fully saturated models (i.e., models that include all possible combinations of interactions between the variables). This indicates that effects significant in the production of discrepancies remain unidentified by the model. However, chi-square-based significance tests are inflated by large sample sizes, so these values may not be indicative of the models’ true sufficiency. Nonetheless, the proportion of total chi-square associated with the model indicates for both measures of discrepancy that the explanatory variables explain a significant proportion of the variation in the data. In addition, diagnostic measures indicate that these best-fitting models are not problematic statistically. For example, values of deviance (a measure of the difference between the observed and estimated probabilities of a discrepancy) are approximately normally distributed, and observations with extreme values of deviance (i.e., with absolute value > 1) are distributed quite randomly with respect to each covariate. 7. Paired t-test: t = - 1.00; sig o f t = ,325; df = 34. 8. Paired t-test: t = -.18; sig o f t = 359; df = 34. 9. Multiple regression with a dummy variable coding for interviewer: number of avoidances: B = .6497; t = .270; sig o f t = .7877; diversity of avoidances: B = -.1486; t = -.560; sig o f t = .5758. 10. Model results are as follows:

Variable Interviewer Experience Forager Sudanic Age Schooling Sex Sex by Forager Sex by Sudanic Age by Forager Age by Sudanic Age by Sex (Constant)

Tolerance

VIF

355996 .065959 .077414 .123485 S22776 .092694 .199310 A46258 .093258 .086447 .149748

1.168 15.161 12.918 8.098 1.913 10.788 5.017 6.837 10.723 11.568 6.678

T

SigT

2.981

.0030 .4765 .0161 .0295 .0035 .0675 .0279 .7875 3887 .0042 .0166

- .713

2.416 2.184 2.933 -1.833 2.206 -.270 -.140 2.875 -2.405

5.505 210

.oooo

REFLEXIVITY IS NECESSARY 11. Number of avoidances: B = .061401; t = 4.816; sig o f t = .OW@ distribution of avoidances: B = .008255; t = 5.941; sig o f t = .0000. 12. Number of avoidances: B = -.202075; t = -3.979; sig o f t = .0001; distribution of avoidances: B = -.012939; t = -2.355; sig o f t = .0197. 13. Inter-interviewer Dissimilarity is an ordinally scaled index that has the value of 1if the two interviewers have the same identity (as when the anthropologist is paired with himself, i.e., NA, and the pair WK), the value of 2 if the two interviewers have the same cultural background (N/K), and the value of 3 if the two interviewers come from different ethnic groups (all such cases include the anthropologist as one of the pair, e.g., N K ) . However, most of the variability of being in the same cultural group is due to pairs of interviewers with the same identity since only IUN, a small sample of observations, distinguishes the same-group-but-different-identity class. 14. Nevertheless, there are suggestions in the humoral literature discussed above that consideration of question-specific response reliability might reduce the difficulty in interpreting variable results. For example, Foster (1979,182) suggests that foods with different “intensities of qualities” have an effect on the consistency of valence assignments for different foods. Mathews (1983), as noted earlier, found that consistency in the humoral valence of a particular food between informants depended on the degree of consistency to which single informants assigned humoral valences to that food across a range of uses. Mathews argued from this result that when a context was not specified to an informant during questioning, the most salient context of use of the food in question would be inferred by that informant, who would then respond with that specific context in mind. If the valence of a food varied between these specific contexts, and if the relative salience of particular contexts of use of a food was variable between informants, then the variability in reported humoral beliefs should reflect the degree of consistency in valence between the specific contexts of use of that food. The salience of particular contexts for different informants could therefore explain much of the observed variability in humoral valence assignments from general questioning in the population as a whole. This is a form of response bias (although Mathews does not refer to this phenomenon from this methodological perspective). 15. Although Cultural Consensus theorists are alone in recognizing this problem, and admit it is serious (Romney, Weller, and Batchelder 1986,332), they simply assume there is no variation in the difficulty of recalling knowledge using different questions as stimuli. 16. In order to examine any type of question-specific effect, the analytical technique must have the ability to simultaneously compare patterns of response to a set of questions. The difficulty of a question can only be measured in comparison to other questions; or, from another perspective, the consistency of one response can only be measured with reference to the degree exhibited by responses to another question. Although the statistical procedures that perform such operations are relatively computer-intensive and less familiar to most researchers (logistic regression is the primary technique used by these schools), their use is advocated here for their ability to provide information concerning the salience of the questions used to measure cognitive variability within cultural groups, 17. During my field season, I conducted extensive interviews with five Lese horticulturalist informants concerning their classifications of animals. For the purposes of this analysis, seven classes of animal were defined, specifying the major named emic categories of animal. These groups are roughly equivalent to the primates, viverrids (genets and mongooses), bovid and suids (antelope and porcine species), rodents, birds, fish, and “other” animals (e.g., reptiles and insects). 211

APPENDIX B 18. Logistic regression models, such as that used here, appear to be adept at the detection of such effects. Hariharan Swaminathan and H. Jane Rogers (1990), for example, using simulation studies, showed that logistic regression models are also quite efficient at uncovering nonuniform “differential item functioning” through the use of interaction terms. Nonuniform differential item functioning occurs when informants, divided into groups on the basis of some characteristic, exhibit different probabilities of answering a given question correctly, even when their cognitive abilities are judged to be the same. Nonuniform differential item functioning would be exhibited in the present case by a result in which the probability of a discrepancy is a function of an interaction between a variable defining groups of informants (e.g., social roles) interacting with some measure of reasoning or recall ability (e.g., Cognitive Load). However, no such effect proved significant, perhaps because such effects were statistically swamped by interactions between Question-Specific Response Bias effects and interviewer characteristics, which are not considered in psychometric models. Such effects are also not often explored in cross-cultural data involving traditional populations. Nevertheless, this is further evidence that differences in reasoning ability are not significant determinants of discrepancy making in this population. 19. As a result, I cannot use the psychometrician’s approach to measure an informant’s cognitive ability. The psychometrician measures cognitive ability as the number of correct responses to a set of questions. However, there is no set of correct answers to the questions used in the food avoidance interviews. Without knowledge of the correct answers, I cannot count their number to measure cognitive ability. Others have assumed that they can infer the answer key to a set of questions whose correct answers are unknown a priori by determining consensus responses (Romney, Weller, and Batchelder 1986). However, assuming consensus in a cultural domain is inappropriate when variability of belief is likely to reflect individual experience rather than the social imposition of uniform, normative beliefs. Such a case is food avoidances in the Ituri. Applications of the Cultural Consensus approach also tend to assume that different questions can be equally weighted in determining consensus beliefs about a domain (e.g., Boster 1985; Boster and Weller 1990). However, the results from the previous section show this assumption is invalid in the present case. 20. Another implication of this analysis applies to statistical tests of ethnographic data. Any analysis that assumes all data are equally reliable (or, equivalently, uniform probability of error across questions) would be invalid. Most statistical analyses make such an assumption. It is therefore important to control for this effect, as I have done here, when reporting on interview-based data using standard statistical tests. 21. Like Animal Group and Animal Familiarity, Cognitive Load can also be interpreted in a purely methodological fashion. All three of these variables control for potential non-independence between observations of discrepancies. For the measurement of informant-specific factors to be reliable, the assumption of the logistic regression model that the different questions are independent measures of informant performance must be satisfied. Thus Question-Specific Response Bias controls for differences in the probability of a discrepant response between questions, while Cognitive Load controls for differences in the probability of a discrepant response between the different individuals in the sample. Allowance for Question-Specific Response Bias is a means of compensating for differences in question difficulty when measuring respondent cognitive abilities. Similarly, Cognitive Load is included because individuals in the sample come from different ethnic backgrounds and sex and age classes. The food avoidance systems of each such

212

REFLEXIVITY IS NECESSARY group are somewhat different. For example, Lese, and particularly Lese women, have more food avoidances than other groups in the sample (Aunger, unpublished data). Inclusion of Cognitive Load thus “levels the field” between individuals of different social roles by controlling for differences in their cultural systems of belieE when I compared the same best-fitting model for each measure of discrepancy without Cognitive Load included, the significance of an informant’s Ethnic Group, Sex, and Age were increased substantially over the model incorporating this effect (other factors were largely unaffected). However, the inclusion of Cognitive Load did not completely absorb all social role effects from the best-fitting models, suggesting there is meaningful variation between groups defined by their social roles in their propensity to exhibit discrepancies (I have used the total number of avoidances reported by each informant during their interview with me, in order to increase the consistency of the measure of Cognitive Load across informants). In sum, inclusion of Cognitive Load as well as the Question-Specific Response Bias factors seems to effectively reduce bias in the estimation of the other factors in the model, and thereby preserve the model’s assumption that observations of discrepancies are statistically independent. 22. This is also true of the “main effects” models (those estimated before the introduction of interaction terms). 23. These figures for the relative significance of particular factors reflect somewhat arbitrary assignment of one-half the variation due to interactions between factors to each of the factors involved. 24. Budu served as the reference, or statistically redundant, category for ethnic group, and so is not included explicitly in the model or results. 25. The independent Age effect is determined from the “main effects” model of mistake making; in the best-fitting model including interactions (shown in table BS), the directionality of this effect is reversed due to the interaction involving Age. 26. This is not to deny that there is considerable evidence in Western societies showing differences between the sexes in verbal, mathematical, and spatial abilities (see Grusec and Lytton 1988 or Kimura 1999 for a summary). However, applying such results to societies with different socialization regimes, as is likely to be the case in the Ituri, is problematic (cf. the Barry, Child, and Bacon 1959.). 27. Since the average level of education in this population is somewhat less than three years of primary school, the observed lack of memory or mental training related to schooling could also be due to the generally low level of exposure to schooling. 28. The first variable measures the effect of using Swahili and/or the degree of exposure to the anthropologist (a value of 2 for the interviewer pairing NA, and a value of 1 for N K , AIM, and A/N;otherwise the measure is set to 0).The second variable separates out the possible effects of a horticulturalist interviewing foragers in the somewhat different horticulturalist language (once for A/N pairings and twice for KlN). The third variable in this set, measuring the frequency of use of a maternal language, is redundant, and so is not included in the model. 29. Studies on interviewer effects in sociological survey data have also shown that the sex of the interviewer had a significant effect on informant responses to questions sensitive to intersexual encounters. For example, Zehner (1970) found that young American women were more inhibited in their responses to questions concerning premarital sex when interviewed by men.

2 13

APPENDIX B 30. I do not believe the observed importance of personal experience in coordinating intercultural communication is due to some idiosyncrasy in my situation during this study. My level of field experience and language proficiencywere probably typical of most ethnographers: I had a full year’s experience at the field site and fluency in Swahili when I began conducting the interviews in this study, which continued for nine months.

214

APPENDIX C

IDEALISM’S FA1LURE

A

s noted in chapters 2 and 4, where Cultural Consensus Analysis (CCA) was discussed, the central methodological goal of CCA is to provide ethnographers with a way to investigate unknown belief systems without having to collect information from a large sample of individuals. Instead, responses from a small number of informants, weighted by their expertise, become sufficient (Romney, Weller, and Batchelder 1986, 326). However, because the ethnographer-as-innocent cannot be trusted, external sources of information for checking CCA output are suspect. It is therefore imperative that the method reliably induces the correct suite of consensual beliefs when in fact there is consensus in the relevant belief system (some domains of belief are simply underdetermined by social forces, and hence are relatively unstructured). Some measures of the technique’s reliability (in terms of unbiased and efficient estimates of informant competency, as well as consistency of consensual responses with a known consensus) have been examined using methods that generate datasets randomized within certain parameters. CCA has generally passed such tests (Maher 1987; Romney, Batchelder, and Weller 1987, 164). Users of the method, relying on these results, have therefore simply run the CCA procedure on whatever data they have available, assuming that the inferred set of consensus values is relatively robust to differences in data collection procedures. However, the many possible combinations of informants produced by the ad hoc selection regimes allowed by CCA (Johnson 1990), together with varied questionnaire designs, may create considerable nonrandom variation in basic data (e.g., interview responses). As a result, my intention in this appendix is not the usual one of providing only the most appropriate analysis, given all the available data. Rather, I examine the robustness of CCA outputs to measured variation 215

APPENDIX C

in inputs. Such sensitivity analyses, using both simulated and real datasets, are common in other areas of anthropology (especiallyin areas such as systematics, where new phylogenetic methods are commonly subjected to such scrutiny), but have been limited in cultural anthropology thus far. In particular, to my knowledge, CCA has not previously been tested using post hoc, quasi-experimental manipulations on a real ethnographic dataset to determine what variation in consensual beliefs results. In fact, when applied to the Ituri Forest food avoidance data, such sensitivity tests produce significant variation in the set of responses determined to be consensual by CCA (see table C.l for a listing of the primary results described in this appendix). Some manipulations of this data even result in the conclusion that this same belief system is not consensual at all. These analyses thus uncover significant problems with the method indeed.

Cultural Consensus Analysis: A Brief HOW-TO The central idea in our theory is the use of the pattern of agreement or consensus among informants to make inferences about their differential competence in knowledge of the shared information pool constituting culture. We assume that the correspondence between the answers of any two informants is a function of the extent to which each is correlated with the truth.

-A. Kimball Romney, Susan Weller, and William Batchelder, “Culture as a Consensus: A Theory of Culture and Informant Accuracy”

Application of the CCA model requires that three conditions be satisfied. First, informants should share a common cultural background to ensure that the consensual responses are applicable to all of them. Second, each response should be independent of all others (including responses to other questions by the same informant). Statistically, this implies each response is an independent draw from the universe of culturally possible responses, or cognitively, a unique recall from memory. Third, each question should be equally difficult to answer correctly (i.e., the competence of an informant should be constant across questions). Employment of the method first requires that the similarity in knowledge between each possible pairing of informants be determined. 216

random subsample

males only females only only individuals over 29 only individuals over 29; 36-valued coding scheme only individuals over 29; ancestral taboos coded as missing equal random samples from the four major ethnic groups “no taboo” coded as missing “no taboo” coded as missing; 36-valued coding ‘(no taboo” coded as missing “no taboo” coded as missing all Sudanics; only questions on familiar animals binary coding scheme

30

I56 I28 I42 I42

I42

I20

284 284

56 82 284

Sudanic Adults

Multi-Group Sample

Sudanics Sudanics

EfeI Bantu Salient Response Sample Sudanics

284

random subsample

Special Sample Characteristics*

284 56 82 27 I00

N

Cultural Consensus in the lturi

Sudanics Efe Bantu Tswa Random Sudanic Sample Random Sudanic Sample Sudanic Men Sudanic Women Sudanic Adults Sudanic Adults

Samble

Table C. I .

.983 .978 I.002

100.0 ( I 4 4 ) 2 100.0 (I44)2

100.0 (I33)2 100.0 ( I43)2 26.9 (41) 65.5 (145)

.477 (. 157) .368 (. 157)

.705 (.24l) .587 (. 184) ,609 (.II I) ,430 (.203)A

4. I92 (7 I.686) 3.268 (45.330) 6.363 (3 I.096) 6.897 (3 I.005) 6.245 (I08.8) I.693 (64.183)

I.oo I .999

I2.4 ( 145)

.650 (. 183)

10.443 (54.692)

(continued)

I.om

.989

.989

41.4 (145)

.618 (.124)

4.680 (56.498)

(145) (I45)

(I45)

.986 .979 .984 .979

.927

1.001 .986 .989 .988 .974

Pseudo-Reliabilitve

(I45)

32.4 64. I 51.7 37.2

,555 (.I10) .5 I 8 (. 134) .543 (.I12) .495 (.I IS)

(I45) (I45) (145) (I45) (145)

4.852 (49.947) 5.33 I (36.632) 4.3 I 9 (43.723) 4.068 (36.7 19)

43.4 26.2 28.3 9.7 50.3 37.9 (I45)

(.I 13) (.I30) (.147) (.083) (.121)

% Taboos in Consensus Setf

.544 (. 109)

.519 .744 .716 .866 .518

Average Combetencet

5.2 I 5 (9.24 I)

4.552 (80.2 19) 9.505 (3 I.922) 18.403 (43.836) 22.390 (20.448) 4.384 (28.325)

Eigenvalue RatioA

N

56 82 27

binary coding scheme binary coding scheme binary coding scheme

Speciol Sample Characteristics*

Cultural Consensus in the lturi (continued)

4.693 (I7.235) 9.41 I (30.259) 5.797 (7.034)

Eigenvalue RatioA

.490 (.260)A S60 (.235)A .SO0 (.101)

Average Cornpetenc&

31.7 (145) 34.5 (I 45) 13.1 (145)

% Taboos in Consensus Sett

.949 .974 .900

PseudeRebbility'

Samples include both sexes, all age groups and taboo types, and taboo coding utilizes the thirteen-valued coding scheme with respect to all I45 interview questions. unless otherwise noted. A Ratio is between the first and second eigenvalues; figure in parentheses is the first eigenvalue. 5 Figure in parentheses is the standard deviation. t Proportion of interview questions in which taboo was consensus response; figure in parentheses is the number of questions over which the percentage was calculated. € Pseudo-reliability is Cronbach's alpha (see text for details). Sample contains three o r four individuals with negative competence. I MINRES procedure failed t o converge on a third eigenvalue in this case. 2 Some items not taboo to anyone in sample.

*

Efe Bantu Tswa

Sample

Table C. I.

