#2989-ASR 71:6 filename:71607-weinreb—p.1014

The Limitations of Stranger-Interviewers in Rural Kenya Alexander A. Weinreb Hebrew University, Jerusalem Virtually all survey data are collected by “strangers,” that is, individuals with no prior social relationship with respondents. Although it has long been recognized that attitudes toward strangers vary cross-culturally, there has been no systematic discussion of how this variation might affect survey data. This article attempts such a discussion, using data from a longitudinal research study in rural Kenya. It reviews qualitative impressions of insider- and stranger-related issues within the specific Kenyan field setting, drawing primarily on field notes. Relevant areas of social theory and the data collection literature are reviewed briefly. Finally, using the project’s longitudinal survey data, empirical tests are presented which allow for an evaluation of differential data quality across insider- and stranger-interviewers on three dimensions: differential response rates, differential reliability of responses, and differential response validity. The results suggest that insider-interviewers increase response rates and collect more consistent data across survey waves. They also suggest that data collected by female insiders in particular appear to be superior for most questions and of equal quality for others.

t has long been recognized that the validity of survey data is affected by the attitudes of

respondents toward the strangers who interview them, and that attitudes toward strangers vary cross-culturally. Yet a century of social research has yielded no systematic review of relevant theoretical or methodological issues concerning the collection of survey data by strangers. Nor has it yielded any estimates of the size and scope of “stranger-interviewer” effects.1 Rather, where the issue has cropped up at all, it has been in either tangential asides in early method-

Direct correspondence to Alexander A.Weinreb, Department of Sociology and Anthropology, Hebrew University, Jerusalem, Israel 91905 ([email protected]). The data used in this article were funded by grants from the Rockefeller Foundation and National Institutes of Health to Susan Watkins and Jere Behrman. The author owes an incredible debt to Susan, who encouraged this research project at every opportunity, and gave freely of her field notes and time. The author also thanks Jerry A. Jacobs and Amelia Weinreb for their extremely helpful substantive comments on a number of versions, and Rafe Stolzenberg, Herb Smith, Randall Collins, and four anonymous reviewers for their comments on the earliest versions.

1 There has been no systematic experimental, empirical, or conceptual treatment of this issue in any single article in the American Journal of Sociology, volumes 1–107; American Sociological Review, volumes 1–69; Public Opinion Quarterly, volumes 1–66; Sociological Methodology, volumes 1–32; Sociological Methods and Research, volumes 1–31; Social Science Research, volumes 1–31; Population Studies, volumes 1–56; or Demography, volumes 1–39. Nor have I found any such treatments in scholarly monographs that in all other ways effectively review and summarize the core literature on response effects (e.g., Fowler and Mangione 1990; Hagenaars and Heinen 1982; Tourangeau, Rips, and Rasinski 2000).

Since there may be marked differences among people in different countries in conceptions of what it means to give information to strangers .|.|. careful comparative studies are imperative before we can know how generally true specific findings may be .|.|. —Sudman and Bradburn (1974:147)

I

AMERICAN SOCIOLOGICAL REVIEW, 2006, VOL. 71 (December: 1014 –1039 )

#2989-ASR 71:6 filename:71607-weinreb—p.1015 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA—– 1015

ological reviews, wherein the possibility of such effects has been acknowledged but then ignored empirically (Mauldin 1965; Stycos 1960; Sudman and Bradburn 1974), or in more specific criticisms of survey research as practiced in the developing world (Heisel 1968; Stone and Campbell 1984). The key aim of this article is to address this gap in the literature by presenting baseline estimates of insider- and stranger-interviewer effects in a single rural area of Kenya. Its structure follows from this. In the first of four main sections, I define “insider” and “stranger” interviewers, briefly describing the perceived advantages of stranger-interviewers over insider-interviewers. In the second section, I briefly review related issues in the extant sociological and data collection literature. In the third section, I describe initial qualitative impressions of insider- and stranger-related issues within the specific Kenyan setting in which this research was conducted, drawing primarily on field notes that describe respondents’ reactions to interviewers and to the research project. Then in the final section, I present empirical tests, noting signs of differential data quality across insider- and stranger-interviewers in the associated survey data on three levels: (1) differential response rates, (2) differential reliability of responses, and (3) differential response validity. BACKGROUND The single most important characteristic that distinguishes insider-interviewers from strangerinterviewers is whether they know or are known by the respondent or the respondent’s family. “Know” refers to a social relationship unrelated to and predating specific interactional roles associated with interviewers and respondents. This definition of “insider” and “stranger” is derived from two main sources. The first source draws on a classical sociological theme, Simmel’s reference to “entanglement in family and party interests” (Simmel 1950a:404) as the key characteristic of an insider. The second source draws on emic perspectives specific to the Kenyan research settings. As described later, a perpetual source of tension during data collection was the extent to which the research project used individuals affiliated with the sampled communities. Those from other villages in

the area, unless they also had some personal, usually familial, connection to someone in the sampled villages, were called “outsiders” or “foreign” (Kawadhgone 2000:112).2 Survey researchers use stranger-interviewers (i.e., interviewers who have no prior social relationship with a respondent or a respondent’s family) for two main sets of reasons. The first is purely practical. An insider is restricted to households that she or he knows. In contrast, a stranger-interviewer can knock on any door or call any number. Stranger-interviewers are therefore more flexible data collectors than insiderinterviewers. This practical difference also has implications for sampling and research design. Over and above these practical concerns, however, more traditional epistemological reasons exist for using stranger-interviewers. These build on one of two ideas about micro level social interaction: either people are “more likely to be reticent with those who are already known to them” (Bulmer 1993:215) as a general rule, or an interviewer–respondent interaction is a distinct type of social interaction in which this is the case. The origins for each of these ideas also can be located in Simmel (1950a). Thus, for example, the argument that strangers in general, but especially “the stranger who moves on .|.|. often receives the most surprising openness—confidences which sometimes have the character of a confessional and which would be carefully withheld from a more closely related person” (Simmel 1950a:404). The reason for this openness, Simmel suggests, is that a stranger typically is deemed to have more objectivity, which arises from both “a particular structure composed of distance and nearness, indifference, and involvement” (p. 404) and “freedom .|.|. from commitments which could prejudice his perception, understanding, and evaluation of the given” (p. 405). Over decades of research activity, the performance of stranger-interviewers, as opposed

2 Kawadhgone (2000:112) refers to page 112 of field notes from the field site of Kawadhgone written during the 2000 wave of data collection. Other site names with associated field notes on which I draw in what follows are Gwassi (1994, 1995, 2000), Oyugis (1994, 2000), and Wakula South (1994, 1996).

#2989-ASR 71:6 filename:71607-weinreb—p.1016 1016 —– AMERICAN SOCIOLOGICAL REVIEW

to the more general types of strangers described by Simmel, has tended to support these hypotheses. In the early days of survey research (e.g., see Hyman et al.’s (1954) discussion of sex surveys in the United States and the United Kingdom), researchers were surprised at how readily respondents seemed to discuss deeply personal issues with strangers. Since then, researchers have tended to forget that early surprise. They also have developed techniques that, by framing either specific questions or the environment in which the question is asked, increase reporting of sensitive, offensive, or threatening behaviors, giving epistemological cover to strangers. The result has been the proven effects on data quality of “confidentiality assurances,” promises of “response anonymity,” telling the respondent that his or her views are important, and attempts to minimize interactional biases by ensuring privacy during the interview or by increasingly substituting noninteractional data collection modes such as self-administered questionnaires and audio- and computer-assisted interview techniques (Audio-CASI & CASI) (Aquilino 1994; Krysan et al 1994; Mangione, Hingson, and Barrett 1982; Mensch, Hewett, and Erulkar 2003; Singer, von Thurn, and Miller 1995; Taietz 1962). GENERAL LITERATURE As noted earlier, no extant methodological literature exists that deals with the relative effectiveness of insider- and stranger-interviewers. However, the more general literature on strangers is, frankly, enormous. I have reviewed relevant themes of this literature in some detail elsewhere (Weinreb 2000). But it is useful to at least summarize two key issues here. First, the basic proposition that respondents are more honest with strangers than with insiders, the old Simmelian claim, draws on assumptions about microsocial relations that at best are inconsistent with and at worst antithetical to a large body of social theory at the micro level. This body of theory includes (1) historical and contemporary equation of truth-telling with intimacy (Collins 2004; Garf inkel 1967; Goffman 1971; Tönnies 1961), (2) the experimental and semiotic literature on microsocial interaction (Breault, Barker, and Lemle 1987; Catania et al 1996; Christenfeld et al. 1997; Dindia, Fitzpatrick, and Kenny 1997; Givens

1978), and (3) the literature on deception (Barnes 1994; Simmel 1950b). It also is inconsistent with (4) elements of macro-oriented theory, particularly ideas about how trust conventions (the propensity to trust or mistrust a stranger) are constituted (Douglas 1970; Durkheim [1926] 1964; Giddens 1990; Levy 1972; Molm, Takahashi, and Peterson 2000). Second, a number of methodological experiments make it clear that survey respondents do not neatly distinguish between interpersonal norms and task-oriented interview-related norms, even in highly differentiated societies for which it can be reasonably assumed that individuals are used to telling quite personal things to professionals of one sort or another (e.g., doctors, lawyers, therapists). Rather, there is considerable overlap between these two types of interpersonal communication, suggesting that communication between interviewers and respondents alternates between formal rule following and tacit practices (e.g., Maynard and Schaeffer 2002; Schaeffer and Presser 2003). Indeed, the fact that mainstream survey methodologists are increasingly adopting more flexible styles of interviewing over fully standardized— and impersonal—interviewing techniques, shows that this overlap can be leveraged to increase the accuracy of survey responses (Cicourel 1974; Dijkstra 1987; Schober and Conrad 1997; Suchman and Jordan 1990). As yet, no systematic methodological research of this type has been conducted in developing societies. But field researchers’ accounts cast considerable suspicion on the assumption that respondents distinguish between interpersonal norms and task-oriented interview-related norms. For example, there are numerous accounts of culturally sanctioned, and even culturally prescribed, deception of stranger-researchers in developing societies, some with comical and others with quite serious results (e.g., Agar 1980:61; Anderson 1986:342–3; Barley 1983:89; Dentan 1968:92; Stone and Campbell 1984:32). Bulmer (1993) refers to these as the “sucker bias,” a phenomenon specific to strangers, because insiders, having mastered local cultural codes, know how to avoid being suckered. From a methodological perspective, Agar (1980) and Stone and Campbell (1984) are particularly significant because they match prior survey reports from