IDEALISM’SFAILURE

A matrix of these similarity values is typically calculated as the proportion of matched responses (or as covariances in the case of binary responses), corrected for an underlying probability of chance agreement.’ In a second step, the internal structure of this agreement matrix is then determined via minimum residual factor analysis (MINRES) for categorical data (Romney, Weller, and Batchelder 1986, 320), or maximum likelihood factor analysis for ordinal data (Romney, Batchelder, and Weller 1987). Three criteria are diagnostic for the model’s goodness of fit: (1)the relative size of the eigenvalues from the factor analysis; (2) the level of mean competence; and (3) the relative absence of negative competence scores. All three are evaluated together as a description of the structure of the dataset. The initial requirement is that the first eigenvalue of the agreement matrix is several times as large as the second (Romney, Weller, and Batchelder 1986,332). As Stephen Borgatti (1992b, 44)notes, two large eigenvalues, or large numbers of negative scores, would be evidence of “systematically different patterns of responses”-that is, two cultural subgroups represented in the sample. A 1O:l eigenvalue ratio is considered strong support for consensus, while a minimum value of 31 has become the de facto standard for concluding that a consensus exists (Romney, Batchelder, and Weller 1987, 174). Second, when only one large eigenvalue is found, the factor loading for each informant on the first eigenvector is assumed to estimate his or her cultural competency. Competence is defined as the correlation of the individual’s knowledge with the estimated aggregate (Romney, Batchelder, and Weller 1987, 166-67). Individual competency thus measures the likelihood that a person’s answer to a question drawn from the domain is culturally correct, while average competence may be considered to estimate the proportion of shared beliefs in the domain (Weller 1987). Mean competency below .30 suggests that questions do not draw on shared cultural knowledge, but rather similar personal predilections for making the same response (Weller and Romney 1987). In a final step, these competency values then serve as weights for the calculation of the correct response set (or “answer key” in psychometric test parlance, from which consensus modeling draws its inspiration). Bayes’s theorem is used to compute the best estimate for a consensus set of responses using both the a priori probability of responses (i.e., chance) weighted by the a posteriori estimates of informant competencies (Romney, Weller, and Batchelder 1986,323). 219

APPENDIX C

In summary, the method factors a matrix that measures the cultural similarity of all possible pairings of sampled individuals in the group. If the factor-analytic output exhibits several characteristics, the domain can properly be considered consensual. Under these conditions, the response to a question most likely to be culturally correct becomes that which gets the highest score when the frequency of each alternative response is weighted by the average competence of those who made that response (individual competence also being estimated by the method). CCA can thus be seen as a means of finding both the answer key and estimates of respondent knowledge, given only a set of responses (Romney, Batchelder, and Weller 1987,164).These inferred beliefs can be reasonably applied to the group as a whole, with observed variation considered as methodological or idiosyncratic in origin (Romney et al. 1996, 4704). Further, Romney, Weller, and Batchelder (1986, 323) argue that the fit of the model is tested by its own performance: “Since the violation of any of the three assumptions [outlined above] affects the factor solution, obtaining an appropriate factor solution is the main requirement in judging the fit of data to the model.” With this ability to test its own assumptions, CCA is indeed a powerful approach to cultural analysis. Just look at what it claims to provide: Cultural consensus analysis gives ethnographers the ability to simultaneously ascertain (1) whether an unknown belief system constitutes a “high concordance code” (Romney, Weller, and Batchelder 1986,316); (2) what the consensual beliefs in a topical domain are; and (3) estimates of individual informants’ competence in this cultural belief system. Further, it performs these inferences based only on standard ethnographic data: responses by a small sample of informants to a suite of questions designed to tap knowledge in some particular topical domain. Since Romney (1994) claims CCA can be used as a method for determining whether variation is significantly patterned, I assume such survey data is appropriate for use with CCA. Subsequently, I argue the problems uncovered here must characterize any approach that seeks to construct an idealized picture of a cultural group.

Some Basic Questions Before we get to the main questions of how CCA can be used to typologize beliefs and values in cultural groups, let us first examine how well CCA can help us with even more basic methodological hnctions, such as 220

IDEALISM’S FAILURE

defining the cultural domain to study and the population of individuals on which to focus that study. How Can a Cultural Domain be Identified?

All knowledge is structured. When deciding to study a particular domain of cultural knowledge, one must first ask, How is knowledge in this domain cognitively organized? And second, given this, How can one best elicit this knowledge from informants? Since the mode of cognitive organization is often unknown, the ethnographer must assume some underlying representational scheme for the information that is elicited, with the choice of elicitation method designed to maximize the reliability and interpretability of the information so obtained. The general ethos of CCA practice is to let informants do as much of the work of bringing a theoretical question to testability as possible, rather than relying on a priori or theory-based choices of the investigator (Alvarado, personal communication). Thus, informants can be responsible for everything from defining the cultural domain (e.g., through free listing and pile sorts) to defining themselves as a cultural group (e.g., by sharing reported ethnicity or roles within a society). Nevertheless, compromises must inevitably be made. For example, CCA characterizes knowledge, regardless of domain, as a set of independent data-points, elicited as answers to questions designed to probe various parts of the domain. This is a minimalist assumption to make the method generally applicable. But, as a result, there is likely to be some degree of mismatch between the choice and form of probes and the actual shape of the cognitive space in each case, with the number of untapped and over-sampled areas depending on the actual form in which knowledge is represented. This means, at minimum, that a certain degree of nonindependence between probes is to be expected in each study. Further, how consensual or shared the cognitive structure of a domain is between group members remains unknown, and must be pursued independently of consensus analysis (e.g., Romney and D’Andrade 1964; Romney et al. 1996; Wexler and Romney 1972). Thus, although the objective must be to minimize the degree of ad hoc inference, it cannot be eliminated altogether. In fact, all social research projects remain something of an art form, with important decisions to be made at several junctures. First, the researcher must ascertain what range of knowledge to include when defining a cultural domain. The adequacy of free listing for this task is 22 1

APPENDIX C

sometimes obvious (e.g., as in defining the ethnofaunal domain of lizards [Atran 1994]), but often is not (e.g., anti-graffiti policy effectiveness [Brewer 1992al or the “general knowledge” of American undergraduates [Romney, Weller, and Batchelder 19861). For example, Johnson and Griffith (1996, 92) were interested in a domain without obvious boundaries (the effects of coastal pollution on seafood safety), and so had to derive their set of CCA questions from open-ended interviews. However, this introduces a considerable degree of investigator-related inference into the framing and selection of these questions. The present study on food taboos involves relatively low inference in this regard. First, the domain itself is both culturally salient and naturally defined by the set of objects to which the beliefs refer (i.e., animals). Thus, developing the proper probes requires little thought: a set of questions covering all the animals in the local fauna, each regarding the edibility of a particular species. Unfortunately, the richness of Ituri forest life would require so many questions as to tire informants excessively, so a principled selection of animals had to be made. This selection was designed to include all animals figuring prominently in everyday experience, as well as those serving most of the cultural and biological functions associated with animals in Ituri society, including all animals that are consumed on a regular basis (i.e., at least once within a year). This resulted in the set of 145 questions described in appendix A (table A.3) concerning different animal foods, including a nearly complete sample of large mammal species, as well as representative smaller animals (e.g., insects, reptiles). Thus, in the present context, the cultural domain is assumed to be “beliefs about the edibility of culturally salient animal foods.” But it might be possible that food taboos are simply not a single cultural domain, thus violating one of the primary assumptions of the CCA: that all questions are of equal difficulty (Romney, Batchelder, and Weller 1987,165). Borgatti (1992b, 44)calls this the “One Domain” assumption because question difficulty can vary between domains informants know more or less well2 One possibility is that only beliefs related to a particular set of animals constitute a cultural domain. To test this, I selected the most cognitively salient animals, because they are the foods about which individuals must most often make consumption decisions, asking themselves: Is this animal edible for me? As a result, they are most likely to share the same, relatively low difficulty of response. Empirically, I selected the forty-one animals 222

IDEALISM’S FAILURE

that accounted for at least 0.3 percent of the animal captures in the area during the field ~ e a s o n . ~ The eigenvalue ratio is significantly higher for these salient animals among Sudanics than for all animals taken together. The proportion of taboos in the consensus set is significantly lower. However, in this case, these differences are probably real and not a function of any methodological problem because salient animals are more edible; it is relatively unfamiliar animals that tend to be tabooed (Aunger, unpublished data). Variance in competence is about the same as for a set of responses to the more complete list of animals. So changing the set of questions considered does not seem to significantly change values associated with consen~us.~ A second possibility is that the food taboo system includes cultural categories that serve different functions and therefore are not equivalent cognitively or culturally. These categories may then be associated with specific subgroups and therefore should not be expected to be widely shared within the entire cultural group. For example, family-based taboos are normatively restricted to particular lineages in the Ituri and probably act to identify them as a related set of individuals.There is thus no reason to expect entire cultural groups to share these taboos; including them in the analysis may bias results against consensus. Nevertheless, if this category of taboo is excluded (by converting all such responses to missing values), there is little effect on the degree or nature of consensus in most cases (see table C.l). The effect is larger among Sudanics, who tend to exhibit more of these taboos than other groups (7.6 percent of consensus values are family-based taboos among Sudanics; the figure is less than one percent in other groups-see table C.2). But nevertheless, the effect on consensus measurement is surprisingly small. One reason is that these taboo systems are sufficiently robust so that when one taboo category is removed from the analysis, a second, less-popular but nevertheless sufficiently widespread taboo becomes the consensus value. Thus the difference in the number of consensus taboos is less than the number of previously designated consensual family-based taboos. Finally, I have thus far assumed that the “OK-to-eat” response can be treated as simply another taboo category. However, this value may be somehow distinct: a default response in the absence of cultural knowledge rather than just a neutral attitudinal categ~ry.~ To test whether the null value category is affecting all results, I removed it from the calculation of consensus by coding all such responses as missing values. Only among Sudanics does 223

71.7 11.0 0.7 3.4 0 0 0.7 2.1 0.7 5.5 4.1 0

73.8 4.8 0.7 0 0 0 1.4 7.6 0.7 7.6 3.4 0

56.6 11.0 7.6 0.7 0 I .4 2. I 8.3 2. I 8.3 2. I 0

Sudanic Men

67.6 10.3 4.8 0 0 2. I 0.7 6.9 0 6.9 0.7 0

Random

30 62. I 10.3 6.2 0 0 I .4 2. I 8.3 I .4 6.9 I .4 0

Random

100 49.7 11.0 13.8 I .4 0 I .4 2. I 8.3 2. I 9.0 I .4 0

Tswa

90.3 0.7 0 2.1 0 0.7 0.7 2.1 0 3.4 0 0

* cell values are the percent of consensus responses of this type (total number of responses = 145)

Bantu

Efe

Sudonics

Distribution of Consensus Responses*

OK-to-eat Attitudinal Family-based Uncertain Ceremonial Non-homeopathic Homeopathic Pregnancy-related Marriage General Sex-specific Other

Table C.Z. Sudonic Adults

48.3 12.4 15.9 0 0 I .4 I .4 8.3 2. I 8.3 2. I 0

Sudanic Women

35.9 12.4 24.8 2. I 0 2. I 2. I 6.9 2. I 9.0 2.8 0

MultiGrouD

87.6 1.4 0.7 I .4 0 0 0 2.8 0 4. I 2. I 0

63.4 9.0 3.4 0 0 2. I 2.8 9.0 2. I 6.9 I .4 0

36-12

Sudanic

0 22.2 29.2 4.2 0 2.8 7.6 18.1 3.5 10.4 2. I 0

Sudanic No OK

IDEALISMS FAILURE

the resulting change (see table C.1) verge on the expected lack of consensus (eigenvalueratio: 4.2:l;3.3:l if the thirty-six-valued coding scheme is used); the ratios for the other groups remain above 6:1.6 So, even when the method is constrained not to find the consensus that I have argued is so powefithat of no taboos at allsignificant degrees of consensus continue to be uncovered by the analysis. Thus, at least four different tests to determine whether food taboos might be improperly considered a single domain have failed to result in significantly different measures of consensus (although consensus sets vary according to the manipulation of the sample undertaken). These results can either be interpreted as indicating the method is very robust with respect to such changes in the conceptualizationof the domain, or that the domain itself is very fuzzy-that is, not materially changed despite such variation. However, whether any of the three assumptions of the model (independence of responses? one culture? one domain?) has been violated cannot be determined strictly from the output. Defining the Cultural Group

Another issue is: Who can be asked questions?What constitutes the relevant cultural group? This is often an implicit choice by the researcher or reflects a tradition of practice. Quite commonly, scholars use self-reported group membership, particularly in ethnic groups, as the foundation for splitting individuals into groups. Certainly, all CCA studies to date have assumed that self-identified ethnic markers are those that partition a population into appropriate cultural groups, even when it is an ethnically diverse population (e.g., Chavez et al. 1995).7However, self-identification along ethnic lines may not distinguish groups with the highest proportion of shared beliefs. Nonethnic clusters of individuals may exhibit higher degrees of consensus, and can thus be considered to share a cultural background. To test whether consensus values are indeed higher within self-identified ethnic groups in the Ituri, I constructed a sample as biased as possible against ethnicity because it is equally balanced in representation between the four major ethnic divisions. It contains thirty individuals of both sexes and all ages from each of these groups.8 One might expect that by aggregating over multiple cultural groups, consensus would disappear. Average competency should also decrease since individuals would be 225

APPENDIX C

compared to the correct response derived from another group’s taboos against that animal. Nevertheless, as table C.1 shows, the multiethnic sample exhibits a higher ratio between eigenvalues than seen in half of the Ituri ethnic groups and high average competence, strongly suggesting consensus on food taboos. The question is whether this is due to inter-cultural similarity or to some other process. It may be that in a sympatric, multiethnic population that shares a grammar of belief, ethnic boundaries are sufficiently porous to the transmission of taboos as to make ethnicity irrelevant. Considerable overlap in consensus values has also been found between identified groups in other cross-cultural consensus studies (e.g., Weller et al. 1993; Chavez et al. 1995). Such results suggest that there may be at least as much variation within as between ethnic groups for many cultural belief systems, a conclusion reached by other studies of cultural variability as well (e.g., Holland 1987; Swartz 1982,1991; Caulkins 1998; Trossett and Caulkins 2001; Aunger in prep.). CCA simply doesn’t pick out the boundaries between groups very well. However, the multiethnic sample, although composed equally of members from the four identified ethnic groups, is not found equidistant from each group, but rather clustered closely with the Tswa (see figure C.l). Thus, this consensus is not an averaging from the equal proportions of individuals in four independent cultural groups, but abstracts from all of the types of taboos distinctive of one or two groups only, reflecting what is characteristic of the Ituri population as a whole (or indeed any population): general, attitudinal, and pregnancy-related taboos (see table C.2). This is a common interpretation of consensus, and so can be considered meaningful. But again, we know there is interesting variation in belief within the composite Ituri population that is obscured here. Which characterization of the Ituri should be preferred? Unfortunately, CCA doesn’t provide an answer to this question. Categorizing Responses

A second kind of methodological issue in ethnography (whether idealistic in intent or not) concerns how to code responses. In some cases, getting to coded responses is relatively straightforward. For example, with regard to food valences (i.e., the “hodcold” system of belie+see Weller 1983), the probes are simple (What is valence of food X?), as are the re226

IDEALISMS FAILURE

SudMen SudRan30 SudAd uIt36 0.5

EFE

-l.O -1.5

I

:

-3

SUDANICS

SudWomen

BANTU I

I

-2

-1

I

I

I

0

1

2

I 3

Dimension 1 Figure C. I. MultidimensionalScaling of lturi Consensus Datasets NOTE Underlined category names indicate sex-specific samples of informants, italicized names indicate samples of differing sizes constructed using randomized informant selection; names in capitals indicate samples specific to single ethnic groups; and names in normal typeface indicate samples using different response coding schemes. See table C. I for more detail about the composition of samples.

sponses (cold or hot). The whole process requires little inference by the re~earcher.~ The CCA ideal is to design questions to which the informant can simply respond yes or no. However, in other domains, it is not so easy to design questions that can be answered in this fashion, and while this avoids inferring a coding scheme, it can introduce its own pitfalls. For example, in Johnson and Griffith‘s (1996) study, questions were posed, sometimes rather artificially, in a form to produce a yes/no response (e.g., “Heated waste water from power and industrial plants tends to kill off bottom-dwelling species in the immediate area’’ [Johnson and Griffith 1996, 105]), so that the covariance estimation method could be used. However, underlying the apparently simple response is a complicated 227

APPENDIX C

conditional calculation-suggesting that informants may interpret the question differently (e.g., by emphasizing different clauses in their consideration of an appropriate response), which makes answers idiosyncratic in ways the researcher cannot control or recover. In the present case, the probes are relatively simple (Can you eat X? If not, why not? Further probes would be used to clarify circumstances under which the food might be refused).” AU questions were posed in the same fashion, changing only the animal referred to. Informants then reported their food taboos with a phrase or a more detailed explanation of their belief. The interviewer recorded their responses as phrases describing the constraints on consumption mentioned by the informant.’* In standard CCA practice, pile sorts from local informants would then be used to determine which responses should properly be classified together from their emic perspective. However, these data are unavailable. Thus, whether there is a cultural consensus that X and Y type reports can be classified together (i.e., are more similar to one another than types X and 2) has not been directly put to indigenous informants.12 As a result, I had to infer some classification scheme for responses. However, the categorization of responses based on informants’ phrased answers was generally quite easy. Since many answers are expressed in stock phrases that different informants repeat word-for-word (e.g., “it stinks”; “my fetus will be born breach”), the whole response can usually be considered a single cognitive unit. Thus, for the purpose of analysis, I converted responses into codes using a categorization scheme that groups certain kinds of response to reduce the dimensions of variation. These categories have proven to vary meaningfully with respect to pattern of transmission (Aunger 2000), the degree to which individuals actually reject reportedly avoided foods in that category, and the social uses to which such avoidances are put (Aunger unpublished data).

Analyses of Coding Schemes With this background now provided, I proceed to a description of the tests performed on the Ituri data to assess the ability of CCA to cope with data variation, beginning with a look at the influence of coding itself on outcomes. 228

IDEALISMS FAILURE

Consensus Changes with Coding Scheme

One virtue of using the Ituri food taboo system as a case study, from a methodological perspective, is that its complexity allows for multiple coding schemes of the same responses to be developed. As a result, the effect of coding schemes themselves can be investigated (see appendix A for discussion of taboo coding schemes). Most of the analyses in the book have been performed using a conservative, low-dimensionality coding scheme having thirteen different values. However, it is just as appropriate in many circumstances to distinguish a greater degree of detail in responses when coding responses. Are differences in the degree and nature of consensus introduced simply by switching from thirteen to thirty-six different response types? The answer is yes. Using the same Sudanic adult sample as in the previous section, the thirty-six-valued scheme has a lower eigenvalue ratio and average competence value. With a wider variety of possible responses, the likelihood of any one response being more common than the no-taboo response decreases. As a result, the number of consensus taboos goes down, and consensus on those non-taboos decreases. Further, the consensus values can be different categories of response for a given question. If I translate the thirty-six-coding scheme responses back into the original thirteen categories, then the difference between the two coding scheme consensuses can be plotted (see figure C.l; differently coded samples are in plain type). Again, the difference between the same sample, with only the coding methodology changed, results in a meaningfully different notion of what the consensus is about-nearly as great as some inter-ethnic differences. Romney, Weller, and Batchelder (1986, 331), acknowledge that such aspects of research design can affect results. For example, through simulation studies they found that “other things being equal, fill-inthe-blank type questions are more reliable than multiple-choice type, which are in turn more reliable than true-false type.” Thus, the method of assessment is important, as is the method of response classification. Such complaints can be made against any method; the point here is that such methodological difficulties argue against the tendency to reifj consensus values: they vary signz5cantly according to researcherspecified parameter^.'^ 229

APPENDIX C Dichotomizing Responses

The Ituri data can also be represented using a binary format. Remember that the answer to the first question concerning each animal food is whether or not there is a restriction on it (yes/no); if yes, then the secondary question is what type of restriction is involved (in effect, a multiple choice question). As a consequence of this division of the interview format into two questions, results from both the covariance and proportional matching estimation methods of the CCA formal process model can be compared. This also alleviates a number of methodological concerns. All of the results presented thus far might be debilitated by an inappropriate combination of method and data: the multiple choice option I have used in calculating the agreement matrix is susceptible to response bias, or the tendency of informants to prefer some response types over others when uncertain about the correct answer (Romney, Weller, and Batchelder 1986, 317). Such effects might be more likely when there are as many alternative response types as here, and hence influence the model’s measures of response likelihood. However, this concern can be addressed by dichotomizing responses (i.e., into only “avoidance” and “OK-to-eat” codes) so that the covariance method, which is less sensitive to response bias (Romney, Weller, and Batchelder 1986,333), can be used. This also alleviates any questions concerning the categorization of responses by the author rather than by local informants, since this binary coding only deals with the informant’s response to the first, direct question: Can you eat X?14 The results from the analysis of binary-coded responses (table C.1) exhibit some surprising characteristic^.^^ Most startling, there is now no consensus among Sudanics (although informants remain quite competent on average); in fact, the eigenvalue ratio is one of the lowest yet published for any CCA study. But the proportion of consensus taboos is the highest reported in table C.l (excluding the “not OK-to-eat” samples). Note that had this been the sole test conducted concerning food taboos among the Sudanics, a conclusion contradicted by all the other evidence presented in this appendix would have been reached with respect to this population. Of course, the 3:l eigenvalue ratio is not a litmus test for the existence of a consensus, but rather a statistical convention (much like the p < .05 criterion of statistical significance for parametric tests), and a careful researcher would be suspicious of results close to either side of the cutoff value. But 230

IDEALISMS FAILURE

the original eigenvalue ratio for Sudanics as a whole is higher than some published studies where consensus was assumed (e.g., Atran 1994; Boster 1991; Brewer, Romney, and Batchelder 1991; Chavez et al. 1995; Garro 1988; Van Raalte et al. 1993;Weller et al. 1993). Certainly it is not a problem of insufficient data: the Sudanic sample is considerably larger than Romney, Weller, and Batchelder (1986) believe necessary to arrive at efficient estimates of consensus values, and probably the largest ever used with this technique. What do these very different results imply about the validity of the earlier proportional matching-based CCA analyses? The dichotomybased results indicate that the average Sudanic should have ninety-five taboos when in fact the average number reported is just seventy-two (see above). This can occur only if, in the preponderance of close cases (i.e., when the frequency of responses approached 50:50), those with higher competence weighed in on the side of avoidance, so that the consensual value became “avoidance”even when actual avoiders were less than 50% of the population. This implies a correlation between competency and the high number of avoidances in this case, just the opposite of what I have argued characterizes previous analyses. Here, conservative traditionalists, probably mostly adult women (rather than men), are granted higher levels of competence. Further, no consensus is found in this ethnic groupagain in agreement with my ethnographic sense that intra-cultural variability in belief is significant. The irony, however, is that with binary coding, the kind of knowledge being described has a different flavor: (implicit) responses concern not the taboo system per se, but rather beliefs about the general edibility of a food. In effect, the definition of the cultural domain has changed. This makes the comparison between algorithms imperfect, since they don’t measure the same kind of knowledge, But can we determine if these differences in outcome are due to response bias? Consensus continues to prevail in the other ethnic groups, although in each case with more consensual taboos than before. However, there is a change in the rank order among ethnic groups with respect to the degree of consensus (as measured by eigenvalue ratios), suggesting that response bias is not constant across the board. In fact, according to eigenvalues, the Tswa are now closer to the Efe than the Bantu, but at the same time continue to have the fewest consensual taboos by a significant margin. Thus, response bias, if present, is affecting these two measures of consensus differently. But all of the groups are using the same vocabulary of 23 1

APPENDIX C

avoidance, and it seems unlikely there would be significant differences in the likelihood that informants from one or another ethnic group would prefer particular responses when uncertain. It seems more likely these results measure consensus in this somewhat different domain of knowledge in these groups. In sum, the comparison of the multiple-choice and dichotomized results suggests that there is no consensus about which animals to avoid eating (although many should be avoided), while there is consensus about which kind of restriction to apply to particular animals (however relatively fewer are “correctly”tabooed)!l6 While thought provoking, this result is not in accord with my sense of how the taboo system works. In fact, it reverses my expectation that there should be no consensus concerning the kind of belief to have about particular animals, but some agreement about which animals are generally inedible. This expectation derives from Mary Douglas’s (1966) idea that animals considered to be anomalous by a cultural group (with respect to their emic classification schemes and/or due to morphological oddities) are more likely to be found disgusting and hence worthy of cultural avoidance-although the exact content of that avoidance will be determined by historical and cultural circumstances. I therefore find it difficult to argue that either the dichotomous- or multiple-choice-based analyses is more true; they are merely different, and both unexpected. The problem with restricting analytic attention to the presence or absence of dietary restriction alone is that one taboo category may not necessarily fall into the same cognitive or cultural domain as another. At least I have argued (Aunger 2000) that two major classes of taboo are quite distinct in social function, distribution patterns, and hence cognitive representation: ancestral and homeopathic taboos. Socially enforced restrictions against the consumption of food items may not, therefore, constitute a single domain, but represent a composite of several kinds of socio-cultural belief systems, all of which happen to involve foods as foci. In conclusion, CCA has not proven to provide us with definitive answers to basic methodological questions about what data to collect, who to collect it from, or how to code the data that is likely to result from any decisions about the cultural domain and population. But perhaps that is too much to expect of any method. After all, many such issues will require information drawn from outside the exercise of any approach to data analysis. However, we will next see that purely analytical questions do not fare much better when addressed using CCA. 232

IDEALISMS FAILURE

Typologizing Groups While these purely methodological issues are interesting, investigating more substantive issues have even greater interest for most researchers. I begin with the first, and most common, goal of an idealistic ethnography: to represent the cultural group through a representative or consensual set of beliefs. How well does CCA fare in performing this function? “Baseline” or Survey Samples Consensus analyses were first performed for each of the four major ethnic divisions in the Ituri (see table C.l).” Since there is a relatively strong consensus (i.e., high eigenvalue ratio, with high average competence and no negative competencies) in each case, we can conclude that the food taboo systems in each ethnic group are cultural. However, the set of consensus responses is quite different in the various ethnic groups, suggesting there is little overlap between these co-resident ethnic groups in the reasons for which a given animal is tabooed. This indicates that taboos might serve to identifl one’s membership in a particular group in this multiethnic population. Roughly 10% to 50% of the major animals among the various ethnic groups exhibit consensus taboos (that is, the response that CCA determines as correct for an animal- or plant-specific question is a type of restriction, not the “OK-to-eat” category). Thus, there is considerable variation in the degree of elaboration of the taboo system between groups. For example, the foraging Tswa show the highest eigenvalue ratio of any sample (22.4:l) and the highest average competence (87%).There are only fourteen consensus taboos (9.7% of responses). The first eigenvector accounts for fully 94% of explained variance. The opposite is true of Sudanics. Generally speaking, as the proportion of animals for which there is a consensus taboo increases, average competence in that group decreases, and eigenvalue ratios declineall signs of decreasing consensus. Thus, as the taboo system becomes more complex, and hence harder for an individual to manage, individual idiosyncrasies in belief increase. Why is cultural competence generally so high in the Ituri? Methodologically, the reason is that the correct answer to the majority of questions is “no taboo” or “OK to eat,” and in populations with relatively less rich taboo systems, most people have fewer taboos, and hence fewer discrepancies with the consensus set of no taboos. This leads to the somewhat 233

APPENDIX C

ironic conclusion that there is a consensus about which animals should not be taboo. This consensus gets stronger as the taboo system itself becomes relatively more impoverished. This has consequences that will become clearer as we proceed. Consensus Changes with Sample Size

The first question to ask is: How would this characterization of group beliefs change if a different set of data from the same population had been collected? For example, what if fewer informants had been questioned? I investigate this possibility by using subsamples taken from the ethnic group with the largest available sample, the Sudanic horticulturalists. Romney, Weller, and Batchelder (1986, 326, table 6) argue that the number of informants required to determine a correct response with considerable confidence under most circumstances (i.e., average competence greater than 5 ) is less than thirty, and believe that “on high concordance culture patterns [i.e., cultural knowledge domains] samples of six to ten informants will work very well as a base to estimate the answer key” (323). This has been a strong selling point for the method, since it seems to justify the traditional use of relatively few informants with respect to a given domain. Since their statement is based on truejfalse questions, minimum sample numbers will normally be higher for questions with more than two alternative responses, as here. To examine whether there were any effects of sample size on CCA results, I produced random subsamples of one hundred or thirty informants from the Sudanic sample using SPSS’s pseudorandom number generator. Results in table C.l show that changing the sample size doesn’t much affect the degree of consensus or estimates of competence. What does change significantly, however, is the set of correct answers to the questions. Are these significant differences? The differences between these subsets of data are as large as those between some ethnic groups. Figure C.l is a spatial representation of the differences between the consensus sets estimated from the various Ituri samples investigated in this appendix, as determined by multidimensional scaling (ethnic group value labels in capital letters; differently sized Sudanic samples are in italics; other samples are discussed in following sections).18 Figure C.l suggests that the two subsamples provide very different pictures of the same belief sys234

IDEALISM’SFAILURE

tem.19 Further, the procedure has the same confidence about both of these consensus response sets: different samples of people from the same ethnic group appear to be answering two different sets of questions with equal degrees of ability. Does the technique produce better answers with better data (i.e., larger, presumably more representative samples)?All we can say is that the answers are different; the method itself provides no heuristic for judging which correct response set to prefer. The beliefs that characterize the Ituri population depend on which set of individuals is used to build the consensus. Thus, at least when there is significant intra-cultural variation in belief, the concept of a minimum but sufficient number of informants to reliably predict consensus values becomes meaningless. Consensus Changes with Sample Composition

I now turn to considerations of sample composition, with sub-samples of data again being taken from the Sudanic horticulturalists. Some lifestyles might lead a particular sex to specialize in a given domain of knowledge. For example,James Boster (1985) found Aguaruna horticulturalist women of Peru were much better than men at naming varieties of manioc, probably due to their greater experience at making such discriminations because of the Aguaruna division of labor. There are also sex differences in knowledge of food taboos in the Ituri (see table C.1). Not only is there a doubling of the number of consensus taboos among Sudanic women when compared to men, but many consensus taboos with respect to particular animals vary: Sudanic men and women have fifty-nine different consensus values, which is 63% of the Sudanic women’s ninety-three total consensus taboos. This is roughly equal to the average proportional difference in consensus values between the Sudanic and Tswa groups, which share neither a language nor subsistence practice (figure C.l; sex-specific sample values are underlined): that is, the two sexes in one ethnic group display differences in belief as great as some differences between ethnic groups in the Ituri. Thus, although there is a consensus in the Sudanic horticulturalist population as a whole, taboo beliefs are actually quite different between the sexes. This difference is masked when the entire population is considered, with the sex balance in the sample becoming important in determining consensus responses (males predominate in the Sudanic sample). Again, CCA itself provides no hints about what sample composition is to be preferred. 235

APPENDIX C

Another possible difficulty with the estimation of consensus may lie with the inclusion of individuals who are not l l l y socialized. Certainly, in the Ituri case, individuals as young as nine years of age appear in the samples, but do not express complete knowledge of their taboos. What happens if immature individuals are excluded from the analysis? Cross-sectional data suggest that knowledge of food taboos increases throughout young adulthood (Aunger 2000). I therefore excluded individuals under age thirty from the present analysis. Nevertheless, this does not appreciably improve the consensus observed among Sudanics (see table C.l)-in fact, the eigenvalue ratio declines compared to the Sudanic group as a whole.2O This is the opposite of expectation, especially if one thinks that children should be similar to each other in their ignorance and different from adults, and hence represent a separate axis of variation. Here, we have the case of a sample being purified of potentially ignorant (i.e., nonconsensual) informants, with the degree of consensus decreasing as a result. This is likely due to the fact that the number of consensus taboos increases when children are excluded from the sample (with a marginal decrease in average competence as a result).

Is the Observed Variation in Consensus Meaningful? Most of the empirical problems with consensus uncovered thus far have revolved around changes in the consensus response set between analyses. Are consensus sets meaningfully labile, or is this variation due to idiosyncrasies in the present dataset? We can first exclude the possibility that the variation is due strictly to methodological idiosyncrasies of the present dataset by returning to figure C.1,which pictorially represents the degree of difference between consensus sets as distances in two dimensions. The first dimension clearly identifies the number of consensus taboos in a sample, with Sudanic women being highest on this scale, and Tswa lowest (see table C.1). The second dimension appears to reflect variation in the specialization by particular samples on different types of taboo (see table C.2). Bantu and Sudanic women emphasize more personal beliefs (general taboos, attitudinal and uncertainty avoidances), while Sudanics and Efe have pregnancy-related, sex-specific, and family-based taboos. Tswa are in between, with both general and pregnancy-related taboos. These tendencies have been reflected in other analyses (e.g., Aunger 1994a), and so are not unexpected. 236

IDEALISMS FAILURE

We can thus conclude that consensus sets vary in the richness of the consensus taboo system and in their relative composition. What then causes changes in these features? To answer this question, we note that Sudanic women and men appear on opposite sides of the Sudanic sample, as do the two random Sudanic samples and the two adult Sudanic samples-all along roughly the same vector. These shifts are due primarily to changes in the distribution of consensus taboos. But some are meaningful differences between subgroups (e.g., between the types of taboos that Sudanic men and women report), while very similar changes are entirely methodological in origin (e.g., between differently coded versions of the same sample). Thus, methodological variation in response alone cannot account for the observed variation in consensus responses. Further, consider that the average Sudanic reports seventy-two taboos (SD = 22.0; min = 11;max = 116).The consensus analysis (column 6 from table C.1) suggests that only sixty-three of these are legitimate, consensual beliefs. The other 13%on average are therefore idiosyncratic. If we assume that the intra-individual error rate from repeated interviews can be used as an indication of purely methodological variation in response, the non-consensual and purely methodological variation rates are roughly equivalent for Sudanics (11.2 percent for the thirteen-value system [Aunger unpublished data], averaged over all groups). But this cannot account for the fact that only 73% of the average number of reported taboos is consensual among Sudanic men (forty-seven of sixty-four taboos), and only 58% of the average Tswa’s twenty-four reported taboos. On the other hand, CCA finds many more taboos to be consensual than the average Sudanic woman actually reports (ninety-three versus eighty-two). I therefore doubt that methodological factors can explain the proportion of taboo reports that CCA indicates are wrong. Are there more substantive reasons for consensus responses being highly labile? For example, is there some peculiarity in the distribution of responses to questions in the Ituri case? Certainly, compared to other studies using CCA, the present one is unusual in the number of alternative responses allowed for each question (most previous studies have used two or three codes). Can this materially affect CCA statistical outcomes? It does allow that there can be a wider variety of response distributions, and in fact, every possible distribution can be found in the Ituri data, from unanimity on a single response, to a roughly normal distribution, to a 237

APPENDIX C

bipolar response set with two categories above 45% of responses, and distributions in between. While increasing the number of response categories from thirteen to thirty-six does decrease consensus values, it does not do so significantly, as we showed earlier. Other studies also suggest that the number of response types cannot be the reason for unstable CCA output. Devon Brewer (1995) uses up to twenty-one alternatives in inter-group comparisons with interpretable consensus values. More importantly, Maher (1987) used simulated tests to show that CCA results were unbiased when up to five response types are allowed (using the same multiplechoice format as here). It therefore seems unlikely that response structure (the number and distribution of response types) can explain variation in consensus sets.21Thus, the Ituri case is not unusual in terms of coding or values. If purely methodological explanations were unwarranted, then we must conclude that the observed variation in consensual representations is meaningfbl, and a problem for interpretation. Quite different notions about the consensus values of a group can be produced from ethnographically realistic variations in the underlying sample of information available about belief in a domain. Which should be considered the true representation is unclear. These difficulties, taken together, do not bode well for the utility of CCA as a method for characterizing groups.