#2989-ASR 71:6 filename:71607-weinreb—p.1017 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA —– 1017

given respondents in rural India and Nepal, respectively, with postenumeration reports gathered from informal interviews in communities where they were embedded. In both cases, respondents reported telling untruths to the survey interviewers because the latter were strangers whose motives were unknown and therefore suspect. Finally, although the substantive focus of this article is on insider- and stranger-interviewer effects, analyses of response patterns also consider the interaction between gender of the interviewer and insider–stranger issues. This builds on a long of tradition of research investigating different types of “role independent” interviewer effects in general (much of it reviewed in Fowler and Mangione [1990], Hyman et al. [1954], Sudman and Bradburn [1974]), and on a small literature comprising interviewer gender effects in developing-country settings (Axinn 1991; Becker, Feyistan, and Makinwa-Adebusoye, 1995; Bignami-van Assche, Reniers, and Weinreb 2003; Blanc and Croft 1992). IMPRESSIONS FROM THE FIELD Following methodological norms, officially sanctioned data collection practice in subSaharan African countries, as elsewhere, includes choosing interviewers from a cohort of professionals who have appropriate language skills, then bringing them into the sampled area for the duration of data collection. In the subsequently discussed research project, this option was too expensive for us, forcing us to use local interviewers. This gave rise, albeit organically rather than through official design, to an interest in insider–stranger issues. In this discussion, it is useful to introduce the research project from which the data in this article are drawn, and to describe two important characteristics of the field setting. I then summarize key impressions about insider–stranger differences that emerged from the fieldwork and use these to anchor ideas that I subsequently test in the analysis section.

THE SETTING I use data from the Kenya Diffusion and Ideational Change Project (KDICP). Based in four sites in South Nyanza District, Nyanza Province, in southwest Kenya, the KDICP col-

lected three rounds of survey data between 1994 and 2000.3 The main focus of the project was on the role of social networks in contraceptive and AIDS-related behavior. It sampled from a population of ever-married women and the husbands of those currently married. Although the specific survey instruments changed somewhat across time, each wave of the survey asked a series of questions about relatively sensitive topics including household economics, gender relations and attitudes, behaviors and conversational networks related to AIDS, and family planning. More generally, as a setting for fieldwork, South Nyanza, as well as the other predominantly Luo districts in Nyanza province (all districts except Kisii and the new Nyamira and Suba districts) is one of the poorest areas in Kenya.4 (Weinreb 2001a). Some of this poverty is attributable to the area’s relative distance from the hubs of economic and nongovernmental organization networks in Nairobi. However, according to both the popular imagination and the popular political discourse among the Luo, the main cause of Nyanza’s impoverishment is the fact that, as an ethnic group, Luo have en masse been associated with the political opposition since at least 1966, which has led to reduced flows of government spending, either directly or through parastatals (Weinreb 2001b). Either way, residents of South Nyanza District are aware of their relative lack of access to various types of capital and public

3 All these data, the survey, and the field notes, are now in the public domain (available at http:// kenya.pop.upenn.edu). Also, more specific information on sample characteristics, data quality, and so on is presented at the beginning of the empirical section. 4 Since the sample was initially drawn, South Nyanza has been subdivided into three districts. Consequently, two sites (Gwassi and Wakula South) currently are in Suba District; one (Oyugis) is in Karachuonyo District; and the fourth (Kawadhgone) is in Homa-Bay District. Irrespective of this redistricting, this article uses the slightly outdated district nomenclature as the summary name, preferring it over both the three new names and less geographically specific alternatives such as “Nyanza Province,” which includes other districts, including the nonLuo Kisii District.

#2989-ASR 71:6 filename:71607-weinreb—p.1018 1018—–AMERICAN SOCIOLOGICAL REVIEW

goods relative to parts of the country that historically have been more favored (particularly areas of the Central and Rift Valley provinces). In each of the four sites and across all three rounds of the survey, we selected interviewers from the large local pool of under- or unemployed Form 4 graduates, that is, high school graduates who lived in or around one of the four research sites, or who happened to be visiting the area when interviewer positions were advertised. Inevitably, this meant that a proportion of the interviews conducted during the project were performed by interviewers who claimed to know the respondent or someone in the respondent’s household well. Because our approach deviated from the widely accepted standard, there was discussion from the outset about the bias that might ensue from using local interviewers.

RESPONDENTS’ VIEWS OF INTERVIEWERS As KDICP data collection proceeded, we became increasingly comfortable with our choice of local interviewers. At the root of this comfort was our growing awareness of two factors. First, tension between local suspicion of our motives, on the one hand, and appreciation for our economic contributions and the opportunities they afforded some local residents, on the other hand, appeared to generate data collection problems. Second, insider-interviewers appeared to be better at minimizing these problems than stranger-interviewers. I discuss each of these in turn. LOCAL SUSPICION. Local suspicion of the KDICP stemmed from three factors. First, the relatively low levels of social differentiation in these communities meant that respondents typically had quite limited experience with stranger-oriented interactions in general, or certainly with anonymous interrogators pushing them to assign motives to our presence in lieu of this experience. As one suspicious motherin-law of a respondent put it: “You people .|.|. you must be after something” (Gwassi 1995:24). Second, because of its partial focus on contraception, the KDICP often was publicly associated with the family planning movement, which hurt it in some quarters, while probably helping it in others. Equally damaging was its association, in local eyes, with overtly feminist

ideas. In relatively traditional patriarchal settings such as Nyanza, these alone were likely to raise at least some hackles. Finally, suspicions also fed on more local political perceptions. Particularly in 10 semistructured interviews with older men (transcripts available at the project Web site) and countless informal conversations, local residents displayed a standardized “political opposition discourse” centered on the idea that Luos had borne a disproportionate share of the costs of independence, both before and after Kenya’s transition to multiparty democracy in 1992. Among the core ingredients of this discourse were conspiratorial theories about the murder of their most able leaders, the relative absence of effective Luo representation in the central government, and systematic discrimination against them in terms of public services provided. Although the specific implications of these factors varied somewhat across informants, all respondents in some way expressed a lack of conf idence in the cur rent authorities. Unfortunately, as with large field projects in many other settings, it was these same authorities from which we—the KDICP—needed to receive official permits to conduct our research. Whether because of a single one of these factors or a combination of them, there was some suspicion of the KDICP among individual respondents. This manifested itself in an unwillingness to believe our mission as laid out in the introductory statement, which, following standard survey protocol, every interviewer made to a respondent before the formal start of the interview. Instead of that stated mission, or perhaps in addition to it, respondents appeared to assign us motives that more neatly fit local discourses, their own ends, or both. The most extreme of these motives were a small number of unusual but influential cases that signaled respondents’ quite serious mistrust of us and our intentions. During a number of interviews, KDICP interviewers were told to “shake the dust from their feet” (Gwassi 1995:24) (i.e., to get out quickly). In a handful of cases, they were chased away by panga (machete)-wielding male respondents who accused them of being agents of the government or devil worshippers. On one occasion, a respondent wielding a panga in one hand and a Bible in the other arrived at our doorstep early one morning, the day after he had been interviewed,

#2989-ASR 71:6 filename:71607-weinreb—p.1019 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA—– 1019

demanding that we return the hard copies of his and his wife’s questionnaires. He claimed not to have slept at all the previous night because of concerns that he and his wife would die if we did not give them to him. We did so (Gwassi 1995:35). More frequently, this devil-worshipping theme played itself out in an outright refusal to participate in the survey, although unaccompanied by even an implicit threat of violence. Most commonly, however, it simply peppered other local conversation about the research project.5 All in all, these responses—and other angry ones such as “those whites, they want to know what we think and they go back and laugh” (Gwassi 1995:12)—resemble colorful examples drawn from early fertility and household surveys in Jamaica (Back and Stycos 1959), Bangladesh (Choldin, Kahn, and Ara 1967), India (Agar 1980), and elsewhere in Kenya (Kearl 1976). In these cases, too, stranger-interviewers were alternately thought to be agents of the secret service, army conscription agencies, or a compulsory population control program, or simply “godless women heralding the end of the world” (Choldin, Kahn, and Ara 1967:251). These issues are directly relevant to debates about how effective stranger-interviewers could be in our setting, for the simple reason that the ulterior motives respondents ascribed to us appeared to be mediated through the social identity of the interviewer. Moreover, in the case of the KDICP, it was the insider-interviewers’ mediation that appeared to work more effectively than that of stranger-interviewers in calming fears or reducing suspicions. Examples of this mediation are given in the following discussion. INTERVIEWS AS AN OPPORTUNITY FOR GAIN. If suspicion caused only a small minority of individuals to refuse outright to participate in the survey (actual nonresponse rates are presented in Section 4), it was largely because the project was also seen as a potential source of economic assistance or as an agent of development. Again, this general problem affects research

5

See Gwassi (1995:21, 25; 2000:25, 81, 92) for more details on why some of the respondents came to associate our project with devil worship.

projects in many developing countries (Belcher et al. 1976; Bulmer 1993; Choldin, Kahn, and Ara 1967; Stone and Campbell 1984). In the KDICP, it manifested itself in a number of ways. First, it drew on the fact that, as noted earlier, we were using people, most of them from the immediate area, and that the general lack of employment opportunities in the area generated intense competition for jobs as interviewers, also pushing chiefs and others into requesting, and sometimes demanding, that we employ certain people (Gwassi 1995:19; Gwassi 2000:58–59; Oyugis 2000:54). Indeed, in some cases, respondents refused to participate in the survey if no one from their own immediate or extended family was hired in any capacity by the project (e.g., Gwassi 1995:22). In the most serious of these cases, a whole clan declared that its members would refuse to participate in the survey after it became known that no prospective candidate associated with the clan had passed the interviewer selection tests. Only after we threatened to pull the project from the whole area—a threat that speedily engaged the intervention of the local chief, assistant chief, and members of other clans (whose children or villagers’ children were directly benefiting from our presence)—were the leaders of the recalcitrant clan persuaded to lift their embargo (Wakula South 1996:70–74). Second, in a similar vein, there was also at least one case in which a chief was accused of taking bribes to change the names on our interviewer selection list (Gwassi 1995:19), although he had no access to the lists. In another case, a local leader was accused of minimizing advertising of the positions outside his clan (Oyugis 1994:18). The KDICP also was considered a potential source of economic assistance or development beyond its role as an employer. Friendly informal conversations with locals, whether they were in the sample or not, often were followed by “invitations” to assist them with school fees, college fees, money for setting up a business or for local development infrastructure, and so on. Less frequent, although not uncommon, were offers of marriage or even land (on which either one of the foreign researchers or Kenyan university-educated field supervisors would settle), each of which would entail the establishment of patron–client ties, and therefore transfer obligations. In short, we repeatedly received the

#2989-ASR 71:6 filename:71607-weinreb—p.1020 1020—–AMERICAN SOCIOLOGICAL REVIEW

impression that locals’ attempts to cultivate relationships with the presumably wealthy and well-connected researchers and senior field staff were, in part at least, instrumental. Properly harnessed, we were their pathway to “barely imaginable opportunities” (Kawadhgone 2000:100). Most commonly of all, however, our presumed development role generated daily demonstrations of respondents’ desire to give us what we wanted so that we might effectively help them as individuals, or else help their communities. At the simplest level, this emerged in the field staff’s daily reports of respondents asking “what is this interview going to do for me?” But this motivation to benefit also emerged in other ways. For example, in a handful of cases, a respondent was impersonated by someone else (e.g., Oyugis 1994:12) or, because of a logging error on our part, we mistakenly interviewed the same respondent twice within the same few days without the respondent acknowledging the preceding interview (Gwassi 1995:47). In each of these cases, the respondents later reported thinking that they might derive some extra benefit from having played these (extra) roles as respondents. However, we also saw signs of a desire to benefit, individually or collectively, in respondents’ sometimes frustrated response to certain questions, including some seemingly innocuous ones. They could not figure out why we were asking a specific question, and consequently could not figure out the best way to respond. In one case, for example, a female supervisor conducted an interview with a difficult male respondent: When she got to the “when were you and your first wife married” question, [the respondent] said “That’s the sort of question I won’t tolerate. What will this help me with?” He answered a few more [questions], but at the radio question [Do you own a radio?] blew up, and said: “Why are you asking about radios? Are you going to give me one?” (Gwassi 1995:11–12)

In another case, a male respondent physically took the questionnaire from the interviewer and told her to leave. When the supervisor visited him to ask for an explanation, and for the questionnaire, he told her: This research is useless; you can’t be asking what we own; you should offer a solution like Bamako and CARE. If you want family planning you should be straightforward about that; if you want AIDS

you should be straightforward about that. (Kawadhgone 2000:125)

Finally, this desire to appear worthy of our presumed development role also emerged in local leaders’ advice to respondents and in their informal comments to us. For example, respondents reported being told by one local health educator “to keep their houses very clean in case we visited” (Wakula South 1994:26). And in informal conversations with field staff, local informants sometimes distanced themselves from the devil-worshipping rumors that they thought might damage the project’s perceived development potential, often painting the accusers as more general troublemakers (Gwassi 1995:35). Alternatively, they exaggerated local attitudes toward core project interests to make the area sound more deserving of our attentions. For example, during the first wave of the survey, when the substantive focus was on contraceptive use, one local leader reported that “everyone” likes family planning (Gwassi 1995:14–15)—a complete falsification according to data on contraceptive prevalence rates for the immediate area (e.g., from both KDICP data and the South Nyanza subsample of the 1993 Kenya Demographic and Health Survey data). Moreover, by the final wave, in which the project’s core substantive interest had shifted to AIDS, a local leader reported that 80 percent of the local population had died of AIDS and that 90 percent of the children younger than 5 years were orphans (Gwassi 2000:86–7). This too was absolutely groundless empirically. But like the family planning comment 5 years earlier, it sent us a message. Local leaders thought this is what we wanted to hear, or that it was the best way to engage us in solving local development problems. I return to some of these issues in the empirical Section 4 because they indicate how social desirability biases were constituted, in predictable ways, thereby allowing for some analytic leverage. THE LIMITS OF PRIVACY. Our comfort with local interviewers also grew as we realized that the effectiveness of the standard methods used by the KDICP to minimize interactional biases in the data also appeared to be affected by the tension between local suspicion of the KDICP and the opportunities represented by the project. For example, we instructed interviewers to

#2989-ASR 71:6 filename:71607-weinreb—p.1021 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA—– 1021

try to minimize interruptions from the outside and to stop any third parties from listening to, or otherwise interfering in, the interview. The problem was that in addition to the layout of the households and their age structure—respectively, dwellings often were small and clustered in extended family (multigenerational) compounds, and high fertility meant that individual families within these compounds also tended to be large—our attempts to restrict the flow of people in and out of earshot were partly undermined by the fact that the very act of participating in the survey as a respondent signaled the ownership of knowledge. Thus, we came across mothers-in-law or husbands who, because they felt threatened by their daughters-in-law or wives being interviewed in private, would try, sometimes successfully, to interrupt or terminate the interview;6 village headmen who directed us to more educated members of the community or failed to list some of the more problematic households during our initial construction of sample lists (they were found later during fieldwork); headmen or other higher-status individuals who demanded to be interviewed, even if they did not fall into the sample frame, to signal their status and, implicitly, their support for our “development” project (see Rudolph and Rudolph [1958] for similar status-related themes in India).7

6 More generally, female field supervisors mentioned university men, complaining about “modern women” being those who “were always talking back” (Wakula South 1994:25). It is reasonable to assume that parts of this attitude also existed at the village level, with implications for the behavior of women respondents. 7 Recently, a small number of researchers who work in developing country settings have sought to minimize interactional biases by experimentally substituting self-administered questionnaires or CASI/Audio-CASI modes of data collection for faceto-face interviewing. The results to date are mixed, generating somewhat higher reports of sensitive behavior, but with some unexpected error (Mensch 2006, personal communication; Mensch, Hewett, and Erulkar 2003; Plummer et al. 2004). This switch to alternative modes does not in itself resolve the insider versus stranger issue. It is possible, for example, that CASI/Audio-CASI would yield even higher reports of sensitive behavior if they were introduced to respondents by insider-interviewers.

Similarly, and perhaps more important, the KDICP also used confidentiality assurances and promises of response anonymity, following what has long been standard protocol for developing societies (e.g., Demographic and Health Surveys 1987:46; United Nations 1961). Here too, however, we began to suspect that the credibility of these assurances was only as strong as the respondent’s confidence in the promise giver, in this case, the interviewer, the organization she or he represented, or both. Without that confidence, no matter how vociferous the promises were, it seemed they did relatively little to negate people’s fear of telling the “wrong people” about an undesirable activity or attitude, where the wrong people increasingly appeared to be strangers over whom respondents had no leverage.8

INSIDERS AS PROBLEM SOLVERS Over and above repeated reports that by hiring locals as interviewers we were providing some tangible “benefits” to the community—considered a positive thing locally, even with the competition it generated—we emerged from the field with considerable anecdotal evidence that insider-interviewers were able to assuage respondents’ fears much more effectively than stranger-interviewers. In some cases, insiders did “damage control” after respondents had demonstrated their unhappiness with strangers. These stories fall into three main categories. COMPLETING INTERVIEWS. Frequent reports suggested that insiders could more successful-

8

No extant literature demonstrates the utility of these types of assurances in developing country settings. For example, all 113 of the studies analyzed in the Singer, Von Thurn, and Miller’s (1995) metaanalysis of the utility of confidentiality assurances were conducted in developed societies. In developing societies, in contrast, it has been asserted that “[t]he anonymity of survey research is generally incomprehensible” (Kroeger 1983:467), and that, therefore, emphasizing the confidentiality of responses in these contexts “may not be necessary or even desirable” (Hershfield et al. 1993:244). There also are recorded cases showing extremely undesirable effects of confidentiality assurances (e.g., Back and Stycos 1959).

#2989-ASR 71:6 filename:71607-weinreb—p.1022 1022—–AMERICAN SOCIOLOGICAL REVIEW

ly leverage the fact that they knew the respondents into completed interviews, either through finding missing respondents or by neutralizing their unwillingness or suspicions. For example, one respondent claimed to be too injured to answer the questionnaire until the supervisor sent an interviewer who was a clan member, and who discovered no injury at all (Oyugis 1994:18). Another respondent told her interviewer, whose family she knew: “If it wasn’t you, the son of so and so, we would demand 50 bob for the interview” (Gwassi 1994:37; see also Kawadhgone 2000:115–116). In another case, the supervisor claimed that the respondents were always friendly to a particular interviewer because they all knew that he was “someone’s son” (Gwassi 1994:49). Moreover, when the questionnaire stealer, described earlier, was eventually persuaded to complete the questionnaire, he justified his change of heart with the statement that he knew both the family of the interviewer, saying “at least its not an outsider” (Kawadhgone 2000:125), and the supervisor’s father.

prescient observation: “One would think that women talking to women .|.|. would get more responsive respondents, but that doesn’t seem to be the case—is it because the interviewers are strange, i.e. not their confidantes?” RESPONSE ACCURACY. In addition to diminishing nonresponse, insider-interviewers’ familiarity with certain individuals or families—over and above the more general information about, say, local politics that all local interviewers had—also made them a fount of information about certain respondents. In particular, the realization that “they [knew] individual histories” (Oyugis 1994:24) meant that somewhat unusual stories reported in interviews could be cross-checked, or that when being interviewed by an insider, the respondents themselves had much less leeway to lie about things that were publicly known, even if they wanted to lie. I return to this point in the analysis. ANALYSIS

ITEM-RESPONSE PATTERNS. Two types of anecdotal evidence also suggested that insider-interviewers were generating better data. First, there were reports from the field. On a verbal autopsy section, for example, a supervisor described respondents’ sour expressions in response to questions about the death of a family member: “You are just somebody coming out of the blue, they know you aren’t related” (Oyugis 2000:55). The implicit message was that without a prior social relationship, stranger-interviewers’ sympathies must be fake; they are not asking for the right reasons. Similarly, in a focus group before the initial survey wave, informants “said they trust the CBD workers [Community Based Distribution agents used to distribute contraception, and in this area almost always from one’s own village], even the men .|.|. said they talk freely because they’ve known them a long time’ (Wakula South 1994:20). Second, there were emerging patterns in the data. During the first wave of the survey, for example, simple tabulations of the entered data—data entry was conducted during fieldwork in all three waves—showed that a higher proportion of women interviewed by women reported no network partners than women interviewed by men. In response, the principal investigator’s field notes contain the following quite

Field impressions and a review of relevant issues in the extant sociological and data collection literature push us in the same direction. Specifically, in a setting such as rural Nyanza, they provide very weak support for the classical position in which stranger-interviewers are seen as outperforming their insider-interviewer counterparts. On the contrary, both field impressions and the general sociological literature suggest that insider-interviewers generate (1) higher response rates, (2) greater response reliability, and (3) greater response validity than their stranger-interviewer counterparts. In addition, field impressions suggest that a given interviewer’s insider versus stranger status visà-vis a given respondent is salient to both sensitive and nonsensitive questions, suggesting that (4) the scope of stranger-interviewer effects extends across an array of variables. I now test these hypotheses empirically, although I switch the order of 3 and 4. These final two sections of the analysis also explore the interaction between interviewers’ insiderness and sex, building on three discrete literatures. First, there is a small but suggestive literature on interviewer gender effects in other subSaharan African settings (Becker et al. 1995; Blanc and Croft 1992), particularly in the KDICP data (Bignami-van Assche et al. 2003).