Comparing Individuals to Ideals The second major goal of CCA is to compare individuals to the idealized group-values, rather than to the beliefs of others in the population. Having found a number of problems with its measurement of consensus, the question remains whether CCA is also troubled by difficulties surrounding its implementation of the notion of cultural competence. Two different characterizations of individuals produced by CCA will be examined in turn. As we will see, this does lead to problems related to those associated with the goal of typologizing the group as a whole. Cultural Competence

Cultural competence is a commonly used notion designed to suggest that individuals vary in their expertise in cultural knowledge. Remember that CCA measures individual competence as the proximity of an individual’s 238

IDEALISM’SFAILURE

set of responses to those considered by the method to be culturally correct. But in fact, an individual’s measured competence can change considerably in CCA tests, depending on which group serves as reference. For example, the competence of the first individual in the Sudanic sample varied from a low of .30 in the multiethnic sample to .48 as a member of the Sudanic women sample. This range of eighteen percentage points in competency values is probably not the most extreme case that could be found, since it was simply the first multi-sample individual I discovered. Competence values are also a function of other choices. For example, competence for the individual just discussed ranged from .36 in the Sudanic thirty-six-valued “not OK” sample, to .56 in the Salient Response sample, to .63 in the Sudanic Random 100 sample. Thus these methodological choices can cause greater changes to competence than are due to different reference groups. In fact, the change in competence is 4.8 points on average among Sudanic adults when the coding scheme is changed (SD = 11.2). Other measurements of a single individual’s competence change to such a degree from the ethnic group sample that they approach a standard deviation in size: for example, 9.0 points on average among Sudanics when only salient animals are considered (SD = 11.1);or 12.9 points among Bantu when “no taboo” values are coded as missing (SD = 18.4). Further, in paired comparisons, an individual with the higher competence according to one coding scheme can be the lower one with respect to the other. For example, differences in competence values for identical pairs were calculated from both the thirteen-valued and thirty-six-valued Sudanic adult samples, and then compared. In 7.7%of pairings, the rank order of the pair members reversed between the two samples, while another 4.7% went from no measured difference to some difference in competence. Therefore, in over 10% of cases, some change in relative standing with respect to the measure of expertise was created by a change in the response-coding scheme. This indicates that inter-individual comparisons of competence are unstable and unpredictable. This is because coding and sampling can change the consensus set of taboos. Since competence measures the degree of approach to these differently constructed ideals, an individual’s distance from one ideal can be different from the distance to another, even though their beliefs don’t change. Further, when individuals are then compared to each other, there can be a switch in their degree of relative approach to 239

APPENDIX C

these moving ideals. Competence, as measured by CCA, is not expertise in a belief system per se, then, but rather similarity of reported belief to a constructed ideal. However, the fact that an individual's degree of centrality depends on the distribution of belief in the reference group is not trivial because, as I showed earlier, it is not necessarily easy to identify the appropriate reference (i.e., cultural) group. A level of indeterminacy thus exists, which is unacceptable for an approach that seeks to construct a single truth from underlying variation (since such variation is also reflected in the estimated set of correct answers, as I have shown previously). However, another perspective, yet to be examined, can be taken on the representation of ethnographic reality that might not be subject to the same sorts of criticism. The Ideal Informant

A strategy common to most, if not

all, of the psychological and social sciences for neutralizing the sting of intentionality and idiosyncrasy is the construction of idealized, artificial minds . . .in the course of formulating normative models of functioning.

-E. Thomas Lawson and Robert N. McCauley, Rethinking Religion: Connecting Cognition and Culture; emphasis in original)

The anthropological version of this construct can be found in the form of an ideal informant. This can be both a model personification of all that is good and right in a culture (e.g., Goodenough's [1965] behavioral ideal of someone who can behave unremarkably in all social settings), or more prosaically, as a rationale for selecting preferable informants from among the many potential ones available to the ethnographer (Johnson 1990). Do problems arise from CCA's conception of an embodied ideal, as in the other CCA idealizations? The Ideal Informant is Unrepresentative

One sense of an ideal informant is normative. In the present context, this implies a hypothetical individual who answers all questions with consensus responses. This person is ideal because he or she has perfect competence (i.e., competence = 1).The first problem that arises using such a representation is that this ideal does not exist. The set of cultural 240

IDEALISMS FAILURE

values that CCA advocates to describe a group does not characterize anyone in the group, at least in the Ituri. The characteristics of the ideal informant must therefore be derived analytically (CCA does not produce an explicit characterization of the ideal informant; it ranks informants relative to the consensus response set). Since optimal scaling can provide a representation of the relationships between various informant characteristics in a common space, I have used this technique on the Sudanic sample to determine what the most likely features of the ideal Sudanic informant would be (see figure C.2; category names are shown using the following conventions: age, sex,schooling, and COM-

PETENCE).22 1.0

HIFH >45

0.5

-

Secondat'y

NoSchool

Men

1

3045 MID

N

.-2 E

0

0.0

-

-0.5

-

E

16-29

*

ii

Women Primary

LOW -1.0

I

-2 .o

-1.5

C M I

-1.o

-0.5

0.0

0.5

1.o

Dimension I Figure C.Z. Optimal Scaling of Sudanic Informant Characteristics NOTE:Underlined category names indicate sex of informant;italicized names indicate whether the informant attended formal schooling; names in capitals indicate level of cultural competence; and names consisting of numeric ranges indicate an informant's age category.

241

APPENDIX C

Since in this case optimal scaling results in a roughly linear vector of increasing competence, extrapolating the competence vector somewhat (since the highest observed competence is only .76) predicts the ideal Sudanic informant would be a well-educated male about forty years of age.23 Is this how an ethnographer would ideally seek to characterize the Sudanic group? Possibly-this characterization is consistent with what some other CCA researchers have found (D’Andrade 1989a, 119-21). But in what sense is it representative of the population, most of whom are poorly educated and many of whom are female (with a very different set of beliefs, as earlier analyses showed) and of other ages? The answer to this question depends, in fact, on whether this individual is likely to have other characteristics that are desirable in an informant. The Ideal Informant is Not Very “Ideal”

CCA studies generally imply that competent individuals tend to have higher levels of education. However, a common result in the CCA literature is that those with greater education exhibit reduced knowledge of fraditional belief systems, presumably due to their more modern lifestyles. For example, the only signdcant intra-cultural variation in knowledge of the folk illness category empacho in Weller et al.’s study (1993, 117) was in the Texas sample, where those with higher education levels exhibited lower competency about empacho treatments. Boster (1985) also showed that school-educated Aguaruna women were less adept at manioc identification than unschooled ones. Thus, to develop an expectation about educatiodintelligence,one first has to determine whether the belief system under study is traditional. I would argue that food taboos are a traditional belief system in the Ituri, presently being replaced by beliefs about food edibility associated with missionary-introduced Christianity (Aunger 1996). Contrary to expectation, higher competence is associated with greater education (figure C.2).24However, in the Ituri case, it is difficult to argue that this result implies more competent individuals are more intelligent or even more Westernized by their experience in school. Although relatively educated informants report a wider variety and number of taboos, I argue from a pattern of empirical results that this increased knowledge is not due to higher intelligence among the educated, but rather to their broader acquaintance with the taboos of different ethnic groups, gained through travel (Aunger 1996). Thus, in the Ituri, education seems to be associated with increased expertise in a traditional domain of knowledge, but not as a 242

IDEALISMS FAILURE

result of relative intellectual advantages. Given the CCA literature, at least one and probably both aspects of this result should be seen as unexpected. DAndrade also argues that those with greater experience of a domain will be more competent. In the present case, experience with taboos is primarily a hnction of age, due to the long period of acquisition of taboo knowledge. Figure C.2 shows that, generally speaking, competence does increase with age, so results in this case conform to expectation. Another expectation is that more reliable informants (i.e., those with more consistent responses) will have higher competence. For example, Weller (1984) and Brewer, Romney, and Batchelder (1991) found highcompetence individuals provide responses with greater internal consistency between questions. Boster (1985) used a small sample of repeat interviews to calculate the correlation between an informant’s reliability (i.e., the ability to duplicate a response) with his or her competence, which is significantly positive. In the Ituri case, interviews repeated on a group of informants can be used to calculate a measure of test-retest reliability similar to B ~ s t e r ’ s . ~ ~ Here, however, the proportion of duplicated responses is inversely correlated with that individual’s competence, measured within their own ethnic group.26This strong and unexpected result-together with the earlier anomaliessuggests we hrther scrutinize those who closely approach the Sudanic ideal informant to determine why they do not exhibit many of the expected characteristics. The ten individuals closest to the Sudanic ideal informant were all interviewed by a field assistant who is himself Sudanic and (it seemed to me) very concerned with deviations from normative responses. His written remarks accompanying these interviews strongly suggest that he found these men particularly unrepresentative of Sudanic beliefs. He writes that he thinks one of the two individuals closest to the ideal informant is particularly startling in what he reports eating, and argues this is because the informant was born to an Efe woman and so doesn’t fear the normal things. My assistant suggests the surprising eating habits of the other individual closest to the ideal informant can be explained by the fact that his brother, rather than father, taught him his taboos. He reports that one of the older individuals in this group of near-ideals didn’t have any children and so doesn’t know his taboos (meaning he had no reason to care about correct knowledge since he had no responsibility to transmit them, nor to fear for protection of offspring by eating only the right foods). Another 243

APPENDIX C

old man fears little, reporting having eaten such shocking items as snake, crocodile, rodents, and snails. In quite a few cases, these individuals experienced unusual life histories, either being raised under atypical family circumstances (i.e., not by their nuclear family in their natal clan’svillage-in particular, two were raised by Efe women), or traveling widely (e.g., to earn money as a plantation worker or as a domestic for a Western plantation owner). One individual spent some time as an itinerant Christian preacher-an occupation guaranteed to result in the ridicule of traditional taboos (Aunger 1996). Several are members of a particular village where, for reasons unclear to me, people malign traditional taboos to an unusual extent. In fact, the only kind of taboos commonly observed in this group as a whole are related to the protection of children, and so are likely to be common in the Sudanic population. Thus, the consensus set of taboos determined by CCA to characterize Sudanics are “lowest common denominator” beliefs in several senses: those close to this ideal never learned other taboos common to their group due to exotic upbringings, disparaged them because of outside influences on attitudes, or simply forgot them. Although DAndrade (1987,196; 1989a, 119) includes normality of experience with respect to the cultural domain as another correlation with competence in the literature, it does not hold in the Ituri case. Finally, Brewer (1992a, b) found that high-competence individuals are more centrally located in the social networks formed by informant groups. This is unlikely to be true in the Ituri, given my personal knowledge of the most competent individuals and the abnormal nature of their life histories. A cluster (UPGMA) analysis of the proportion of matching responses between the 284 Sudanic informants, with the ideal informant added as a hypothetical individual (not presented here due to the size of such a graphical representation), also places the cluster including the ideal informant at the point most removed from the root of the branching diagram. This suggests that the most competent individuals are unlikely to have played central roles in the transmission of food-related beliefs between individuals in this population. At least in this sense, they are true outliers. Romney (1994, 270) argues that the associations between competence and these various desirable traits “dramatically increase the validity of the whole theory.” But what happens when the expected correlations do not appear or cannot be interpreted to mean what is desired? Also, recent research has shown that some of these relationships 244

IDEALISMS FAILURE

have not held up with additional research. For example, Boster’s (1991) survey of CCA ethnofaunal classification studies shows that expertise and competence are not consistently correlated. Further, major problems have appeared with both consensus and the qualities embodied in the ideal informant concept: most of the results have been inconsistent with expectations from the CCA literature, and it has not been possible to make the standard inferences from those results that were expected. On closer inspection, the individuals being held up as ideal are highly unusual in a variety of respects: they are significantly less reliable, atypical, and although educated, not more intelligent. Thus, CCA researchers cannot necessarily rely on a high correlation between competence and other desirable qualities among informants, nor weight the responses of high-competence individuals assuming this will produce the most representative suite of beliefs. In sum, this investigation of the second means of idealizing cultural representation provided by CCA seems not to have fared any better than the first. Thus, our analyses of the sensitivity of this idealistic methodology for representing cultural beliefs have turned up significant problems, for which an immediate source of remedy is not evident. However, it might be argued that the present study is an inappropriate test of CCA because it engages in unwarranted manipulations of data (Romney 1999). Such a charge would vitiate my conclusion that the method itself is at fault. If I have unfairly tested the method, then my conclusions do not follow. At least I cannot attempt to generalize them from their application to the present dataset. Methodologically, in the Ituri study, I did most work with respect to the coding of responses, which may result in some response bias. However, the need to form a coherent domain is much more crucial to consensus analysis than question response bias, since CCA is robust to violations of the equal item difficulty assumption (Romney, Weller, and Batchelder 1986, 332). Thus, the kinds of inferences I have made presumably cause smaller faults than might arise from imperfectly probing informant knowledge due to an ad hoc set of questions. In any case, the overall degree of inference and methodological control in the present study falls within the range of existing CCA studies. O n this basis, I argue that the results obtained here can be interpreted as a general warning to those who would conduct CCA on other datasets with the express purpose of finding the one true consensus of belief or value for a cultural group. No 245

APPENDIX C

such set of beliefs can be found reliably. Further, no such set need exist except as a statistical construct. In the end, I don’t believe ethnographers really want to characterize a group of people with such a construct when it abstracts from what anyone in the group actually entertains as their personal value. Surely, the lesson of the recent debates about ethnographic authority have taught us that second-party representations by social scientists are not to be preferred over native representations if they require us to typologize groups or abstract from the facts on the ground. Let us heed this lesson by remaining realists who honor variation in cultural beliefs and values as one of our own professional standards.

Notes 1. The formal process model described here (Batchelder and Romney 1988, 1989) holds for dichotomous and multiple choice formats, while a more ad hoc data level model (Romney, Batchelder, and Weller 1987) has been devised for formats where the assumption that questions are independent is violated (e.g., triads, ranking, and item matching tests, where multiple questions require sampling from the same set of responses without replacement). 2. Appendix B reports an analysis that the questions about food taboos are probably not of equal difficulty, since an informant’s ability to repeat an earlier response in duplicated interviews varies between questions. However, CCA’s authors themselves note that this assumption is not expected to be widely upheld, and Kathleen Maher (1987) and Batchelder and Romney (1988) show, using Monte Car10 simulations, that MINRES based on the proportion of matched responses (the technique used here) is robust against fairly sizeable deviations from this condition (Batchelder and Romney 1989,239). 3. The meat consumption data shows that meat diet is heavily skewed toward a relatively few, often-captured animals. 4. This result can also be considered an indirect test of the equal difficulty assumption of CCA analysis, since this subset of animals should be of approximately equal salience to informants, and hence any avoidances of these animals equally difficult to recall. Given my argument that the difference in eigenvalue ratios is due to real changes in edibility rather than a real difference in consensus, I believe this result suggests that the entire set of questions do not signijicantly violate the equal difficulty assumption, but see further discussion of this issue in the section on applicability of the Ituri data to CCA analysis in the text that follows. 5. I assume that, in the absence of a cultural restriction, nearly all tabooed foods would be eaten because most such items would be freely eaten by others in the local population (thanks to the high degree of intra-cultural variation in taboos). These foods are familiar to informants (and so unlikely to cause fears about poisoning or distastefulness),and may be refused only during certain times of the life cycle (e.g., during pregnancy) but accepted by the same individual at other times. This is not an assumption that adults continue to 246

IDEALISM’S FAILURE exhibit a child’s “eat-anything”strategy. Thanks to Nancy Alvarado for requesting clarification on this point. 6. Some questions had to be eliminated from this analysis because all respondents agreed a particular food was OK to eat: converting all these “no taboo” values to missing would imply a “missing”consensus value, which is illegal and causes ANTHROPAC to terminate. Similar problems caused the Tswa case to be excluded altogether. 7. I exclude here CCA studies that compare groups defined by status or occupation within an ethnicity (e.g., McMullin, Chavez, and Hubbell 1996). 8. Since only twenty-seven Tswa were available, the addition of three Efe balanced their under-representation. Otherwise, thirty individuals were randomly selected from each ethnic group. 9. But even here, most researchers assume that all foods have a valence, even though for some respondents a given food may be unvalued. Including a third response category (“neutral”)might show that some foods are neutral for the cultural group as a whole, with the binary characterization of consensus proving to be an incorrect representation. 10. Since there are more animals than types of avoidance, it seems best to prompt for knowledge using animal names as probes rather than avoidance types, for two reasons. First, this places less stress on the informant in terms of the expected number of items to be recalled; the result should be fewer errors of omission. Second, this reduces the number of possible response types for any given question, reducing the likelihood of response bias. 11. By framing questions in this fashion, my interview format was open-ended with respect to response types, avoiding another potential problem. In standard CCA practice, a small number of informants are used prior to consensus interviewing to define the range of response types, with the structure of the domain then determined by consensus interviewing. However, this may not elicit the relevant universe of possible responses since the investigator may not have found informants with the widest range of knowledge. Because I did not prejudice the set of possible responses, informants could always introduce a new category of avoidance (even after four hundred prior interviews), which would be missing from a study that followed the more traditional CCA route. 12. The idea that everyone in a cultural group shares a particular categorization of knowledge in a domain is another idealistic notion. Having conducted interviews with various Sudanic informants to determine their ethnofaunal classification schemes, I found considerable variation in these taxonomies as well (Aunger, unpublished data). But CCA applies the same classification scheme-presumably derived from pile sorts by one or more informants-to everyone in the group, whether or not each individual has been ascertained to share this categorization.Of course, I have had to make the same assumption in my use of CCA. Presumably, at least one individual in the local population shares my perception of how the taboo domain should be classified, and so could legitimize my use of it here. 13. Even more dramatic differenceswere produced by Boster (1986,431), who had two assessment tasks for manioc identification among the Aguaruna: an “easy” task and a “hard” task, which included more detailed discriminations.There is consensus on manioc identification, but only if the easy task is used, since the results from the hard task don’t fulfill the 3:l rule of thumb. This order of magnitude difference in eigenvalue ratios between one of the highest and lowest values for competence recorded in the literature is simply due to two different methods for assessingthe same knowledge domain in the same set of informants. Boster (1991) also reports a twofold difference in eigenvalue ratios for passerine bird similaritiesdue simply to two different methods of pile sort techniques (free versus successive); average competence changed 50% between tasks.