#2989-ASR 71:6 filename:71607-weinreb—p.1023 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA —– 1023

Second, the fact that the sampled population is patrilineal, patrilocal, and patriarchal, with relatively fixed gender roles across a range of everyday tasks, implies that an interviewer’s sex would be more salient to a larger number of questions than other role-independent characteristics such as marriage, educational level, or fertility.9 Third, there is some anecdotal evidence from informal conversations with supervisors that respondents preferred talking to a male interviewer they knew than to an anonymous female interviewer (e.g., Wakula South 1994:20). I begin, however, by introducing relevant aspects of the KDICP data not yet described.

THE DATA Three rounds of data collection were conducted on the KDICP: in 1994–1995, 1996–1997, and 2000. The data used in the following analysis are from the second and third waves, henceforth referred to as K2 and K3, because only in these waves were data on interviewers’ characteristics collected. The sample design used a simple clustersampling strategy. That is, four rural sublocations (administrative areas that include multiple villages) in South Nyanza were selected. Within each of these sublocations, villages were randomly selected, and all ever-married women of reproductive age (younger than 50 years) known to be de jure residents during the initial survey wave (K1) were interviewed. The villages then were resampled before K3 to introduce a new respondent panel. This yielded interviews with 924 women, who constitute the main sample for most of the analyses. The exception is a series of models that explore the effects of selectivity between K3 and the prior 1996–1997 round (K2) using merged K2/K3 data files. In K2, 779 women were interviewed, 611 of whom were successfully interviewed in K3, for a follow-up rate of 78 percent.

9 Exploratory analyses showed this to be the case. Parallel models using other observed interviewer characteristics, some of which are allegedly important in developing-country research (e.g., Ware [1977] recommends using married interviewers in fertility surveys) found much weaker effects (results available from the author).

INTERVIEWERS’ SELECTION, CHARACTERISTICS, ASSIGNMENTS

AND

At all four sites and in all rounds of the survey, male and female interviewers were selected from a pool of Form-4 graduates, described earlier. Altogether, 97 interviewers (61 men and 36 women) were employed (in K3). In general, the mean age of the interviewers was 24 years. Approximately one third of the interviewers were married. One third had at least one child, and three fifths had worked as interviewers before (the vast majority in prior waves of this project). In addition, consistent with the norms of patrilocal exogamy in this population, none of the married women interviewers and only 23 percent of the unmarried ones were born in the sampled sublocations. The effects of these social characteristics on responses in these data—“roleindependent” interviewer effects (Sudman and Bradburn 1974)—have been assessed elsewhere (Bignami-van Assche, et al. 2003). To assess their level of acquaintance with a given respondent, interviewers were instructed to answer the following question after completing each interview: How well do you know the respondent’s family? 1. Not at all 2. By name only 3. Quite well 4. Very well

In the following analysis, stranger-interviewers are those who claimed not to know the respondent’s family at all (Code 1), and insider-interviewers are those who reported knowing the respondent’s family quite well or very well (Codes 3 and 4). These two are combined because of the relatively small cell size. There also is an intermediate category of “acquaintance-interviewers,” which comprises those who claimed to know respondents families “by name only.”10 Table 1 shows the distribution of completed K3 women’s interviews by level of interview-

10 These data may therefore underestimate the total stranger-interviewer effect because all interviewers were from the area, and thus fluent in local dialects and idioms, whereas normal practice in developing-country research, as noted, is to bring in professional stranger-interviewers from the outside, usually from urban areas.

#2989-ASR 71:6 filename:71607-weinreb—p.1024 1024—– AMERICAN SOCIOLOGICAL REVIEW

Table 1. Distribution of Female Respondents, by Type of Interviewer Interviewer Knows the Respondent N (Row %a) Stranger

Acquaintance

Insider

Totalb

Obisa Owich Kawadhgone Wakula South

115 (41.8) 091 (50.0) 095 (36.0) 059 (29.1)

100 (36.4) 064 (35.1) 121 (45.8) 063 (31.0)

060 (21.8) 027 (14.9) 048 (18.2) 081 (39.9)

275 182 264 203

Female Interviewer Male Interviewer

169 (41.4) 191 (37.0)

140 (34.3) 208 (40.3)

099 (24.3) 117 (22.7)

408 516

Total

360 (39.0)

348 (37.7)

216 (23.4)

924

a

Note: Row % may not sum to 100.0 because of rounding. b Totals refer to completed interviews.

er’s insiderness, and by interviewer’s sex and sublocation. In all, nearly 40 percent of the interviews were conducted by stranger-interviewers, and slightly more than 20 percent by insider-interviewers. The remainder fall into the acquaintance category. Notwithstanding local norms of patrilocal exogamy, little of this insider–stranger variation is associated with the interviewer’s sex. There is, however, more significant variation by research site. For example, the percentage of interviews conducted by insiders ranges from 14.9 percent in Owich to 39.9 percent in Wakula South. This distribution fits with general characteristics of these areas because Wakula South is on the small island in Lake Nyanza-Victoria. It makes more sense, therefore, that the social networks in which these interviewer–respondent relations are embedded would tend to be different. Interviewer training followed standard procedures. In addition to working through the nuts and bolts of the survey instrument over several days, trainers repeatedly emphasized the importance of neutral responses and nondirective probing. The questionnaire also included a confidentiality assurance at the beginning of the interview, and three reminder assurances at later points. It should be noted that although these are not experimental data, it is reasonable to have confidence in the magnitude of observed response variability. Not only have methodological studies frequently used nonexperimental data to explore broad patterns of response effects (e.g., Berk and Bernstein 1988; Schaeffer 1980; Sudman et al. 1977; van Tilburg 1998), but postfieldwork selectivity checks in this case also confirm that interviewer assignments were large-

ly random with respect to interviewers’ insiderness. In particular, interviewer’s insiderness in K3 is totally independent of the respondent’s observed socioeconomic and demographic characteristics (educational attainment, wealth, income, family size), as well as AIDS-related attitudes and behavior and gender attitudes, as measured in K2, even at a liberal 10 percent significance level. In fact, the only K2 variables related to K3 insiderness are four indicators of mobility and attitudes toward family planning.11

INSIDER- AND STRANGER-INTERVIEWERS’ EFFECTS ON RESPONSE RATES Our impression that an insider-interviewer was more likely than a stranger-interviewer to return from the field with a completed questionnaire after any of the up to three reported visits to a given respondent is reflected in actual response rates. This was because of differences in both the overt and the numerically much more significant covert nonresponse. By overt nonresponse, I refer to an explicit refusal to participate in the survey, amounting to 2.9 percent of all household

11 The K3 interviewers were less likely to know respondents who previously reported having been to Nairobi (chi2[3df] = 10.6; p = 0.014), or being able to speak Kiswahili, the main African language in Kenya (chi2[3df] = 15.7; p = 0.001). They also were more likely to know respondents who had previously reported larger conversational networks (chi2[57df] = 141.2; p < 0.001) and, related to this, respondents who had talked to their husbands about family planning (chi2[3df] = 6.35; p = 0.012).

#2989-ASR 71:6 filename:71607-weinreb—p.1025 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA —– 1025

Table 2. Response Consistency Across Survey Waves, by Type of Interview Stranger

Acquaintance

Insider

Speaks Kiswahili Speaks English Has Ever Been to Nairobi Has Ever Used a Method of Family Planning

.79 .89 .86 .89

.86 .98 .89 .90

.89 1.00 .89 .94

Number of Children Ever Born Number of Deceased Children Size of Family-Planning Related Conversational Network Size of AIDS-Related Conversational Network

.74 .73 .50 .62

.74 .72 .42 .51

.85 .90 .65 .65

Mean Consistency Score

.75

.75

.85

Note: A consistent answer is defined in one of two ways. Either the respondent gave the same response in both waves. Else the respondent, in K2, (a) claims not to be able to speak Kiswahili or English, or to have never been to Nairobi or ever used contraception, but reports the opposite in K3; or (b) reports a lower (lifetime) number of children ever born, children deceased, or network size in K2 than K3.

visits for strangers and 1.3 percent for insiders. By covert nonresponse, I refer to respondents claiming to be too busy to be interviewed, or to having a family member reporting that the respondent is temporarily away. The covert nonresponse rate hovered around 45.4 percent for strangers and around 36.1 percent for insiders across all three visits.12 Together, these differences in overt and covert nonresponses across levels of interviewers’insiderness meant that stranger-interviewers returned with a completed questionnaire after 52 percent of their visits to respondents’households, as compared with 63 percent for insider-interviewers.