247

APPENDIX C

14. I have used the default probability of (either) response = .5, since I have no a priori theory for the appropriate likelihood of responses (as is typical of anthropological research). The comparability with other results in this book is also maximized by this choice because the other tests implicitly make the same assumption of no response bias. 15. In comparison, Alvarado (1994,67) dichotomized responses to pictorial stimuli from the Thematic Apperception Test, originally presented as a multiple-choice option along a five-point Likert scale (an emotional intensity rating scheme that combined positive/negative affect with intensity measurement), even though each of the responses was considered to be a separate answer, and individuals are known to use values on this scale differently. Average competence values in a sample of thirty-two undergraduates was .392 when calculated as multiple choice, and .499 using the binary coding, a difference within the range of those seen in the present study for the same kind of manipulation. 16. This statement is strictly true only for Sudanics, but the sense is the same for the other ethnic groups, with a smaller difference in the degree of consensus between conditions. 17. All consensus analyses were performed with ANTHROPAC 4.0 (Borgatti 1992a), using the multiple-choice option for calculating inter-informant agreement values, unless otherwise noted. 18. The SPSS ALSCAL procedure was used for this analysis, based on Euclidean differences between consensus value sets (coded as numbers). A two-dimensional solution yielded an S-Stress value (Young’s measure) of .108, with squared correlation = .949. A three-dimensional solution produced a somewhat better fit (S-Stress = .075), but the first two dimensions were very similar to those reproduced here. 19. Since I am only interested in relative differences between sample consensus sets, I represent these differences spatially for ease of interpretation. I assume that the appropriate standard of comparison is inter-ethnic differences, which are real and significant because they are produced by unique cultural histories involving different language groups and subsistence practices, with low rates of inter-marriage. Differences in standard measures of agreement such as Cramer’s V (a standardized Chi-square-based measure correcting for sample size and degrees of freedom) or Cohen’s kappa (the normalized proportion of agreement, corrected for chance) show the same relative proportions between samples as indicated by the spatial distances in figure C.l. For example, the measures for Cramer’s V and Cohen’s kappa, respectively, for several comparisons are as follows: SudanicdEfe: .65329, S2166; Sudanic Women/Sudanic Men: .62599, .46469; Sudanic Women/Sudanic Women NOK .70902, S6138; Sudanics/Random 100: .85795, .79882. SudanicdEfe have 28.3% different consensus responses-certainly significant on any measure. These results confirm that sampling, sex-based, and coding differences (discussed below) produce significance values of the same relative magnitude seen in figure C.l, some ofwhich approach the differences seen in inter-ethnic comparisons. Are these in fact statistically significant differences between consensus sets? Pseudoreliability (see table C.1; a variant of Cronbach‘s alpha or the Spearman-Brown reliability formula) is a measure of the validity of the cultural consensus set, expressed as a function of the number of informants and the correlation among their responses. The value of this measure increases with sample size as long as correlations among informants are positive and can be considered to measure the adequacy of the informant pool for finding the

248

IDEALISM’S FAILURE culturally valid response set (Weller 1987, 190). For example, the full Sudanic sample is sufficiently large as to never produce a pseudo-reliability value less than 1.00. The lowest pseudo-reliability value in table C.l is .900, for Tswa (N=27) in the binary analysis (see below); however, most values are greater than .975, suggesting the differences between consensus sets (displayed graphically in figure C.1) are real, and not just noisy variants of each other. It might be argued that I cannot suggest consensus sets are significantly different without providing measures of the confidence attached to the responses deemed consensual by the CCA procedure in the different analyses. But let us consider the possibilities. If low confidence in the consensus response set were coupled with the observed high measures of average competence and eigenvalue ratios, then, rather paradoxically, the situation would be one of high certainty about the existence of some consensus, but with no idea about which consensus is referred to. Alternatively,if there were high confidence in the consensus taboos, this would imply that the method was quite certain in each case that the correct suite of responses for the group had been identified, which still leaves the ethnographer in the unenviable position of determining which analysis to prefer. Either way, my general point is further established: the CCA analyses consistentlyinsist that there is a consensus where I expect none, even though there is no “consensus”about which consensus is meant. I therefore have not included probability measures for consensus responses in table C.l because this would only add another level of complexity to an already complex argument. 20. Throughout this section on consensus, I emphasize eigenvalue ratio as the most salient indicator of degree of consensus. Competence values are discussed in the section devoted to “cultural competence”in the text that follows. In any case, average competence remains sufficiently high in all analyses, with no negative individual competencies, to justify my disregard of these other criteria of consensus here. 21. Some might continue to worry that a “neutral” coding value was included in the response set. However, of the ten individuals with the highest values of competence in the “not O K Sudanic sample, seven were in the top ten for competence in the standard Sudanic sample. Further, those new to the list were closely related to those on these others lists. In addition, removing the “OK-to-eat” option simply increases the number of consensus taboos of those types that Sudanics already specialize in: family-based taboos and pregnancy-related taboos (as well as attitudinal beliefs-see table C.2). It is thus apparent that removing the “OK-to-eat” category does not materially change either the gross CCA measures or the pattern of variation found to underlie these composites. 22. OVERALS was again used in this case. All 284 Sudanic individuals were included in the analysis. AU variables except Sex were treated as multivariate nominal. Low and high competence categories each contain one-quarter of Sudanic informants. Overall fit in three dimensions was 1.05, with eigenvalues of .424, .324 and .302. 23. The specific age prediction derives from the UPGMA cluster analysis described in the section below, “The ideal informant is not very ‘ideal”’;the ten individuals clustering most closely to the ideal informant were educated males with an average age of 40. 24. This trend is complicated somewhat by the fact that some middle-aged men have secondary-level schooling, but most above age forty-five have none, due to the unavailability of schooling when these individuals were young (i.e., before 1945). A further complication is that some younger people in the sample have not yet completed schooling.

249

APPENDIX C 25. See appendix B for details about repeat sampling protocols. Here, I have simply counted any difference in the category of the two interview responses to a question as a discrepancy. Similar results (both overall and within ethnic groups) are obtained when the “mistake”coding of discrepancies is used (see appendix B), although “forgetll” discrepancies show no relationship to cultural competence at any level. The thirty-six-valued coding scheme is used in this analysis, as it more closely corresponds to meaningful emic categories, although similar results were again obtained with the thirteen-valued scheme. Since reliability is available for fewer than sixty Sudanic informants, it could not be included in the OVERALS analysis above. 26. r = -.736; t = -8.490; p < .0001;N = 63.

250

REFERENCES

Abbott, A.

1988 “Transcendinggeneral linear reality.” Sociological Theory 6: 169-86. Abbott, A.

1992 “From causes to events: Notes on narrative positivism.” SociologicalMethods and Research 20:428-55. Abraham, C. 1998 “The Ika redeemed.” New Scientist, Scienzia Nuova Issue 2, May 1998. (www.scienzanuova.itnumeri/n02/n02~083.htm). Abu-Lughod, L. 1992 Writing women? worla!~:Bedouin stories. Los Angeles: University of California Press. Acker, J., K. Barry,and J. Esseveld 1991 “Objectivity and truth: Problems of doing feminist research.” In Beyond methodology: Feminist scholarship as lived research, pp, 133-53, ed. M. M. Fonow and J. A. Cook Bloomington: Indiana University Press. Agar, M.

1980 “Getting better quality stuff: Methodological competition in an interdisciplinary niche.” Urban L@ 9:34-50. Agresti, A.

1990 Categorical data analysis. New York John Wiey. Albert, S. M. 1992 ”The autonomy of lay and professional knowledge in home health-care.”Journal ofAging Studies 6 (3):227-41. Albert, S. M., and M. Moss 1990 “Consensus and the domain of personal relations among older adults.”Journal $Social and Personal Relationships 7:353-79. Altheide, D. L., and J. M. Johnson 1994 “Criteria for assessinginterpretive validity in qualitative research.”In Handbook of Qualitative Research, pp. 485-99, ed. N. K. Denzin and Y. S. Lincoln. Newbury Park, C A Sage.

251

REFERENCES Alvarado, N. 1994 “Empirical validity of the Thematic Apperception Test.”Journal of Personality Assessment 63(1): 59-79. Alvarado, N. 1996 “Congruence of meaning between facial expressions of emotion and selected emotion terms.” Motivation and Emotion 20: 33-61. Alvarado, N. 1997 “Arousaland valence in the direct scaling of emotional response to film clips.” Motivation and Emotion 21: 323-48. Anderson, D. A., and M. Aitkin 1985 “Variance component models with binary response: Interviewer variability.” Journal of the Royal Statistical Society B 47: 203-10. Archer, M. S. 1988 Culture and agency: The place of culture in social theory. Cambridge: Cambridge University Press. Argyris, C., R. Putnam, and M. C. Smith 1985 Action science: Concepts, methods, andskillsfor research and intervention. San Francisco: Jossey-Bass. Ashmore, W. 1989 The rejexive thesis: Writing the sociology cfscient9c knowledge. Chicago: University of Chicago Press. Atran, S. 1990 Cognitivefoundations of natural history: Toward an anthropology of science. New York: Cambridge University Press. Atran, S. 1994 “Coredomainsversus scientific theories: Evidence from systematics and Itza-Maya follcbiology.”In Mapping the mind Domain spec$ciq in cognition and culture, pp. 31640, ed. L. A. Hirschfeld and S. A. Gelman. New York Cambridge University Press. Aunger, R. 1992 “The nutritional consequences of rejecting food in the Ituri Forest of Zaire,” Human Ecoloa 2 0 263-91. Aunger, R. 1994a “Sources of variation in ethnographic interview data: Food avoidances in the Ituri Forest, Zaire.” Ethnology 33: 65-99. Aunger, R. 1994b “Are food avoidances maladaptive in the Ituri Forest of Zaire?”JournalofAnthropologicalResearch 50: 277-310. Aunger, R. 1995 “On ethnography: Story-telling or science?” CurrentAnthropology 36: 97-130. Aunger, R. 1996 “Acculturation and the persistence of indigenous food avoidances in northeastern Zaire.” Human Organization 55: 206-18. 252

REFERENCES Aunger, R. 1999 “Against idealisdcontra consensus [Theory in anthropology forum on “Culture as consensus’’]’’ Current Anthropology 40: S93-Sl01. Aunger, R. 2000 “The life history of culture learning in a face-to-face society.” Ethos 28(2): 1-38. Austin, J. L. 1962 How to Do Things with Words. Oxford Clarendon Press. Axinn, W. G. 1991 “The influence of interviewer sex on responses to sensitive questions in Nepal.” Social Science Research 20: 303-18. Axinn, W. G., T. E. Fricke, and A. Thornton 1991 “The microdemographic community-study approach.” SociologicalMethods and Research 20(2): 187-217. Baddeley, A. D. 1986 Working memory. New York Oxford University Press. Bailey, R. C. 1991 The behavioral ecoZogy of Efepygmy men in the Ituri Forest, Zaire, Anthropological Papers, 86. Ann Arbor, MI: Museum of Anthropology. Bailey, R. C., and I. DeVore 1989 “Research on the Efe and Lese populations the Ituri Forest, Zaire.” American Journal of PhysicaZAnthropology 78(4): 459-71. Bailey, R. C., M. R. Jenike, P. T. Ellison, G. R. Bentley, A. M. Harrigan, and N. R. Peacock 1993 “The ecology of birth seasonality among agriculturalists in central Africa.” Journal of Biosocial Science 2 4 393-412. Bakker, J. W. 1992 “How useful is anthropological self-reflection?”In Cognitive relativism and social science, pp. 35-50, ed. D. Raven, L. van Vucht Tijssen, and J. de Wolf. New Brunswick, NJ: Transaction Publishers. Barry, H. A., I. L. Child, and M. K. Bacon 1959 “Relation of child training to subsistence economy,”AmericanAnthropologist 61: 51-63. Barth, F., ed. 1969 Ethnic groups and boundaries: The social organisation of cultural dtferenre. Oslo: Universiteforlaget. Barth, F. 1987 Cosmologies in the making: A generative approach t o cultural variation in inner New Guinea. Cambridge: Cambridge University Press. Barth, E 1993 Balinese worlds. Chicago: University of Chicago Press. 253

REFERENCES Batchelder, W. H., and A. K. Romney. 1988 “Test theory without an answer key.” Psychometrika 53: 71-92. Batchelder, W. H., and A. K. Romney 1989 “New results in test theory without an answer key.” In Mathematicalpsychology inprogress,pp. 229-48, ed. E. E. Roskam. Berlin: Springer Verlag. Beck, B. E. E 1969 “Colour and heat in South Indian ritual.” Man (NS)4 553-72. Behar, R. 1993 Translated woman: Crossing the border with Esperanzak story, Boston: Beacon. Behar, R. 1997 The vulnerable observer: Anthropology that breaks your heart. Boston: Beacon. Behrens, C. A. 1990 “Qualitative and quantitative approaches to the analysis of anthropological data: A new synthesis.”Journal of Quantitative Anthropology 2: 305-28. Bernard, H. R. 1988 Research methods in cultural anthropology. Beverly Hills, C A Sage. Bernard, H. R., and P. D. Killworth 1977 “Informant accuracy in social network data, 11.”Human Communication Research 4 3-18. Bernard, H. R., P. D. Killworth, D. konenfeld, and L. Sailer 1985 “The problem of informant accuracy: The validity of retrospective data.” Annual Review OfAnthropology 13: 495-517. Bernard, H. R., P. D. Killworth, and L. Sailer 1980 “Informant accuracy in social network data, IV: A comparison of clique-level structure in behavioral and cognitive data.” Social Networks 2: 191-218. Bernard, H. R., P. D. Killworth, and L. Sailer 1982 “Informant accuracy in social network data, V: An experimental attempt to predict actual communication from recall data.” Social Science Research 11: 30-66. Bernard, H. R., and J. S. Pedraza 1989 Native ethnography: A Mexican Indian describes his culture. Newbury Park, CA: Sage. Berry, J. W., Y. H. Poortinga, M. H. Segall, and P. R. Dasen 1992 Cross-culturalpsychology: research and applications. New York Cambridge University Press. Bhaskar, R. 1979 Thepossibilityof natura2ism:Aphilosophicalrriti~ue of the contemporary human sciences. Brighton, England: Harvester. Bidney, D. 1967 Theoreticalanthropology.New York: Schocken Books. Biemer, P., R. M. Groves, L. E. Lyberg, N. A. Mathiowetz, and S. Sudman, eds. 1991 Measurement errors in surveys. New York: Wiley.

254

REFERENCES Blanc, Ann K., and Trevor N. Croft 1992 “The effect of the sex of the interviewer on responses in fertility surveys: The case of Ghana.” Poster presented at Population Association of America meeting. Bohman, J. 1991 New philosophy of social science. Cambridge: MIT Press. Borgatti, S. P. 1992a ANTHROPAC 4.0. Columbia, SC: Analytic Technologies. Borgatti, S. P. 1992b ANTHROPAC 4.0 methods guide. Columbia, SC: Analytic Technologies. Borgatti, S. P. 1994 “Cultural domain theory,”Journal of Quantitative Anthropology 4 261-78. Borofsky, R., ed. 1994 Assessing cultural anthropology. New York McGraw-Hill. Boster, J. S. 1985 “‘Requiem for the omniscient informant’: There’s life in the old girl yet.” In Directions in CognitiveAnthropology,pp. 177-98, ed. J, W. D. Dougherty. Chicago: University of Illinois Press. Boster, J. S. 1986 “Exchange of varieties and information between Aguaruna manioc cultivators.” American Anthropologist 88: 428-36. Boster, J. S. 1987 “Intracultural variation [Introduction to Special Issue].” American Behavioral Scientist 31(2): 150-62. Boster, J. S. 1991 “The information economy model applied to biological similarity judgement.” In Socially Shared Cognition, pp. 203-25, ed. J. Levine, L. Resnick, and S. Behrend. Washington, D.C.: American Psychological Association. Boster, J. S., and R. G. D’Andrade 1989 “Natural and human sources of cross-cultural agreement in ornithological classification.”American Anthropologist 91: 132-42. Boster, J. S., and J. C. Johnson 1989 “Form or function: A comparison of expert and novice judgements of similarity among fish.” American Anthropologist 91: 866-89. Boster, J. S., and S. C. Weller 1990 “Cognitive and contextual variation in hot-cold classification.” American Anthropologist 92: 171-78. Bourdieu, P. 1977 Outline ofa theory ofpractice. London: Cambridge University Press. Bourdieu, P. 1984 Distinction: A social critique ofthejudgement oftaste. Cambridge: Harvard University Press. 255

REFERENCES Bourdieu, P. 1991 Language and symbolicpower.Cambridge: Harvard University Press. Bower, G. H., and J. P.Clapper 1994 “Experimental methods in cognitive science.” In Foundations of Cognitive Science, pp. 245-300, ed. M. I. Posner. Cambridge: MIT Press. Boyd, R., and P. J. Richerson 1985 Culture and the evolutionaryprocess.Chicago: University of Chicago Press. Boyer, P. 1994 The naturalness ofreligious ideas:A cognitive theory of religion. Berkeley: University of California Press. Bradburn, N. M., S. Sudman, and Associates 1979 Improving interview method and questionnairedesign: Response efects to threatening questions in survey research. San Francisco:Jossey-Bass. Brady, I. 2000 “Anthropological poetics.” In Handbook of Qualitative Research (2nd ed.), pp. 949-79, ed. N. K. Denzin, and Y. S. Lincoln. Thousand Oaks, C A Sage. Brenner, M. 1982 “Responseeffects of ‘role-restricted’characteristics of interviewers.”In Response Behaviour in the Survey-Interview, pp. 131-65, ed. W. Dijkstra, and J. van der Zouwen. Oxford: Academic Press. Brewer, D. D. 1992a “Hip hop graffiti writers’ evaluations of strategies to control illegal graffiti.” Human Organization 51(2): 188-96. Brewer, D. D. 1992b “A note on the relationship between centrality and cultural knowledge in a professional network.” Connections 15: 21-28. Brewer, D. D. 1995 “Cognitive indicators of knowledge in semantic domains.”Journal of Quantitative AnthropoIogy 5: 107-28. Brewer, D. D., A. K. Romney, and W. H. Batchelder 1991 “Consistency and consensus: A replication.”Journal of Quantitative Anthropology 3: 195-205. Brigs, C. L. 1986 Learning how t o ask. New York Cambridge University Press. Brown, M. F. 1976 “Patterns of variability in two folk systems of classification.” University Michigan Discussions in Anthropology 2: 76-90.

of

Bryk, A. S., and S. W. Raudenbush. 1992 Hierarchical linear models. Newbury Park, CA: Sage. Burawoy, M., A. Burton, A. A. Ferguson, K. J. Fox, J. Gamson, N. Gartrell, L. Hurst, C. Kurzman, L. Salzinger,J. Schiffman, and S. Ui, eds. 1992 Ethnography unbound: Power and resistance in the modern metropolis. Berkeley: University of California Press. 256

REFERENCES Burnstein, E., and A. Vinokur 1973 “Testing two classes of theories about group-induced shifts in individual choice.”Journal of Experimental Social Psychology 9: 123-37. Campbell, D. T., and D. W. Fiske 1959 “Convergent and discriminant validation by the multitrait-multimethod matrix.” Psychological Bulletin 56: 81-105. Carspecken, P.F., and M. Apple 1992 ”Critical research: Theory, methodology, and practice.” In The Handbook of Qualitative Research in Education, pp. 507-54, ed. M. D. LeCompte, W. L. Millroy, and J. Preissle. New York: Academic Press. Caton, H., ed. 1990 The Samoa reader:Anthropologists take stock. Lanham, MD: University Press of America. Caulkins, D. D. 1998 “Consensusanalysis: Do Scottish business advisers agree on models of success?” In Using Methods in the Field A Practical Introduction and Case Book, pp.175-95, ed. Victor DeMunck and Elisa Sobo. Walnut Creek, C A AltaMira Press. Caulkins, D. D. 2001 “Consensus, clines, and edges in Celtic cultures.” Cross-Cultural Research 35(2):109-26. Caulkins, D. D., and S. B. Hyatt 1999 “Using consensus analysis to measure cultural diversity in organizations and social movements. Field Methods l(1): 5-26. Cavalli-Sforza, L. L., and M. W. Feldman 1981 Cultural transmission and evolution: A quantitative approach. Princeton: Princeton University Press. Chapman, A. 1992 Masters of animals: Oral traditions ofthe Tolupan Indians, Honduras. New York Gordon and Breach. Chavez, L. R., F. A. Hubbell, J. M. McMullin, R. G. Martinez, and S. I. Mishra 1995 “Structure and meaning in models of breast and cervical-cancer risk-factors: A comparison of perceptions among Latinas, Anglo women, and physicians.” Medical Anthropology Quarterly 9(1): 40-74. Clifford,J. 1983 “On ethnographic authority.”Representations l(2): 118-46. Clifford, J. 1986 “Partial truths.” Introduction to Writing culture: Thepoeticsandpolitics of ethnography, pp. 1-26, ed. James Clifford and George E. Marcus. Berkeley: University of California Press. Clifford,J. 1988 The predicament of culture: Twentieth-century ethnography, literature, and art. Cambridge: Harvard University Press. 257

REFERENCES Clifford, J., and G. E. Marcus, eds. 1986 Writing culture: The poetics and politics California Press.

of ethnography. Berkeley: University of

Cole, M., J. Gay, J. Glick, and D. Sharp 1971 The cultural context oflearning and thinking. New York Basic Books. Collins, H. M. 1982 Sociology ofscient$ic knowledge:A sourrebook. Bath, England Bath University Press. Collins, H. M. 1985 Changing order: Replication and induction in scientrfirpractice. Chicago: University of Chicago Press. Collins, R. 1975 ConfIict sociology: Toward an explanatory science. New York Academic Press. Converse,J. M., and S. Presser 1986 Survey yuesfionr: Handcrafting the standardized yuestionnaire. Beverly Hills, C A Sage. Corman, S. R., and L. Bradford 1993 “Situational effects on the accuracy of self-reported organizational communication behavior.” Communication Research 2 0 822-40. Cote, J. E. 1992 “Was Mead wrong about coming of age in Samoa? An analysis of the MeaaFreeman controversy for scholars of adolescence and human development.” Journal of Youth andAdolescence 21: 499-527. Crane, G. 1991 “Composing culture: The authority of an electronic text.” Current Anthropology 3 2 293-311. Crapanzano, V. 1980 Tuhami: Portrait of a Moroccan. Chicago: University of Chicago Press. Cronk, L. 1993 “Signals, manipulation, and Darwinian models of culture.” Paper presented at the Annual Meeting of the American AnthropologicalAssociation,Washington, D.C. Culler,J. D. 1983 On deconstruction: Theory and criticism afferstructuralism. London: Routledge. DAndrade, R. G. 1987 “Modal responses and cultural expertise.”American Behavioral Scientist 31(2): 194-202. DAndrade, R. G. 1989a “Some propositions about the relations between culture and human cognition.” In Culturalpsychology: Essays on comparative human development, pp. 65-129, ed. J. W. Stigler, R. A. Shweder, and G. Herdt. Cambridge: Cambridge University Press. DAndrade, R. G. 1989b “Cultural cognition.”In Foundations ofcognitive science, pp. 795-830, ed. M. I. Posner. Cambridge: M I T Press. 258

REFERENCES DAndrade, R. G. 1995 The development ofognitive anthropology. Cambridge:Cambridge University Press. Davies, C. A. 1999 Rejexive ethnography: A guide t o researching selves and others. New York: Routledge. Davison, J., and Women of Mutira 1989 Voices of Mutira: Lives of rural Gikuyu women. Boulder, CO: Lynne Rienner. Del Pino, G. 1989 “The unifying role of iterative generalized least squares in statistical algorithms.” Statistical Science 4 394-408. Denzin, N. K. 1992 Symbolic interactionism and cultural studies. Cambridge, England Basil Blackwell. Denzin, N. K. 1997 Interpretive ethnography: Ethnographic practices f i r the 21st century. Newbury Park, CA: Sage. Denzin, N. K., and Y. S. Lincoln, eds. 1994 Handbook of qualitative research. Newbury Park, C A Sage. Dijkstra, W., and J. van der Zouwen 1982 Introduction to Response bebaviour in the survey-interview, pp. 1-12, ed. W. Dijkstra and J. van der Zouwen. Oxford Academic Press. Dilley, R., ed. 2002 Theproblem of context: Perspectivesfiom social anthropology and elsewhere. Oxford: Berghahn Books. Douglas, M. 1966 Puriv and danger:An analysis of concepts ofpollution and taboo. London: Routledge and Kegan Paul. Dressler, W. W., and J. R. Bindon 2000 “The health consequences of cultural consonance: Cultural dimensions of lifestyle, social support, and arterial blood pressure in an African American community.” American Anthropologist 102: 244-60. Dube, E. F. 1977 A Cross-cultural study of the relationshipbetween Snte2ligence”leveland story recall. Ph.D. diss., Cornell University. Duranti, A., and C. Goodwin, eds. 1992 Rethinking context: Language as an interactivephenomenon. Cambridge: Cambridge University Press. D y e r , K. 1982 Moroccan dialogues. Baltimore: Johns Hopkins Press.

Edgington, E. S. 1987 Randomization tests. New York: Marcel Dekker. 259

REFERENCES Fabian, J, 1983 Time and the other: How anthropology makes its object. New York Columbia University Press. Fabian,J. 1991 “Dilemmas of critical anthropology.” In Constructing Knowledge:Authority and Critique in Social Science, pp. 180-202, ed. L. Nencel and F! Pels. London: Sage. Fabrega Jr., H., C. W. Ahn, J. S. Boster, and J. E. Mezzich 1990 “DSM 111 as a systemic culture pattern: Studying intracultural variation among psychiatrists.”Journal of Psychiatric Research 24(2): 139-54. Fals-Borda, O., and M. A. Rahman, eds. 1991 Action and knowledge: Breaking the monopoly with participatory action research. New York Intermediate Technology/Apex. Fardon, R. 1990 General introduction to Localizing strategies: Regional traditions of ethnographic writing, pp. 1-35, ed. R. Fardon. Washington, D.C.: Smithsonian Institution Press. Foddy, W. 1993 Constructing guestionsfor interviews and questionnaires: Theory andpractice in social research. New York Cambridge University Press. Foerstel, L., and A. Gilliam, eds. 1992 Con$-ontingthe Margaret Mead legacy: Scholarship, empire, and the South Pacific. Philadelphia:Temple University Press. Fonow, M. M., and J. A. Cook, eds. 1991 Beyond methodology: Feminist scholarship as lived research. Bloomington: Indiana University Press. Ford, C. S., ed. 1967 Cross-cultural approaches: Readings in comparative research. New Haven, CT: HRAF Press. Foster, G. 1979 “Methodologicalproblems in the study of intracultural variation: the hodcold dichotomy in Tzintzuntzan.” Human Organization 38: 179-83. Foster, G. 1987 “On the origin of humoral medicine in Latin America.” MedicaIAnthropology Quarterly 1: 355-93. Fowler, F. J. 1995 Improving survey guestions: Design and evaluation. (Applied Social Research Methods Series, Vol. 38). London: Sage. Fowler, F. J., and T. W. Mangione 1990 Standardized survey interviewing: Minimizing interviewer-related error. London: Sage. Fox, J. 1984 Linear statistical models and related methods, with applications to social research. New York Wiley. 260

REFERENCES Freeman, D. 1983 Margaret Mead and Samoa: The making and unmaking of an anthropological myth. Cambridge: Harvard University Press. Freidenberg, J., M. Mulvihd, and L. R. Caraballo 1993 “From ethnography to survey: Some methodological issues in research on health seeking in East Harlem.” Human Organization 52: 151-61. Gardner, H. 1985 The minds new science:A history ofthe cognitive revolution.New York Basic Books. Gardner, P. M. 1991 “Foragers’pursuit of individual autonomy.” CurrentAnthropology 32: 543-72. Garro, L. C. 1986 “Intracultural variation in folk medical knowledge: A comparison between curers and noncurers.” American Anthropologist 88: 351-70. Garro, L. C. 1988 “Explaining high blood pressure: variation in knowledge about illness.”American Ethnologist 15(1): 98-119. Garro, L. C. 2000 “Remembering what one knows and the construction of the past: A comparison of cultural consensus theory and cultural schema theory.” Ethos 28: 275-319. Geertz, C. 1988 Worksand lives: The anthropologist as author. Stanford Stanford University Press. Gergen, K. J. 1989 The saturated seF Dilemmas of identity in contemporary I$. Books.

New York Basic

Gergen, K. J., and M. M. Gergen 1991 “Toward reflexive methodologies.” In Research and Reflexivity, pp. 76-95, ed. F. Steier. London: Sage. Giddens, A. 1979 Centralproblemsin social theory:Action, structure and contradictionin social analysis. New York Basic Books. Giddens, A. 1984 The constitution ofsociety: Outline of a theory of structuration. Berkeley: University of California Press. Giddens, A. 1991 “Structuration theory: Past, present and future.” In Giddens’theory ofstructuration: A criticalappraisal,pp. 201-21, ed. C. G. A. Bryant and D.Jary. London: Routledge. Gifi, A. 1990 Nonlinear multivariate analysis. New York: Wiey. Goldstein, H., and R. Wood 1989 “Five decades of item response modelling.” British Journal of Mathematical and Statistical Psychology 42: 139-67. 26 1

REFERENCES Goodenough, W. 1965 “Rethinking status and role.” In The relevance of modelsfor social anthropology, pp. 1-22, ed. M. Banton. London: Tavistock. Gopnik, A., and H. Wellman 1994 “The ‘theory theory.”’ In Domain spectfirity in culture and cognition, ed. L. Hirschfield and S. Gelman. New York: Cambridge University Press. Grinker, R. R. 1990 “Images of denigration: Structuring inequality between foragers and farmers in the Ituri Forest, Zaire.”American Ethnologist 17: 111-30. Grinker, R. R. 1994 Houses in the rain forest: Ethnicity and ineguality amongfarmers and foragers in CentralAfrica. Berkeley: University of California Press. Gross, D. R. 1984 “Time allocation: A tool for the study of cultural behavior.” Annual Review of Anthropology 13: 519-58. Grusec, J. E., and H. Lytton 1988 Social development:History, theory and research. New York: Springer-Verlag. Guba, E. G. and Y. S. Lincoln 1994 “Competing paradigms in qualitative research.” In Handbook ofgualitative research, pp. 105-17, ed. N. K. Denzin and Y. S. Lincoln. Thousand Oaks, C A Sage. Guglielmino, C. R., C. Viganotti, B. Hewlett, and L. L. Cavdi-Sforza 1995 “Cultural variation in Africa: The role of mechanisms of transmission and adaptation.”Proceedings of the NationalAcademy of Sciences ofthe United States ofAmerica 92 (16): 7585-89. Hagenaars,J. A., and T. G. Heinen 1982 “Effects of role-independent interviewer characteristics on responses.” In Response behaviour in the survey-interview, pp. 91-130, ed. W. Dijkstra and J. van der Zouwen. Oxford: Academic Press. Hambleton, R. K., H. Swaminathan, and H. J. Rogers 1991 Fundamentals of item response theory. Beverly Hills, C A Sage. Hammersley, M. 1990 Reading ethnographic research: A criticalguide. London: Longman. Handwerker,W. Penn 2002 “The construct validity of cultures: Cultural diversity, culture theory, and a method for ethnography.”American Anthropologist 1 0 4 106-22. Haraway, D. J. 1988 “Situated knowledges: The science question in feminism and the privilege of partial perspective.”Feminist Studies 1 4 75-99. Hartsock, N. 1983 “The feminist standpoint: Developing the ground for a specifically feminist historical materialism.”In Discovering ReaZity, pp. 283-310, ed. S. Harding and M. B. Hintinkka. Amsterdam: D. Reidel. 262

REFERENCES Haugeland, J. 1985 Art$icial intelligence: The very idea. Cambridge, MA: MIT Press. Hewlett, B. S., and L. L. Cavalli-Sforza 1986 “Cultural transmission among the Aka pygmies.” American Anthropologist 88(4): 922-33.

Hirschfeld, L. A., and S. A. Gelman, eds. 1994 Mapping the mind: Domain spec$icity in cognition and culture. Cambridge: Cambridge University Press. Holland, D. 1987 “Culture sharing across gender lines: An interactionist corrective to the statuscentered model.” American Behavioral Scientist 31(2): 234-49. Holland, D. C., and N. Quinn, eds. 1987 Cultural models in language and thought. Cambridge: Cambridge University Press. Holmes, L. D. 1987 Quest for the real Samoa: The MeaYFreeman controversy and beyond. South Hadley, MA: Bergen and Gamey. Holmes, L. D., and E. R. Holmes 1992; 2nd ed. Samoan village then and now. Fort Worth, TX Harcourt Brace. Hosmer, D. W., and S. Lemeshow 1989 Applied logistic regression. New York Wiley. Hox, J. J. 1994 “Hierarchical regression models for interviewer and respondent effects.” Sociological Methods and Research 22: 300-18.

Hox, J. J., and I. G. G. Kreft 1994 “Multilevel analysis methods.” SociologicalMethods and Research 22: 283-99. Hubert, L. 1987 Assignment methods in combinatorialdata analysis. New York Marcel Dekker.

Huizer, G. 1979 “Research-through-action: Experiences with peasant organizations.” In The politics of anthropology: From colonialism and sexism toward a view from below, ed. G. Huizer and B. Mannheim. The Hague: Mouton.

Hull, V. J. 1986 “Dietary taboos in Java: Myths, mysteries, and methodology.” In Shared wealth and symbol: Food, culture, and society in Oceania and Southeast Asia, pp. 237-58, ed. L. Manderson. Cambridge: Cambridge University Press.