INSIDER- AND STRANGER-INTERVIEWERS’ EFFECTS ON RESPONSE RELIABILITY The collection of interviewer-related information in the K2 and K3 waves of the survey allows us to explore the differential reliability

12

Low overt refusal rates are typical in developing-country surveys, most likely because projects typically require government permits and the blessings of all local headmen (as did the KDICP). This means that only those with a history of run-ins with the local headman or those who otherwise care little about being considered a “troublemaker” overtly refuse. In contrast, covert refusal is a much more legitimate avoidance mechanism. People slip out the back door as interviewers approach the front, claim to be someone else, say that they are about to go to a funeral, or the like.

of responses across insider- and stranger-interviewers across two survey waves. This measure of test–retest reliability, also known as the “simple response variance,” is an important component of the mean square error of the estimated mean (Bailar 1976) and a standard indicator of valid measurement (Zeller and Carmines 1980). The key problem in estimating this type of test–retest reliability is that many characteristics, behaviors, and attitudes can legitimately change between two waves, especially where, as in this case, there is a lengthy gap between the waves (Litwin 2003). To avoid this, I restricted the analysis to a small subset of questions about relatively fixed characteristics, the only such questions asked in either wave. For each of these questions, I defined a logically consistent answer in one of two ways. Either the respondent gave the same answer in both waves, or she reported a transition that might reasonably have occurred. For example, between the two waves, she might well have visited Nairobi, learned more Kiswahili or English, used contraception for the first time, and so on. Or she might well have given birth to more children, had more children die, or developed a larger conversational network size for ever having talked to people about AIDS or family planning. In contrast, logically inconsistent answers were defined as those in which the woman dropped her earlier admission that she had ever visited Nairobi, could speak some Kiswahili or English, had ever used contraception, and so on. Table 2 presents the results from this test–retest analysis, with the reliability expressed

#2989-ASR 71:6 filename:71607-weinreb—p.1026 1026—–AMERICAN SOCIOLOGICAL REVIEW

as the proportion of respondents who gave a consistent answer, as defined earlier, in both rounds. The comparison is made across the three categories of respondents (Columns 1 to 3): those who in both K2 and K3 were interviewed by stranger-interviewers, by acquaintance-interviewers, or by insider-interviewers. The sample sizes are too small to allow much leverage (only 20 individuals were interviewed by insiders in both rounds), but the trend is unambiguous. Across all eight questions, there is a clear and consistent pattern in which women interviewed during both rounds by insider-interviewers gave more reliable responses than those interviewed during both rounds by strangerinterviewers. The respective mean consistency across the eight questions is 0.85 for insiderinterviewers and 0.75 for stranger-interviewers. In contrast, those interviewed by acquaintance-interviewers appear no more nor less reliable than those interviewed by strangerinterviewers.

THE SCOPE OF INSIDER- AND STRANGERINTERVIEWER EFFECTS To assess the scope of stranger-interviewer effects, a series of multivariate models were estimated in which an assortment of response variables were regressed on stranger-interviewer characteristics. Specifically, 127 variables were selected from eight distinct sections of the questionnaire: general characteristics, household

wealth, earnings, use of and attitudes toward family planning, family planning–related conversational networks, gender attitudes, AIDSrelated behavior and attitudes, and AIDS-related conversational networks. Among the 127 variables were both the main dependent and independent variables of the project and the core sociodemographic control variables. Standard estimation procedures were used: logit regression on dichotomous variables, tobit on continuous earnings data, and Ordinary Least Squares (OLS) on other continuous variables. Two types of models were estimated. The Model 1 group explored variation across three types of interviewers: strangers, acquaintances, and insiders. Stranger-interviewers, representing the standard type of interviewer in survey research, were defined as the reference group. The Model 2 group introduced an interaction term between each of the insider and stranger classifications and the interviewer’s sex, yielding six distinct types of interviewers: female stranger, female acquaintance, female insider, male stranger, male acquaintance, and male insider. Here, female stranger-interviewers, the standard type of interviewer in demographic and health surveys, were defined as the reference group. Summary results are presented in Table 3 (full results for the Model 2 group of estimates are available at Web site Appendix A). With respect to the scope of the stranger-interviewer effects, the results are unambiguous. In the

Table 3. Response Variation by Type of Interviewera Variable Category

Number of Variables

Model 1

Model 2

General Characteristics Household Wealth Individual Earnings Use of and Attitudes to Family Planning Family Planning Related Conversational Networks Gender Attitudes AIDS Related Behavior and Attitudes AIDS Related Conversational Networks

017 021 015 007 017 020 013 015

11 07 03 02 05 10 06 07

15 11 07 03 05 12 07 07

Total

127

51

67

Note: All models were run using logit regression on dichotomous variables, Tobit on reported earnings data, and OLS regression on other continuous variables. The ‘Model 1’ group explores response variation across 4 levels of insiderness; the ‘Model 2’ group explores variation across 6 categories, representing the interaction between 3 levels of insiderness and the interviewer’s sex. a Full model results available at ASR website (url: .|.|.|. ).

#2989-ASR 71:6 filename:71607-weinreb—p.1027 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA—– 1027

Model 1 group, different responses are given to stranger- and insider-interviewers in 51 (40 percent) of the variables, with significance levels set at the 5 percent level. This includes about two thirds of the selected general characteristics variables, notwithstanding the fact, as discussed earlier, that an interviewer’s insiderness in K3 is almost completely independent of a respondent’s observed socioeconomic and demographic characteristics, as measured in K2. It also includes about half of the gender attitudes and AIDS-related variables, and about one third of the wealth and family planning variables. Allowing for the interaction with sex of the interviewer in Model 2 increases the observed scope of stranger-interviewer effects. Specifically, different responses are given to stranger- and insider-interviewers in 67 (53 percent) of the 127 selected variables, with particularly strong effects on reported general characteristics and gender attitudes. In summary, there is a generalized strangerinterviewer effect. Respondents tended to report different things to stranger- and insider-interviewers on about half of the selected variables. Moreover, whereas some of these were core variables of interest, and some could be thought of as sensitive variables, many others were neither core variables of interest nor overtly sensitive variables.

VALIDATING RESPONSES To identify which of the six interviewer types used in the Model 2 group of Table 3 appear to generate the most accurate type of data, we must be able to validate survey responses. The absence of external data against which given responses can be compared means that we must do this inferentially: (1) by drawing on standard models of the response process that describe how respondents evaluate a series of possible responses to any given question (Cannel, Miller, and Oksenberg 1981; Tourangeau, Rips, and Rasinski 2000), and (2) by conforming to a standard assumption in the response effects literature, that is, that better methods of data collection generate higher reports about undesirable phenomena or lower reports about desirable phenomena (e.g., Aquilino 1994; Axinn 1991;

Becker et al. 1995; Blanc and Croft 1992; Krysan et al. 1994).13 The key to inferring response validity is therefore to identify the direction of the social desirability bias for any given behavior or attitude, and then to assess the relative proximity of any group’s aggregated responses to questions about that behavior or attitude. To this end, 11 questions were selected from those used to construct Table 3. Three are related to sources of income, five to gender attitudes, and three to AIDS. Some of the questions are behavioral, whereas others are attitudinal. Most important, for all 11 questions, the trajectory of the social desirability bias can be identified with some certainty. SOURCES OF SOCIAL DESIRABILITY. Daily meetings with supervisors, who reported local attitudes toward the project and frequent conversations with elders, and others, who approached the research team for conversation during fieldwork clarified views of the research project and helped define the factors that created socially desirable responses to given questions in the Nyanza setting. Here I list three of these factors. The first and dominant factor was poverty, specifically, the lack of labor opportunities, lack of access to capital, general hunger for development, and respondents’ difficulty distinguishing the KDICP research goals from more development-oriented projects. These led respondents to tend to exaggerate their poverty as a strategy to trigger material assistance. I refer to it as the “poverty response motive.”

13 The external data that do exist cannot help here.

For example, the field notes on which I drew in the first part of the article can sometimes be matched with specific survey respondents, but tend to focus on more general background characteristics of the individuals than on their responses to individual questions. Similarly, other surveys have been administered in the area (e.g., the South Nyanza subsample of the 1993, 1998, and 2003 Kenyan Demographic and Health Surveys), but the sampling frame was different, and all interviewers were the equivalent of female strangers, undermining comparisons between aggregate response patterns.

#2989-ASR 71:6 filename:71607-weinreb—p.1028 1028—–AMERICAN SOCIOLOGICAL REVIEW

The second factor drew on public health promotions, ubiquitous in Kenya throughout the 1990s, including messages and programming on the radio, posters for clinics and schools, billboards, and other public advertisements. These messages often emphasized fidelity, the importance of open discussions about sexual practice between spouses, and the use of condoms.14 They also built on approximately 20 years of family planning campaigns and a more recent focus on women-specific programs in general, including efforts to change gender attitudes (Watkins 2000). The prevalence of these public health discourses meant that respondents would tend to assume that attitudes promoted by modern public health campaigns should be normative, and therefore that foreign and/or urban development workers wanted modern responses if they were to continue assisting local populations. I refer to this as the “modernity response motive.” The third factor, whose effects were very powerful but limited to a small subset of variables, was HIV/AIDS-related stigma. In the KDICP research setting, there were numerous signs of AIDS-related stigma. Frequently, references to AIDS used local code words (e.g., chira, a condition resulting from the breaking of traditional taboos). Respondents systematically underreported family members’ mortality from AIDS or HIV infection, even while they acknowledged its high prevalence in the community, or excessive mortality. Moreover, wives known to be HIV+ frequently were kicked out of their homes by their husband or in-laws and told to return to their natal families. In short, the strength of this stigma suggested that respondents would tend to underreport anything relat-

14 The intensity of AIDS-related social marketing was particularly pertinent in Nyanza because the latter has the highest HIV prevalence in Kenya (Office of the President et al. 2004). In a 1999 seroprevalence survey in neighboring Kisumu district, for example, also dominated by Luos, 30.1 percent of the sampled adults had positive test results for HIV (Buvé, Caraël, Hayes et al. 2001). At the research sites represented in these data, adult mortality appears to have increased threefold during the 1990s (Doctor and Weinreb 2005), as reflected during data collection in the constant cycle of postponed interviews attributable to respondents being “away at a funeral.”