Hymes, D. H. 1969 Reinventing anthropology. New York Pantheon. Iannucci, A. L., and A. K. Rornney 1994 “Consensus in the judgement of personality traits among friends and acquaintances.”Journal of Quantitative Anthropology 4 279-95. 263

REFERENCES Ingold, T. 1993 “A social anthropological view [Comment on Tomasello et al. “Cultural learning’’].” Behavioral and Brain Sciences 16: 526-27. Irvine, S. H., and J. W. Berry 1988 “The abilities of mankind: A revaluation.” In Human abilities in rult u r d context, pp. 3-59, ed. S. H. Irvine and J. W. Berry. Cambridge: Cambridge University Press. Jameson, K., and A. K. Romney 1990 “Consensus on semiotic models of alphabetic systems.”Journal of Quantitative Anthropology 2: 289-303. Janis, I. L. 1982 Groupthink. Boston: Houghton Mifflin. Johnson, A. 1975 “Time docation in a Machiguenga community.”Ethnology 1 4 310-21. Johnson, J. C. 1990 Selecting ethnographic informants. Newbury Park, C A Sage. Johnson, J. C., and D. C. Griffith 1996 “Pollution, food safety, and the distribution of knowledge.” Human Ecology 24(1): 87-108. Johnson, K. E., C. B. Mervis, and J. S. Boster 1992 “Developmentalchanges within the structure of the mammal domain.” Developmental Psychology 28(1): 74-83. Kane, E. W., and L. J. Macaulay 1993 “Interviewer gender and gender attitudes.” Public Opinion Quarterly 57: 1-28. Keesing, R. M. 1974 “Theories of culture.”Annual Review of Anthropology 3: 73-97. Keesing, R. M. 1976 Cultural anthropology. New York Holt, Rinehart and Winston. Keesing, R. M. 1987 “Anthropology as interpretive quest.” Current Anthropology 28(2): 161-76. Keil, F. C. 1995 “The growth of causal understandings of natural kinds: Modes of construal and the emergence of biological thought.” In Causal cognition, pp. 234-62, ed. A. Premack and D. Sperber. Oxford: Oxford University Press. Kempton, W. 1981 Thefolk class9cation ofreramics: A study ofcognitive prototypes. New York Academic Press. Kenny, D. A., and D. A. Kashy 1992 “Analysis of the multitrait-multimethod matrix by confirmatory factor analysis.” Psychological Bulletin 112: 165-72. 264

REFERENCES Kimura, D. 1999 Sex and cognition. Cambridge: MIT Press. Knorr-Cetina, K. D. 1981 “The micro-sociological challenge of macro-sociology: Towards a reconstruction of social theory and methodology.” I n Advances in social theory and methodology, pp, 1-47, ed. K. D. Knorr-Cetina and A. Cicourel. New York Methuen. Krippendorff, K. 1991 “Reconstructing (some) communication research methods.” In Research and rej’exivity, pp. 115-42, ed. F. Steier. London: Sage. Kroeber, A. L., and C. Kluckohn 1952 “Culture: A critical review of concepts and definitions.” Papers of the Peabody Museum ofAmerican Archaeology and Ethnology 47( 1):1-223. Kuhn, T. 1970 The structure ofscientifc revolutions. Chicago: University of Chicago Press. Laboratory for Comparative Human Cognition 1982 “Culture and intelligence.” In Handbook ofhuman intelligence,pp. 642-719, ed. R. J. Sternberg. New York Cambridge University Press. Laderman, C. C. 1983 Wivesand midwives: Childbirth and nutrition in rural Malaysia. Berkeley: University of California Press. Lakatos, I. 1970 “Falsification and the methodology of scientific research programmes.” In Criticism andthegrowth ofknowledge, ed. I . Lakatos and N. Musgrave. Cambridge: Cambridge University Press. Lal, J. 1996 “Situating locations: The politics of self, identity, and ‘other’in living and writing the text.” In Feminist dilemmas infieldwork, pp.185-214, ed. D. L. Wolf Boulder, CO: Westview. Latour, B., and S. Woolgar 1979 Laboratory l@:The social construction ofscientlficfacts. London: Sage. Lawson, E. T., and R. N. McCauley 1990 Rethinking religion: Connecting cognition and culture. Cambridge: Cambridge University Press. Layder, D. 1990 Realist image in socialscience. London: Palgrave MacMiian. Levy-Bruhl, L. 1923 Primitive mentality. Boston: Beacon Press. Lewis, 0. 1951 L @ in a Mexican village: Tepoztlan revisited. Springfield: University of Illinois Press. 265

REFERENCES Linton, R. 1940 “Acculturation.” In Acculturation in seven American Indian tribes, ed. R. Linton. Gloucester, MA: Peter Smith. Logan, M. H., and W. T. Morrill 1979 “Humoral medicine and informant variability: An analysis of acculturation and cognitive change among Guatemalan villagers.” Anthropos 7 4 785-802. Lorenz, K. 1950 “The comparative method in studying behavior patterns.” Symposia ofthe Societyfor Experimental Biology 4 221-68. Mack, J. E. 1994 Abduction: Human encounters with aliens. New York: Ballantine. Maher, K. M. 1987 A multiple choice modelfor aggregating group knowledge and estimating individual competence. Ph.D. diss., University of California, Irvine. Manganaro, M., ed. 1990 Modernist anthropology: From3eldwork to text. Princeton, NJ: Princeton University Press. Marcus, G. E. 1986 “Contemporary problems of ethnography in the modern world system.” In Writing culture: The poetics andpolitics Ofethnography, pp. 165-93, ed. J. Clifford and G. E. Marcus. Los Angeles: University of California Press. Marcus, G. E. 1994 “What comes (just) after ‘post’?The case of ethnography.”In Handbook ofqualitative research, pp. 563-74, ed. N. K. Denzin and Y. S. Lincoln. Thousand Oaks, CA: Sage. Marcus, G. E. 1998 Ethnography through thick and thin. Princeton, NJ: Princeton University Press. Marcus, G. E., and M. J. Fischer 1986 Anthropology as cultural critique. Chicago: University of Chicago Press. Marks, G., and N. Miller 1987 “Ten years of research on the false consensus effect: An empirical and theoretical review.” Psychological Bulletin 102: 72-90. Marsh, H. W. 1989 “Confirmatory factor analysis of multitrait-multimethod data: Many problems and a few solutions.”Applied Psychological Measurement 13: 335-61. Mascia-Lees, F. E., P. Sharpe, and C. B. Cohen 1993 “The postmodernist turn in anthropology: Cautions from a feminist perspective.” In Anthropology and literature,pp. 225-48, ed. P. Benson. Urbana: University of Illinois Press. Mathews, H. F. 1983 “Context-specific variation in humoral classification.”American Anthropologist 85: 826-43. 266

REFERENCES Mayr, E. 1982 The Growth 0f biological thought: Diversity, evolution, and inheritance. Cambridge, MA: Beknap Press. McCullagh, P., and J. A. Nelder 1989 Generalized linear models. London: Chapman Hall. McMullin, J. M., L. R. Chavez, and F. A. Hubbell 1996 “Knowledge, power and experience: Variation in physicians’ perceptions of breast cancer risk factors.“MedicalAnthropology 16: 295-317. Mead, M. 1928 Coming of age in Samoa: A psychologicalstudy ofprimitive youthfor western civilization. New York William Morrow. Meyerhoff, B., and J. Ruby 1982 Introduction to A crack in the mirror: Reflexive perspectives in anthropology, pp. 1-35, ed. J. Ruby. Philadelphia: University of Pennsylvania Press. Mishler, E. G. 1986 Research interviewing: Context and narrative. Cambridge: Harvard University Press. Molony, C. H. 1975 “Systematic valence coding of Mexican ‘hot’-‘cold’ food.” Ecology of Food and Nutrition 4: 67-74. Moore, R., I. Brodsgaard, M. L. Miller, T. K. Mao, and S. F. Dworkin 1997 “Consensusanalysis: Reliability, validity, and informant accuracy in use of American and Mandarin Chinese pain descriptors.”Annalc ofBehavioralMedicine19:295-300. Murdock, G. P. 1980 Theories 0fillness:A world survey. Pittsburgh: University of Pittsburgh Press. Neisser, U., and I. E. Hyman 1999; 2nd ed. Memory observed Remembering in natural contexts. New York Worth Publishing. Newell, A. 1990 Un;fied theories of cognition. Cambridge: Harvard University Press. Okely, J., and H. Callaway 1992 Preface to Anthropology andAutobiography,pp. xi-xiv ,ed. J. Okely and H. Callaway. London: Routledge. Okonji, 0.M. 1971 “The effects of familiarity on classification.”Journal of Cross-Cultural Psychology 2: 39-49. OMeara, T. 1989 “Anthropologyas empirical science.”American Anthropologist 91: 354-69. Orans, M. 1996 Not even wrong: Margaret Mead, Derek Freeman, and the Samoans. Novato, C A Chandler and Sharp. 267

REFERENCES Orso, E. 1970 Hot and cold in thefolk medicine of the Island of Chira, Costa Rica. Baton Rouge: Latin American Studies Division, Louisiana State University. Outhwaite, W. 1987 New philosophies of social science: Realism, hermeneutics and nitical theory. Hong Kong: Macmillan Education. Pelto, l? J. 1992 “Anthropologicalresearch methods and applications:Taking stock.” In Anthropological Research:Process andApplication, pp. 259-70, ed. J.J. PoggieJr., B. R. DeWalt, and W. W. Dressler. Albany: State University of New York Press. Pelto, P. J., and G. H. Pelto 1975 “Intra-cultural diversity: some theoreticalissues.”American Ethnologist2(1): 1-18. Poggie Jr., J. J., B. R. DeWalt, and W. W. Dressler, eds. 1992 Anthropological research: Process and Application. Albany: State University of New York Press. Potter, M. C. 1990 “Remembering.”In Thinking (An Invitation to Cognitive Science, vol. 3), pp. 3-32, ed. D. N. Osherson and E. E. Smith. Cambridge: M I T Press. Pratt, M. L. 1986 “Fieldwork in common places.” In Writing culture: The poetics and politics .f ethnography,pp. 27-50, ed. J. Clifford and G. E. Marcus. Berkeley: University of California Press. Redfield, R. 1930 Tepoztlan:A Mexican village:A study offolk I$. Chicago: University of Chicago Press. Reed-Danahay, D. E., ed. 1997 Auto/ethnography: Rewriting the selfand the social. Oxford: Berg. Reinharz, S. 1992 Feminist methods in social research. New York Oxford University Press. Rhoads,J. K. 1991 Critical issues in social theory. University Park Pennsylvania State University Press. Rimer, G. 1992 Sociological theory. New York McGraw-Hill. Roberts, J. M. 1951 “Three Navaho households: A comparative study in small group culture.” Papers of the Peabody Museum ofAmerican Archaeology and Ethnology 40 (3): Roberts, J. M. 1964 “The self-management of cultures.” In Explorations in Cultural Anthropology: Essays in Honor ofGeorge Peter Murdock, pp. 433-54, ed. W. Goodenough. New York McGraw-Hill. 268

REFERENCES Romney, A. K. 1994 “Cultural knowledge and cognitive structure.” In The making ofpsychologicalanthropology 11,pp. 254-83, ed. M. M. Suarez-Orozco, G. Spindler, and L. Spindler. New York Harcourt Brace. Romney, A. K. 1999 “Cultural consensus as a statistical model.” Current Anthropology 40: S103-Sl15. Romney, A. K., W. H. Batchelder, and S. C. Weller 1987 “Recent applications of cultural consensus theory.” American Behavioral Scientist 31(2): 163-77. Romney, A. K., J. P. Boyd, C. C. Moore, W . H. Batchelder, and T. J. B r a d 1996 “Culture as shared cognitive representations.” Proceedings $the National Academy $Sciences, USA 93: 4699-705. Romney, A. K., and R. G. DAndrade 1964 “Cognitive aspects of English kinship terms.” American Anthropologist 26: 146-70. Romney, A. K., and C. C. Moore 1998 “Towards a theory of culture as shared cognitive structures.” Ethos 26(3): 314-37. Romney, A. K., and S. Weller 1989 “Systematic culture patterns in high concordance codes.” In Roberts Festchrij?, ed. R. Bolton. Washington, D.C.: HRAF Press. Romney, A. K., S. C. Weller, and W. H. Batchelder 1986 “Culture as consensus: a theory of culture and informant accuracy.”American Anthropologist 88(2): 3 13-38. Rosaldo, R. 1989 Culture and truth: The remaking of social analysis. Boston: Beacon. Rosenberg, A. 1988 The philosophy of social science. Boulder, CO: Westview. Rosenthal, R. 1991 “Replication in behavioral research.” In Replication research in the social sciences, pp. 1-30, ed. J. W. Neuliep. Newbury Park, CA: Sage. Rubin, D. 1986 Autobiographicalmemory. Cambridge: Cambridge University Press. Ruebush 11,T. K., S. C. Weller, and R. E. Klein 1992 “Knowledge and beliefs about malaria on the pacific coastal-plain of Guatemala.” American Journal of TropicalMedicine and Hygiene 46(4): 451-59. Ruebush II,T. K., S. C. Weller, and R. E. Klein 1994 “Qualities of an ideal volunteer community malaria worker: A comparison of the opinions of community residents and national malaria service staff.” Social Science andMedicine 39: 123-31.

269

REFERENCES Said, E. W. 1978 Orientalism. New York Pantheon. Said, E. W. 1989 “Representing the colonized Anthropology’s interlocutors.” Criticalhquiry 15: 205-25.

Sanjek, R., ed. 1990 Fieldnotes: The making of anthropology. Ithaca, Ny: Cornell University Press. Sapir, E. 1934 “The emergence of the concept of personality in a study of cultures.”Journal of SocialPsychology 5: 408-15. Saussure, F. de 1974 Course in general linguistics. London: Peter Owen. Scheper-Hughes, N. 1984 “The Margaret Mead controversy: Culture, biology and anthropological inquiry.” Human Organization 43: 85-93. Scheper-Hughes, N. 1995 “The primacy of the ethical: Propositions for a militant anthropology.” Current Anthropology 36(3): 409-20. Scholte, B. 1974 “Towards a reflexive and critical anthropology.” In Reinventing Anthropology, pp. 430-57, ed. D. Hymes. New York Vintage Books. Schwarz, N. 1994 ”Judgement in a social context: Biases, shortcoming, and the logic of conversation.” Advances in Experimental Social PsychologV 2 6 123-62. Schwarz, N., and S. Sudman, eds. 1992 Context efects in social andpsychological research. New York: Springer Verlag. Scott, J. 1988 Social network analysis. Newbury Park, C A Sage. Sheeran, M. J. 1983 Beyond majority rule: Voteless decisions in the Religious Society $Friends. Phdadelphia: Philadelphia Yearly Meeting of the Religious Society of Friends. Sieber, S. D. 1973 “The integration of fieldwork and survey methods.” AmericanJournal $Sociology 78: 1335-59. Smith, D. E. 1989 The everyday world as problematic. Boston: Northeastern University Press. Smith, R. B. 1987 “Linking quality and quantity, Part I: Understanding and explanation.” Quality and Quantity 21: 291-311. 270

REFERENCES Soltis, J., R. Boyd, and P.J. Richerson 1995 “Can group-functional behaviors evolve by cultural group selection?An empirical test.” Current Anthropology 36(3): 473-94. Sperber, D. 1985 On anthropological knowledge. Cambridge: Cambridge University Press. Sperber, D. 1991 “The epidemiology of beliefs.” In Psychological studies ofwidespread belit$, pp. 25-43, ed. C. Fraser and G. Gaskill. Oxford: Oxford University Press. Sperber, D. 1996 Explaining culture:A naturalistic approach. Oxford: Blackwell. Sprague,J., and M. K. Zimmerman 1989 “Quality and quantity: Reconstructing feminist methodology.”American Sociologist 2 0 71-86.

SPSS Inc. 1990 SPSS 4.0.Chicago: SPSS Inc.

Steier, F., ed. 1991 Research and reflexivity. London: Sage. Stewart, A. 1998 The ethnographer)method. Newbury Park, C A Sage. Strauss, A. L., and J. Corbin 1990 Basics of qualitative research: Grounded theory procedures and techniques. Newbury Park, CA: Sage. Strauss, C., and N. Quinn, N. 1994 “A cognitive/cultural anthropology.” In Assessing cultural anthropology, pp. 284-97, ed. R. Borofsky. New York McGraw-Hill.