ed to their own infection or chances of infection. I refer to this as the “stigma response motive.” I TEM -S PECIFIC D IRECTION OF S OCIAL DESIRABILITY BIAS. Given the strength of these attitudes toward poverty, public health, and HIV/AIDS, I assumed that these three determinants of response affected all KDICP respondents irrespective of whether they were interviewed by an insider- or a stranger-interviewer, or by a male or female interviewer. Thus, insider- and stranger-interviewer effects would emerge as intermediate effects, mediating the extent to which the desire to benefit would affect an actual response. I now describe the expected trajectory of social desirability on each item as well as the expected mediating effects of insiders and strangers. I assumed that the expected social desirability bias on sources of income questions would be driven by the poverty response motive, depressing reports about these sources (e.g., “No, I don’t get any income from selling things from my shamba [agricultural plot], from a small business, or from membership in a credit merry-go-round”). However, because these questions address a type of information that is partly public knowledge—the types of business activity listed here can be observed directly, and the cyclical beneficiaries of credit merrygo-rounds are likely to be known—I also assumed that respondents interviewed by insiders would have less leeway to lie comfortably than those interviewed by strangers. As noted in the prior qualitative section, supervisors identified this quite early in fieldwork as an important advantage of insider-interviewers (see also Barnes 1994). Gender attitude questions also seemed to be affected by the dominant poverty motive, although in this case it related to a poverty of perceived rights vis-à-vis men rather than financial resources. Thus, socially desirable responses would emphasize the relative lack of rights because that might trigger programs (and hopefully an inflow of resources) to establish womenspecific business initiatives, or “women’s groups” in general. Consistent with the popular “women in development” paradigm (Jacquette 1990), this was an important part of development on the ground in Kenya during in the 1980s and 1990s, such that by 2000, there were 25,000 registered women’s organizations,

#2989-ASR 71:6 filename:71607-weinreb—p.1029 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA—– 1029

Table 4. Frequency with which Observed Responses were Furthest from, or Closest to, the Expected Socially Desirable Response, by Type of Interviewer. Distance from Socially Desirable Response Furthesta

Closest

Female-strangers 00 1b Female-acquaintances 00 0b Female-insiders 11 0 Male-strangers 00 3 Male-acquaintances 00 2 Male-insiders 00 4b Notes: a Source for furthest and closest are models in Appendix A. b These three types of interviewers were joint closest to the expected socially desirable response on one of the eleven items.

almost all of which had been established in the preceding two decades (Morel-Seytoux 2000). In addition, because Luo society is traditionally patriarchal, patrilineal, and patrilocal, emphasizing the relative lack of women’s rights also made cultural sense. 15 Finally, as with the sources of income questions, I assumed that the extent to which a respondent would be able to exaggerate her poverty of rights would be limited in interviews with insiders. The expected social desirability of responses across the three AIDS questions appeared to draw on a different combination of motives. I assumed that responses to the first two questions, which asked how frequently the respondent had conversations about AIDS with her husband and about her comfort suggesting that he use condoms, drew primarily on the modernity response motive, pushing any given women toward answering “yes” to both questions. The other AIDS-related question addressed how likely the respondent thought she was to become infected in the future. Here I assumed that the stigma response motive would depress reported risk, implying that higher reports of perceived risk would tend to be more accurate.

For expository ease, the predicted responses across all six categories of interviewer’s insiderness and sex are presented in two ways. First, they are graphed in Figures 1a to 1c (tabulated results with associated levels of statistical significance are presented in Appendix A). Second, the overall frequency of each type of interviewer’s proximity to the most and least socially desirable responses is summarized in Table 4. In each case, it should be noted that these estimated coefficients are net of controls for other observed interviewer characteristics, the most significant of which (despite varied effects across variables) tended to be previous experience working as an interviewer and marital status. In the absence of stranger-interviewer effects, we expect horizontal lines in Figures 1a to 1c either across all six interviewer categories or, where there is an interviewer gender effect, with a step between interviewer categories 1, 2, and 3 and categories 4, 5, and 6. That is not the case here. None of the response categories are horizontal across the different types of inter-

15 In theory, the modernity response motive may have pushed a relatively small group of respondents in the opposite direction, because by exaggerating their rights vis-à-vis men, they could signal their internalization of egalitarian discourses, thereby winning points from potential aid “investors.” On the aggregate level, this is not a convincing argument. It would have had to overcome both the dominant poverty response motive and culturally entrenched gender attitudes.

16 Steps taken to assess the possible effects of selectivity (e.g., the extent to which nonrandom interviewer assignment affected estimated response variability by interviewer’s insiderness) are discussed on Web site Appendix B. Models also were specified that explored variation in insider–stranger effects across the four research areas. Because there were very few differences between sites, and nothing systematic, the results presented in this section are “main effects” of interviewers’ insiderness.

RESULTS16

#2989-ASR 71:6 filename:71607-weinreb—p.1030 1030 —– AMERICAN SOCIOLOGICAL REVIEW

Figure 1a–1c.—Predicted Response Values by Category of Questions and Type of Interviewer.2 Notes: 1 FS = female stranger; FA = female acquaintance; FI = female insider; MS = male stranger; MA = male acquaintance; MI = male insider. 2 Predicted values are net of other interviewer characteristics.

#2989-ASR 71:6 filename:71607-weinreb—p.1031 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA—– 1031

viewers. Moreover, as shown in Appendix A, and consistent with the high prevalence of effects noted in Table 3, in 9 of the 11 cases, the differences between female stranger-interviewers (the reference category in the regressions from which these values are derived) and either female or male insider-interviewers are statistically different, at the 5 percent level. More specific to the main aim of this section, which is to compare responses across interviewers’ levels of insiderness and their sex, three broad explanatory patterns can be identified in the predicted values. MORE VALID DATA REPORTED TO FEMALE INSIDER-INTERVIEWERS. This is the dominant pattern among the three broad explanatory patterns.. As summarized in Table 4, for all 11 variables, the responses to female insider-interviewers appear least prone to the social desirability bias. In seven of these, differences in reports to female insiders and in the reference category are statistically significant. It is also useful to note the scale of these reporting differences. Among the respondents interviewed by female insider-interviewers, for example, 31 percent agree that a husband has the right to leave a wife who fights him frequently. Among the respondents interviewed by all other types of interviewers, 52 to 60 percent agree with this statement. The responses to sources of income and AIDS-related questions, presented in Figures 1a and 1c, respectively, tell a similar story. The responses of women interviewed by female insiders were furthest from the most socially desirable response for all six questions. Thus, women interviewed by female insider-interviewers appear to have the most sources of income, but also appear to be least comfortable suggesting condom use to their spouses, to report talking with him about AIDS the least frequently, and to have the highest perceived risk of HIV infection. In a poor setting with high HIV prevalence and repeated public calls for marital fidelity, these differences are conceptually important. SHIFTING DIFFERENCES BETWEEN MALE AND FEMALE INSIDERS. None of the 11 variables show a statistically significant gender-of-interviewer effect across stranger-interviewers

(between female stranger and male stranger) in Appendix A. Rather, virtually all gender-ofinterviewer effects are found in the differential pattern of responses to insider-interviewers, and only in one case is there a difference across acquaintance-interviewers. This appears visually across the figures in the tendency of the male and female interviewer slopes to move in different directions, a suspicion largely confirmed in Table 4. The latter shows that in contrast to female insiders’ success in avoiding socially desirable responses, described earlier, a plurality of the most socially desirable responses—5 out of 11—were given to male insider interviewers. ABSENCE OF AN INSIDER PREFERENCE OR A PREFERENCE FOR STRANGERS. The is the most interpretively conservative explanatory pattern. There is no statistically significant response variation of any type for 2 of the 11 selected variables, and none related to female insiderinterviewers for 4 of the 11. Therefore, even if the general shape of the distribution mirrors statistically significant variation in similar variables, stranger-interviewers are frequently no better and no worse than insider-interviewers. DISCUSSION The communicative rituals from which these response effects emerge, or not, as described in the last few paragraphs, are the unknown in this analysis. Without data to explore the linguistic and general semiotic architecture of the survey interaction at the most micro level, we have no way of knowing how what Maynard and Schaeffer (2002:4) call “situational oscillations” and “analytic alternations” vary by insider- and stranger-interview across any of the variables in these Kenyan data. But the product of those oscillations and alternations is clear. At least in this setting, there appears to be a considerable insider–stranger response bias for many variables, and a systematic bias across specific types of questions, especially if we examine the aggregated patterns of slope effects presented in Figures 1a, 1b, and 1c, instead of restricting ourselves to independent items significant at the 5 percent level. More specifically, in accordance with the social desirability criteria, the results suggest that data collected by female insider-interview-

#2989-ASR 71:6 filename:71607-weinreb—p.1032 1032 —–AMERICAN SOCIOLOGICAL REVIEW

ers are superior for many questions and of equal quality for several. Conversely, as described from the other side of the coin, data collected by female strangers and, to a lesser degree, data collected by male strangers often is inferior to, but never better than, data collected by female insiders. In addition, and more generally, data collected by insiders in general appear to be more consistent across survey waves. Moreover, insiders also are more successful at actually finding respondents and returning with a completed questionnaire. In this setting, therefore, we can reasonably conclude that insider-interviewers, especially female insiders, appear to be generally better data collectors than strangerinterviewers. The key question, of course, is whether we can generalize from this single Kenyan setting to other settings. In combination with the established theoretical literature and other accounts of fieldwork, I think it is reasonable to do so, although conservatively. For example, I would expect that parallel effects could be identified in other areas of the developing world in which there are similar structural characteristics to those found in South Nyanza. These characteristics would include a general mistrust of strangers and groups associated with the government. To the extent that this is true, it suggests that the results presented here may also inform data collection practices in developed countries with relatively large immigrant populations or other minorities in which, for a number of reasons, there is continued suspicion of outsiders. For until or unless members of these groups develop the same tolerance for pesky strangers, and more specifically for the strangerinterviewers who, albeit politely, demand to know personal things, response patterns may be affected by insider–stranger concerns of the type illustrated in this analysis. This argument is in line with trends in recent methodological literature. In fact, it may be possible to coopt some parts of that literature into a more general insider–stranger paradigm. For example, recent research on ways to facilitate emotional comfort between interviewers and respondents have tended to emphasize one of two strategies, both recommended in some form by Cicourel (1974) three decades ago. The first, focused on research in developed country settings, has emphasized the development of interpersonal interviewing styles. These

have been shown to increase both the content and accuracy of responses (Dijkstra 1987; Schober and Conrad 1997; Suchman and Jordan 1990), and therefore have been welcomed, albeit with some residual suspicion, in the methodological mainstream (e.g., Beatty 1995; Schaeffer 1991; Tourangeau 1990). The second strategy, which has relied less on formal methodological tests than on reviews of field experiences, has largely been a by-product of research in developing countries and has focused on building social relationships in a preliminary interview (e.g., Axinn, Fricke, and Thornton 1991) or on using interviewers from the same setting (e.g., Ross and Vaughan 1986). Mapped onto the framework described in this analysis, the effectiveness afforded by each of these methodological strategies derives from the fact that they aim to more closely approximate interactional patterns among insiders than among strangers, an approach already modeled by a caste of qualitative interviewers (Jansen 1980; Holstein and Gubrium 1995; Mayoni 1983). The possibility remains, however, that the marginal gains from using an interpersonal interviewing style will be limited in certain research settings. Training interviewers to assume the stance of a friend by being simultaneously professional and informal will not be enough. It will not negate the respondent’s suspicion of strangers and, consequently, facilitate the collection of higher quality data. Only the “real thing,” an insider known by the respondent, or one whose family is known by the respondent, will be able to do this. CONCLUSION As I have argued, the fact that we almost always employ stranger-interviewers to collect data does not mean that we should do so uncritically. In certain areas of the world, insider-interviewers may collect more data, as well as more reliable and accurate data, on a surprisingly large range of topics. Foundational social theory suggests that this relative success of insiderinterviewers over stranger-interviewer counterparts is related to variation in the visibility of and tolerance for strangers, and that the very presence of strangers asking questions indexes a certain level of social differentiation. This begs full empirical verification. However, this single Kenyan case, in which I have pre-

#2989-ASR 71:6 filename:71607-weinreb—p.1033 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA —– 1033

sented baseline estimates of insider- and stranger-interviewer effects, is consistent with the underlying hypothesis. Important conceptual gaps remain. It is useful to at least mention a few of them. First, as noted earlier, recent research suggests that communication between interviewers and respondents alternates between formal rule following and tacit practices (Schaeffer and Presser 2003). To the extent that this is the case, it seems likely that these alternations are somewhat different where interviewers are strangers as opposed to insiders. If this represents the microsocial origins of these response effects, it would be useful to observe the enactment of these alternations, seeing precisely where they differ across these two types of interviewers. Second, the 11 questions on which the most detailed analysis was undertaken deal with relatively sensitive or threatening topics: sources of income, women’s autonomy, and AIDS-related questions. But it is important to recall the generalized insider–stranger effects noted in Table 3: statistically significant variation in response distributions across levels of insiderness for 67 of 127 variables. In other words, there is reason to believe that insider–stranger effects equally influence both sensitive and nonsensitive questions, or at least what we think are sensitive or nonsensitive questions in given contexts. This is in contrast to interviewer effects more generally, which tend to be limited to sensitive variables in relation to which interviewers’ social identity is the most “salient” (e.g., Schaeffer and Presser 2003; Sudman and Bradburn 1974). Third, another question raised by the analysis relates to the differential quality of data collected by female and male insiders. Although the matching of interviewers and respondents according to sex has been criticized in developing-country research (Becker et al. 1995; Blanc and Croft 1992), it remains standard practice for survey research in general, and for developing countries in particular, mainly because of the widely held belief in what Warren (1988) calls the “focal gender myth of field research,” that is, the belief that women interviewers are less threatening and have better rapport, and that this leads to higher quality data, especially in interviews with women (e.g., Oppenheim 1992:84). The presented analysis supports a middle position between the two

sides of the debate. The response distribution on most of the variables suggests that, at least in this setting, women respondents do not favor female stranger-interviewers over male interviewers, but that they do favor female insider-interviewers. Thus, there are no gender-of-interviewer effects per se in this setting. Rather, these function through insider–stranger effects. Again, further research is necessary to explore the extent to which this type of interaction in interviewer effects exists in other research settings. At the very least, however, it raises a more general question about the extent to which roleindependent interviewer effects, long observed in social sciences, sometimes index other unobserved characteristics. Fourth, we also can expect stranger-interviewer effects to vary across time in any given setting because one of the most important structural determinants of openness to strangers in general may be the level of social density and diversity (Douglas 1970; Durkheim [1926] 1964), which is not fixed at either the general macro level of community norms or the individual level. Indeed, we can expect increasing interactions with strangers in general and increasing familiarity with stranger-interviewers and their scripts in particular to reduce the scope of stranger-interviewer effects, perhaps even reversing them in those conversational domains in which people become accustomed to interacting with other types of professional strangers. On the other hand, the increasing realization within the response effects literature that interviewer–respondent interaction draws heavily on normal types of interactional rules (e.g., Maynard and Schaeffer 2002; Schaeffer and Presser 2003) suggests that intimate insiders, in this case female insiders, may always have a marginal advantage over strangers in certain conversational domains, no matter how skilled the latter. Finally, and this may be the most important point in terms of the practical implications of this research, failure to address the potential for insider–stranger effects has analytic implications where such effects are significant and unobserved. Besides the effects on response reliability described earlier, two such effects are likely. The first includes the standard effects of response bias: undermining of univariate statistics accuracy (and hence comparability) and distorting of analytic relationships (e.g., Blalock

#2989-ASR 71:6 filename:71607-weinreb—p.1034 1034 —–AMERICAN SOCIOLOGICAL REVIEW

1961; de Leeuw, Mellenbergh, and Hox 1996). Indeed, the potential for such implications is perhaps all the greater in this case because strangerinterviewer effects appear to offer particularly fertile ground for response bias. In these data, for example, the respondents who emerge from interviews with female insider-interviewers appear different from those who emerge from interviews with female stranger-interviewers. The former have fewer traditional gender attitudes and more sources of income, but their responses to AIDS-related questions are less likely to match the messages of public health campaigns. In contrast, the latter (women interviewed by strangers) appear more downtrodden (by their husbands) and poorer, but more in tune with AIDS-related messages. The difference between these two indicates two different types of populations, each of which blends modern and traditional characteristics in a very different way, and each of which, in preliminary analysis using K2 data only, appears to generate different multivariate estimates (Weinreb 2000). However, and this is the important point, each of these populations appears to be a simple product of different types of interviewer stimuli. This in turn points to the most insidious effects of insider–stranger biases. On the structural level, foundational social theory implies that tolerance for stranger-interviewers is related to tolerance for other types of behavioral innovations from the outside. This makes the choice of insider- or stranger-interviewers endogenous to subsequent analyses of behavioral innovation. Given that the identification of such causal mechanisms is the explicit aim of much social science research, this is a

potentially serious analytic problem. At the very least, it suggests that researchers need to incorporate ways to measure the extent of the potential insider–stranger bias in their research design, especially where they are conducting research among populations that are structurally more susceptible to such effects. Ultimately, it is hoped that careful comparative studies will allow researchers to identify such settings empirically, and to identify how shifting structural boundaries, such as fluctuations in the levels of social density and diversity, affect the generation of these biases. Until then, our sole guide is theory and, perhaps less reliable, variable researcher instincts and experience. This article suggests that these two sources lend themselves to somewhat greater skepticism about the efficacy of stranger-interviewers than has been acknowledged to date, certainly by practitioners of survey research. But it also suggests that just as there are limits to what we can expect of stranger-interviewers, there also are, at least in some settings, alternatives. Alexander A.Weinreb is a lecturer in the Department of Sociology and Anthropology, Hebrew University. A social demographer by training (Ph.D. from the University of Pennsylvania; doctoral committee constituted by Susan Watkins, Herbert Smith, and Randall Collins), his primary substantive research focuses on family and community networks and AIDS, and his primary geographic focus is East Africa. Over several years, Weinreb was Field Director of two NIHfunded longitudinal research projects in Kenya and Malawi. He currently teaches courses on the sociology of Africa, the sociology and demography of AIDS, data collection, and microsociological theory.

.47 .39 .30

.46 .44 .60 .88 .62

.34 .26 .48

.48 .40 .27

.44 .40 .60 .86 .64

.40 .34 .39

Female acquaintance

Note: Significantly different from predicted value for “female stranger” at: * p = .10; ** p = .05; *** p = .01.

Source of Income in the Last Month —Respondent claims that she: ——01. Sold things grown on her shamba (agricultural land) ——02. Received money from a small business ——03. Received money from a credit merry-go-round Gender Attitudes —Agrees that a husband has the right to leave his wife if: ——04. She disrespects his family ——05. She neglects household chores ——06. She fights him frequently ——07. She is sexually unfaithful ——08. She drinks too much AIDS —Respondent claims to: ——09. Talk to her husband about AIDS frequently ——10. Feel comfortable suggesting the use of condoms to her husband ——11. Be at moderate or great risk of catching AIDS

Model # and Dependent Variable

(Reference) Female stranger

.19* .11** .66*

.29 .24 .31** .53* .38

.60* .64*** .52

Female insider

.34 .31 .33

.50 .38 .52 .90 .68

.44 .42 .20

Male stranger

Type of Interviewer

Table A1.—Predicted Value of Selected Dependent Variables by Type of Interviewer, Net of Effect of Other Interviewer Characteristics

APPENDIX

.41 .27 .25**

.52 .46 .58 .87 .64

.45 .38 .25

Male acquaintance

.56** .31 .44

.62** .27* .60 .86 .62

.36** .32 .21

Male insider

#2989-ASR 71:6 filename:71607-weinreb—p.1035 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA —– 1035