Stull, D. D., and J. J. Schensul, eds. 1987 Collaborative research and social change: Applied anthropology in action. Boulder,

CO:Westview. Suchman, L., and B. Jordan 1992 “Validity and the collaborative construction of meaning in face-to-face surveys.” In Questions about questions: Inquiries into the cognitive bases of surveys, pp. 241-67, ed. J. M. Tanur. New York Russell Sage Foundation. Sudman, S., and N. M. Bradburn 1974 Response effects in surveys: Review and synthesis. Chicago: Aldine. Sudman, S., and N. M. Bradburn 1982 Asking questions:Apracticalguide t o questionnairedesign. Chicago: Aldine. Swaminathan, H., and H. J. Rogers 1990 “Detecting differential item functioning using logistic regression procedures.” Journal of Educational Measurement 2 7 361-70. 271

REFERENCES Swartz, M. J. 1982 “Cultural sharing and cultural theory: Some findings of a five-society study.” American Anthropologist 8 4 314-38. Swartz, M. J. 1991 The way the world is: Cultural processes and social relations among the Mombasa Swahili. Berkeley: University of California Press. Tinbergen, N. 1951 The study of instinct. New York Oxford University Press. Tomasello, M., A. C. Kruger, and H. H. Ratner 1993 “Cultural learning.” Behavioral and Brain Sciences 16: 495-552. Tooby,J., and L. Cosmides 1992 “The psychological foundations of culture.” In The adapted mind evolutionary psychology and thegeneration of culture, pp. 19-136, ed. J. H. Barkow, L. Cosmides, and J. Tooby. New York Oxford University Press. Torbert, W. R. 1991 The power of balance: Transforming selJ society, and scientiJir inpuiry. Newbury Park, C A Sage. Trosset, C., and D. Caulkins 2001 “Triangulation and confirmation in the study of Welsh concepts of personhood.” Journal ofAnthropoIogicalResearch 57( 1):61-81. Tyler, S. A. 1986 “Post-modern ethnography: From document of the occult to occult document.” In Writing culture: Thepoeticsandpolitics $ethnography, pp. 122-40, ed. J. Clifford and G. E. Marcus. Los Angeles: University of California Press. Tyler, S. A. 1991 “A post-modern in-stance.’’ In Constructing knowledge: Authority and critique in socialscience, pp. 78-94, ed. L. Nencel and P.Pels. London: Sage. Van Leeuwen, M. S. 1978 “A cross-cultural examination of psychological differentiation in males and females.” International Journal of Psychology 13: 87-122. van Maanen, J. 1988 Tales ofthefield: On writing ethnography.Chicago: University of Chicago Press. Van Raalte, J. L., D. D. Brewer, and D. E. Linder 1992 “NCAA Division-I1 college football players perceptions of an athlete who consults a sport psychologist.”Journal of Sport and Exercise Psychology 14(3): 273-82. Van Raalte, J. L., D. D. Brewer, B. W. Brewer, and D. E. Linder 1993 “Sport psychologists’ perceptions of sport and mental health practitioners.” Journal ofApplied Sport Psychology 5: 222-33. van W a g e n , J., B. Rylko-Bauer, and A. McElroy, eds. 1989 Making our research useJitL Case studies in the utilization of anthropological knowledge. Boulder, CO: Westview. 2 72

REFERENCES Vayda, A. F! 1994 “Actions, variations, and change: The emerging anti-essentialist view in anthropology.”In Assessing culturafanthropofogy,pp. 320-29, ed. R. Borofsky. New York McGraw-Hill. vygotsky, L. s. 1978 Mind in society: The development of higher psychological processes. Cambridge: Harvard University Press. Wallace, A. F. C. 1952 “Individualdifferences and cultural uniformities.”American SociologicalReview 17: 747-50. Wallace, A. F. C. 1970 Culture andpersonafity. New York Random House. Wasserman, S., and K. Faust 1993 Social network analysis: Methods and applications. New York Cambridge University Press. Watson, G. 1987 “Make me reflexivebut not yet: Strategies for managing essential reflexivity in ethnographic discourse.”Journal ofAnthropofogicalResearch 43: 29-41. Webb, E. J., D. T. Campbell, R. D. Schwartz, L. Sechrest, and J. B. Grove 1981 Nonreactive measures in the social sciences. Boston: Houghton Miffin. Weber, M. [1949] 1971 “Ideal types and theory construction.”In Readings in thephilosophy ofthe social sciences, pp. 496-507, ed. M. Brodbeck. London: Macmillan. Weller, S. C. 1983 “New data on intracultural variability: The hodcold concept of medicine and illness.” Human Organization 42(3): 249-57. Weller, S. C. 1984 “Consistency and consensus among informants: Disease concepts in a rural Mexican town.”American Anthropologist 86: 966-75. Weller, S. C. 1987 “Shared knowledge, intracultural variation, and knowledge aggregation.”American Behavioral Scientist 31(2): 178-93. Weller, S. C., L. M. Pachter, R. T. Trotter 11, and R. D. Baer 1993 “Empacho in four Latino groups: A study of intra- and inter-cultural variation in beliefs.” MedicalAnthropology 15: 109-36. Weller, S. C., and A. K. Romney 1987 Systematic Data Collection. Beverly Hills, C A Sage. Weller, S. C., and A. K. Romney 1990 Metric scaling. Beverly Hills, C A Sage. Weller, S. C., A. K. Romney, and D. P. Orr 1987 “The myth of a sub-culture of corporal punishment.” Human Organization 46: 39-48. 273

REFERENCES Werner, O., and G. M. Schoepfle 1987 Systematicfieldwork. Newbery Park, C A Sage. Wexler, K. N., and A. K. Romney 1972 “Individual variations in cognitive structures.” In Multidimensional scaling: Theory and applications in the behavioral sciences, pp. 73-92, ed. A. K. Romney. New York Seminar Press. White, D. R. 1990 “Reliability in comparative and ethnographic observations: The example of high inference father-child interaction measures.”Journal of Quantitative Anthropology 2: 109-50. White, W. F. 1991 Participatory action research. London: Sage. Wilson, C. S. 1970 Food beliefs andpractices of Malayjshermen: A n ethnographic study of diet on the east coast of Malaya. Ph.D. diss., University of California. Wolf, M. 1992 A thrice-told tale: Feminism,postmodernism, and ethnographic responsibility.Stanford, CA: Stanford University Press. Wulff, R. M., and S. J. Fiske, eds. 1987 Anthropologicalpraxis: Translating knowledge into action. Boulder, CO: Westview. Young, E W., and R. C. Young 1961 “Key informant reliability in rural Mexican villages.” Human Organization 20: 141-48. Zalewski, D. 2000 “Anthropology enters the Age of Cannibalism.” New York Times. Zehner, R. B. 1970 “Sex effects in the interviewing of young adults.” SociologicalFocus 3: 75-84.

274

INDEX

Numbers in italics refer to tables or figures. Abbot, Andrew, 80-81 Abraham, Curtis, 2 African Americans, 92n3 Agta, 116 Aguaruna Jivaro (Peruvian horticulturalists), 52, 120-21,201, 235,242,247n13 Alvardo, Nancy, 246-47n5, 248n15 Americans, 75,79,92n2,139,199, 214n29,222 Argyis, Chris, 140 Ashmore, Wendy, 131 Atran, Scott, 132 Aunger, Robert, 203 Axinn, William, 34,109 Bantu (in the Democratic Republic of the Congo), 17,145,201. See aZso Lese Barth, Fredrik, 70,137 Batchelder, William, 75,79,92n2, 220,229,231,234,243,246112 Batswa, 145 Beck, Brenda, 24 Behar, Ruth, 9,19n2 behavior, unconscious, 46-47 Behrens, Clifford, 141

beliefs, 16; cultural, 64-91,134; cultural, as language, 95; food related (see food taboos); humoral, 21,23-26,28-31,39,74-75, 120-27,200,211n14; and speech, 46-47; transmission of, 79, 84, 133-34,138; and values, 17; variation 23,25,28,39,55,62, 69-71,76-89,139,165,190 Bernard, Russell, 25,110,144114, 187 Boas, Franz, 1 , 2 Borgatti, Stephen, 219 Borofsky, Robert, 83 Boster,James, 28,33,37n3,40,52, 56,72, 75,97, 120-21, 143nl,235, 242,243,245,2471113 Boyer, Pascal, 86,132,142,144115 Bradburn, Norman, 33,209112 Brewer, Devon, 238,243,244 Brown, Michael, 26,39 Budu (Sudanic horticulturists), 48, 145,150,163,213n24 Campbell, Donald, 128nl Caulkins, Douglas, 70, 77,144112 Chavez, Leo, 92n2 Child, Irving, 203 275

INDEX

chi-squared test, 104, 117,210n6, 248n19 Christianity, 242,244 Clifford, James, 7 4 , 1 2 0 Cognitive Load, 195-97,212n21 Cohen’s Kappa, 248n19 Collins, Randall, 118

computers. See ethnographic practice, computers in Connecticut Yankees, 70

context, social, 99; as effect on observation, 35-37,42,59; and variability, 23-24,35-36,47,61, 89,169,181-204 Cramer’s V, 248n19 Crane, G., 19n3 Cross-Cultural Psychological Model, 27,29-30 Cultural Consensus Model, 30-32,

37nl,66-91,143nl, 190,2111-115, 217-18,224,227; critique of, 28,

33,67,75-78; as ethnographic method, 27-28,67,115n8,216-28 See afso idealism, and Cultural Consensus Models culture, defined, 94-95

Culture-Centered Models, 26-27; critique of, 36

DAndrade, Roy, 71,72-73,76,84, 93n4,144n2,243,244 Darwin, Charles, 137, 138-39

data collection: methods (see methods, ethnographic); as situation, 41-47

Davis, Charlotte, 15,101 The Democratic Republic of the Congo, 17,48,140,207 Denzin, Norman, 9,114115 Dilley, Roy, 99

Douglas, Mary, 232

Dressler, William, 92n3

educational testing, 53

Efe (pygmies of the Democratic Republic of the Congo), 48,145, 150,163,201-2 Elapsed Time effect, 54-55,197-200 elicitation effects (during data collection), 32-35, 89

ethnographic practice, 105-6; change in, 6, 12, 13-14, 18, 105-10;

computers in, 103-4,107,153; traditional, 4,12,13,25,32,59,61, 97-98,108 ethnography: authority of, 1-19,41; autobiographic, 9; children in, 70; defined, 1,64; methods of (see methods, ethnographic); as “soft science,” 113; writing, 5-9, 99

ethnomedical beliefs, 21-22, 146

ethnoscience, 37nl Fabian, Johannes, 139-40

Fischer, Michael, 7

Fiske, Donald, 1281-11

folk illness, 242

food taboos: as cultural beliefs, 23,53; definition, 146; in the Democratic Republic of the Congo, 49-62,70, 120-27,129n8,145-62; pregnancy and, 49,146,147 formal methods, 100-101,246nl. See afso reflexive analysis, methods of Foster, 30-31,121-23,128n6,200, 211n14 Freeman, Derek, 1,12,15, 119

functionalism, 95

Gardner, Peter, 201

Garro, Linda, 52, 74, 77

Geertz, Clifford, 10-11,19n3,120,130 General Linear Model, 80431,102,103 general realist approach, 138

276

INDEX

Gergen, Kenneth, 20n5,134 Goodenough, Ward, 13

Griffith, David, 77,79, 139,222,227 Guatemalan villagers, 23

Hagenaars, J. A., 34

Hammersley, Martyn, 1141-5

Haraway, Donna, 115n5 Hardwerker, Penn, 70, 74,78,97 Heinen, T. G., 34

hierarchical linear models, 103

Hippocrates, 21,74 Holmes, Lowell, 19nl Hull, Valerie, 125

humoral belief systems, 21,23-26, 28-31,39,74-75,120-27,200, 211n14 Hyatt, Susan, 77

Iannucci, Anita, 92n2 idealism, 64-91,135,137; and Cultural Consensus Models, 67-68,78-81,215-46; scientific, 66

Ika (Uganda), 2

informant effects: of knowledge in statistical models, 53-55,56-59, 60,73-74,87,165,171,200-204; of memory in statistical models, 53,54,171,173-74,195-200; question-specific response bias, 189-95,212n21; of reasoning in statistical models, 53, 87; in

statistical models, 46,52-56, 63n4, 165,188-204 Informant Error School, 24-25

informants, 36,43,45,47; ideal, 68,

72-74,88-90,240-46; influencing factors on, 44,62n3,93n4; selection of, 91n1, 110-12; “strength of belief,” 116-27

interaction effects. See informant effects; statistical models interviewer: as biasing influence on responses, 30,34,60, 187; effects (see statistical models) interviewing: change in techniques, 44,45-46; critique of use in ethnography, 42-44; as an ethnographic method, 22,32,34, 41-43; the Ituri, 148-52

item response theory, 371-12

Ituri, 49-62,68-70,73,78-80, 120-27,145-61,149,163 Johnson, Jeffrey, 77,79,110-11, 139,

222,227 Kashy, Deborah, 128111

Kenny, Davis, 128111

Kessing, Roger, 139

KiNgwana (Swahili), 149

knowledge, informant. See informant effects Kurosawa, Akira, 97

Laderman, Carol, 39,125 Lal, Jayari, 19112

Latin America, 39

Leeuwen, Van, 203

Lese (Bantu horticulturists), 48, 145,

149-50,163,201,212n17 Lewis, Oscar, 2

Lincoln, Yvonna, 114115

linear assignment model, 104

Logan, Michael, 23,30,39 loglinear models, 103

Maher, Kathleen, 238,246112 Major Culture Difference, 205-8

Malays, 23

Malaysia, 39

277

INDEX

Marcus, George, 7,9,98,114-15n5 Marsh, H. W., 128nl Mathews, Holly, 25-26,28,30,37n3, 39-40,121-23,125,128n6, 211n14 maximum likelihood fictor analysis, 219

Mayr, 137

McMullin, Juliet, 77

Mead, Margaret, 12,15, 119; debate about, 1-2

memory, informant. See informant effects methodological collectivism, 16

methodological factors, 16,52-53, 60-62,87,166,172,230 methodological individualism, 15-16

methodological situationalism, 15-16

methods, ethnographic: crisis in, 2-11; criticism of, 22,29-35,37,39; interviewing as, 22,32,34, 41-43; reflexivity in, 61,102-13, 114n4,167,170; types of traditional, 21-36,37; and variability, 41

Mexicans, rural, 200

minimum residual factor analysis, 219,

2461-12

Minor Cultural Difference, 205-8

Molony, Carol, 26

Monte Carlo, 246n2 Moore, Carmella, 89

Morrill, Warren, 23,30,39 Mountain People, 2

multitrait-multimethod approach, (MTMM), 128nl Oaxaca, Mexico, 25,26,28,121,125 observation, 41-42; observer as influence on, 35,99,119-20 optimal scaling, 103

Orso, Ethelyn, 24

Pelto, Gretel, 96

Pelto, Pertii, 96

Peru, 39,121,235 Pichataro, Mexico, 52

Plato, 137

primary data, 20n6,41-42,46,127 “primitive mentality” (Levi-Bruhl), 49

psychological testing, 27,29,53,60, 61,82,132,193 psychometric models, 27, 190,

193-94,212n19 Puerto Ricans, 70

Pure Context Models, 23

Pure Informant Model, 22-23,30 Putman, Robert, 140

Pygmies, 3. See also Efe; Tswa quadratic assignment, 104

question-dependant effects, 60

questionnaires, 26; design of, 34

Quinn, Naomi, 133

random sampling, 112

Rasch model, 37n2 Rashomon, 97

realism, 83,112,135 reasoning, informant. See informant effects Redfield, Robert, 2 , 3 Reed-Danahay, 9

reflexive analysis, 15-17,98-102, 114n5,133-43; methods of, 61,

102-13,114n4,167,170 Reflexive Analytic Model, 61,102-13, 114n4,167,170 reflexivity, 101; defined, 15,16,102; need for, 14,31,35,36; and science, 15, 18, 98; and textualists,

15

reliability of data, 118-23,128n2 278

INDEX

repeatability bias index (RBI), 116-18

Roberts, John, 57,96 Rogers, Jane, 2121118

Romney, A. Kimball, 75,77,79,80, 89-90,92n2,144nS, 220,229,231, 234,243,244,246n2 Rosaldo, Renato, 7

Samoa, 12,19111,119; adolescent sexuality and Margaret Mead, 1 , 3 Schoepfle, G. M., 110

schooling effects, 203-4,242 science: and ethnography, 18,98,112, 130,140-42; nature of, 16

Setswana, Botswana, 199

Sieber, Sam, 109

Situational Variation School, 23-24,

31

slash-and-burn horticulture, 48

Smith, Diana, 140

social context. See context, social social roles: as influence on interview responses, 52-53,57,200-204 social surveys, 107-10

sociology, 33

sociology of science, 32,98 South Americans, 122

South Indians, 24

Sperber, Dan, 84, 132,133, 138, 142

SSPS logistic regression procedure, 168-69,209n3 standpoint theory, 9

statistical models, 111;informant effects, 46,52-56,63n4,165, 188-204; interaction effects between interviewer and informant, 50,55-56,164,170, 205-8; interviewer effects, 33-34,

44,45,51-52,60,179-88; measurement error in, 47, 105;

multivariate, 50-51,80, 103, 136,

164,167,169,196; unexplained variation in, 61,95 Steier, Fredrick, 102

Stein, Gertrude, 12

Stewart, Alex, 20n7,128n2 Strauss, Claudia, 133

structuralism, 95

Sudanic horticulturalists, 48, 145,151, 182,201,229-46 Sudman, 33,209112 Swahili, 52,149, 150,187 Swaminathan, Hariharan, 212n18 taboos, food. See food taboos Tepoztlan, Mexico, 2 , 3 Texas, 242

textualists, 4,97-98; critique of ethnographic practice, 4-7,

19-20n4,105,137; defined, 4

Thematic Apperception Test, 2471111

Tlaxcala, Mexico, 28,40 Torbett, William, 140

Trosset, Carol, 70

Tswa (pygmies), 48

Turnbull, Colin, 2 , 3 typological thinking, 96

Tzintzuntzan, Mexico, 30,121 valence, of food. See humoral belief systems variation: between belief systems, 116-27; cognitive, 53-54,57, 88-91; in consensual belief, 69-71,

78,88,139; contextudsituational, 23-24,35-36,47,61,89,169, 181-204; cultural, in belief, 23,25, 28,39,55,62,76-89, 165, 190;

cultural, and ethnography, 4,12, l8,96; cultural, between informants, 31,39,53,56,58, 113nl,131-34,136,143; and data

2 i'9

INDEX

collection, 41; definition of ethnographic, 5; within informants, 25,30,31,39-40 Wales, 70 Wallace, Anthony, 57 Watson, Graham, 115n5 Weber, Max, 65,136,143 Weller, Susan, 28,33,37n3,40,56, 72,75,77,79,92n2,97,220,229, 231,234,242,243

Werner, Oswald, 110 White, Douglas, 118, 119-20 Wilson, Christine, 23 Young, Forrest, 122 Young, Robert, 122 Zaire, 122 Zehner, R. B., 214n29 Zuni, 96, 116

280

ABOUT THE AUTHOR

Robert Aunger trained as a biological anthropologist; his Ph.D. is from the University of California, Los Angeles. He has taught psychology, biology, and anthropology at Northwestern University, the University of Chicago, and King’s College, Cambridge. He has been working for fifteen years on the theoretical, methodological, and empirical problems of studying cultural evolution. This has resulted in a series of academic articles, as well as a book, The Electric Meme:A New Theory of How We Think (2002), and an edited volume, Darwinizing Culture: The Status of Memetics as a Science (2001). He spent two years living with the foragers and horticulturalists of the Ituri Forest, Democratic Republic of Congo. Dr. Aunger has discussed his work on radio, TV, and the Web. He is currently a lecturer at the London School of Hygiene and Tropical Medicine.

281