#2989-ASR 71:6 filename:71607-weinreb—p.1036 1036 —–AMERICAN SOCIOLOGICAL REVIEW

REFERENCES Agar, Michael. 1980. The Professional Stranger: An Informal Introduction to Ethnography. New York: Academic Press. Anderson, Myrdene. 1986. “Cultural Concatenation of Deceit and Secrecy.” Pp. 323–8 in Deception: Perspectives on Human and Nonhuman Deceit, edited by Robert W. Mitchell and Nicholas S. Thompson. New York: State University of New York Press. Aquilino, William S. 1994. “Interview Mode Effects in Surveys of Drug and Alcohol Use.” Public Opinion Quarterly 58:210–40. Axinn, William G. 1991. “The Influence of Interviewer Sex on Responses to Sensitive Questions in Nepal.” Social Science Research 20:303–18. Axinn, William G., Thomas E. Fricke, and Arland Thornton. 1991. “The Microdemographic Community-Study Approach: Improving Survey Data by Integrating the Ethnographic Method.” Sociological Methods and Research 20:187–217. Back, Kurt W. and J. Mayone Stycos. 1959. The Survey Under Unusual Conditions: Methodological Facets of the Jamaica Human Fertility Investigation. Ithaca, NY: Society for Applied Anthropology, Cornell University. Bailar, Barbara A. 1976. “Some Sources of Error and Their Effect on Census Statistics.” Demography 13(2):273-85. Barley, Nigel. 1983. The Innocent Anthropologist: Notes From a Mud Hut. London, England: Penguin Books. Barnes, John A. 1994. A Pack of Lies: Towards a Sociology of Lying. Cambridge, England: Cambridge University Press. Beatty, Paul. 1995. “Understanding the Standardized/Non-Standardized Interviewing Controversy.” Journal of Official Statistics 11:147–60. Becker, Stan, Kale Feyistan, and Paulina MakinwaAdebusoye. 1995. “The Effect of Sex of Interviewers on the Quality of Data in a Nigerian Family Planning Questionnaire.” Studies in Family Planning 26:233–40. Belcher, D. W., F. K. Wurapa, I. M. Lourie, K. Kwabia, and S. K. Avle. 1976. Experience in Selecting, Training and Supervising Interviewers in a Rural Health Project: Danfa Project, Ghana, Danfa Project Monograph Series, No. 11. Los Angeles, CA: UCLA School of Public Health. Berk, Marc L. and Amy B. Bernstein. 1988. “Interviewer Characteristics and Performance on a Complex Health Survey.” Social Science Research 17:239–51. Bignami-van Assche, Simona, Georges Reniers, and Alexander A. Weinreb. 2003. “An Assessment of the KDICP and MDICP Data Quality: Interviewer

Effects, Question Reliability and Sample Attrition.” Demographic Research SC1:31–75. Blalock, Hubert M. 1961. Causal Inferences in Nonexperimental Research. Chapel Hill: University of North Carolina Press. Blanc, Ann K. and Trevor N. Croft. 1992. “The Effect of the Sex of Interviewer on Responses in Fertility Surveys: The Case of Ghana.” Presented at the annual meeting of the Population Association of America, Apri1 30–May 1, Denver, CO. Breault, K.D., Chris Barker and Russell Lemle. 1987. “Informal Helping in Partner and Stranger Dyads.” Journal of Marriage and the Family 49(3):541-47. Bulmer, Martin. 1993. “Interviewing and Field Organization.” Pp. 205–17 in Social Research in Developing Countries: Surveys and Censuses in the Third World. 2d ed., edited by Martin Bulmer and Donald P. Warwick. London, England: UCL Press. Buve, A., M. Carael, R.J. Hayes, et al. 2001a. “Multicentre Study on Factors Determining Differences in Rate of Spread of HIV in subSaharan Africa: Methods and Prevalence of HIV Infection.” AIDS 15, (Suppl 4): S5–S14. Cannell, Charles F., Peter V. Miller, and Lois Oksenberg. 1981. “Research on Interviewing Techniques.” Sociological Methodology 12:389–437. Catania, Joseph A., Diane Binson, Jesse Canchola, Lance M. Pollack, Walter Hauck, and Thomas J. Coates. 1996. “Effects of Interviewer Gender Interviewer Choice, and Item Wording on Responses to Questions Concerning Sexual Behavior.” Public Opinion Quarterly 60:345–75. Choldin, Harvey M., A. Majeed Kahn, and B. Hosne Ara. 1967. “Cultural Complications in Fertility Interviewing.” Demography 4:244–52. Christenfeld, Nicholas, William Gerin, Wolfgang Linden, Mara Sanders, Jennifer Mathur, James D. Deich, and Thomas G. Pickering. 1997. “Social Support Effects on Cardiovascular Reactivity: Is a Stranger as Effective as a Friend?” Psychosomatic Medicine 59:388–98. Cicourel, Aaron V. 1974. Theory and Method in a Study of Argentine Fertility. New York: John Wiley and Sons. Collins, Randall. 2004. Interaction Ritual Chains. Princeton, NJ: Princeton University Press. de Leeuw, Edith, Gideon J. Mellenbergh, and Joop J. Hox. 1996. “The Influence of Data Collection Method on Structural Models: A Comparison of a Mail, a Telephone, and a Face-to-Face Survey.” Sociological Methods and Research 24:443–72. Demographic and Health Surveys (DHS). 1987. Interviewer’s Manual: For Use With Model “A” Questionnaire For High Contraceptive Prevalence Countries (Basic Documentation–5). Columbia, MD: Institute for Resource Development. Dentan, Robert Knox. 1968. The Semai: A Nonviolent

#2989-ASR 71:6 filename:71607-weinreb—p.1037 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA—– 1037

People of Malaya. New York: Holt, Rinehart, and Winston. Dijkstra, Wil. 1987. “Interviewing Style and Respondent Behavior: An Experimental Study of the Survey Interviewer.” Sociological Methods and Research 16:309–34. Dindia, Kathryn, Mary Anne Fitzpatrick, and David A. Kenny. 1997. “Self-Disclosure in Spouse and Stranger Interaction: A Social Relations Analysis.” Human Communications Research 23:388–412. Doctor, Henry V. and A. A. Weinreb. 2005. “Mortality Among Married Men in Rural Kenya and Malawi.” Etude de la Population Africaine 20:165–77. Douglas, Mary. 1970. Natural Symbols: Explorations in Cosmology. New York: Pantheon Books. Durkheim, Emile. [1926] 1964. The Division of Labor in Society. New York: The Free Press. Fowler, Floyd J. and Thomas W. Mangione. 1990. Standardized Survey Interviewing: Minimizing Interviewer-Related Error. Newbury Park, CA: Sage Publications. Garf inkel, Harold. 1967. Studies in Ethnomethodology. Englewood Cliffs, NJ: Prentice-Hall. Giddens, Anthony. 1990. The Consequences of Modernity. Stanford, CA: Stanford University Press. Givens, David. 1978. “Greeting a Stranger: Some Commonly Used Nonverbal Signals of Aversiveness.” Semiotica 22:351–67. Goffman, Erving. 1971. Relations in Public: Microstudies of the Public Order. New York: Basic Books. Gwassi. 1994. Unpublished KDICP field notes by Susan Watkins (KDICPJournal-HH1-A-Dec1994.pdf). Available at http://kenya.pop.upenn.edu/ level%203/Kenya/level3_kenya_fieldnotes. Gwassi. 1995. Unpublished KDICP field notes by Susan Watkins (KDICPJournal-HH1-B-Jan1995.pdf). Available at http://kenya.pop.upenn.edu/ level%203/Kenya/level3_kenya_fieldnotes. Gwassi. 2000. Unpublished KDICP field notes by Susan Watkins (KDICPJournal-HH3-K3.pdf). Available at http://kenya.pop.upenn.edu/ level%203/Kenya/level3_kenya_fieldnotes. Hagenaars, Jacques A. and Ton G. Heinen. 1982. “Effects of Role-Independent Interviewer Characteristics on Responses.” Pp. 91–130 in Response Behavior in the Survey-Interview, edited by W. Dijkstra and J. van der Zouwen. London, England: Academic Press. Heisel, Donald F. 1968. “Attitudes and Practice of Contraception in Kenya.” Demography 5:632–41. Hershfield, Allan F., Niels G. Rohling, Graham B. Kerr., and Gerald Hursch César. 1993. “Fieldwork in Rural Areas.” Pp. 241–52 in Social Research in Developing Countries: Surveys and Censuses in the Third World. 2d ed., edited by Martin Bulmer

and Donald P. Warwick. London, England: UCL Press. Holstein, James A. and Jaber F. Gubrium. 1995. The Active Interview. Thousand Oaks, CA: Sage Publications. Hyman, Herbert H., William J. Cobb, Jacob J. Feldman, Clyde W. Hart, and Charles Herbert Stember. 1954. Interviewing in Social Research. Chicago, IL: University of Chicago Press. Jacquette, J. 1990. “Gender and Justice in Economic Development.” Pp. 54-69 in Persistent Inequalities, edited by Irene Tinker. Oxford, England: Oxford University Press. Jansen, Sue Curry. 1980. “The Stranger as Seer or Voyeur: A Dilemma of the Peep-Show Theory of Knowledge.” Qualitative Sociology 2:22–55. Kawadhgone. 2000. Unpublished KDICP field notes by Susan Watkins (KDICPJournal-HH3-K3.pdf). Available at http://kenya.pop.upenn.edu/ level%203/Kenya/level3_kenya_fieldnotes. Kearl, B., ed. 1976. Field Data Collection in the Social Sciences: Experiences in Africa and the Middle East. New York: Agricultural Development Council, Inc. Kroeger, Axel. 1983. “Health Interview Surveys in Developing Countries: A Review of the Methods and Results.” International Journal of Epidemiology 12:465–81. Krysan, Maria, Howard Schuman, Lesli Jo Scott, and Paul Beatty. 1994. “Response Rates and Response Content in Mail Versus Face-to-Face Surveys.” Public Opinion Quarterly 58:381–99. Levy, Marion J., Jr. 1972. Modernization: Latecomers and Survivors. New York: Basic Books. Litwin, M. S. 2003. How to Assess and Interpret Survey Psychometrics: The Survey Kit. 2d ed. London, England: Sage Publications. Mangione, Thomas, Ralph Hingson, and Jane Barrett. 1982. “Collecting Sensitive Data: A Comparison of Three Survey Strategies.” Sociological Methods and Research 10:337–46. Mauldin, W. P. 1965. “Application of Survey Techniques to Fertility Studies.” Pp. 93–118 in Public Health and Population Change: Current Research Issues, edited by Mindel C. Sheps and Jeanne C. Ridley. Pittsburgh, PA: University of Pittsburgh Press. Maynard, Douglas W. and Nora Cate Schaeffer. 2002. “Standardization and Its Discontents.” Pp. 3–46 in Standardization and Tacit Knowledge: Interaction and Practice in the Survey Interview, edited by Douglas W. Maynard, Hanneke HoutkoopSteenstra, Nora Cate Schaeffer, and Johannes van der Zouwen. New York: John Wiley and Sons. Mayoni, Joseph R. 1983. “Eager Visitor, Reluctant Host: The Anthropologist as Stranger.” Anthropologica 25:221–49. Mensch, B. S., P. C. Hewett, and A. Erulkar. 2003. “The Reporting of Sensitive Behavior Among

#2989-ASR 71:6 filename:71607-weinreb—p.1038 1038—– AMERICAN SOCIOLOGICAL REVIEW

Adolescents: A Methodological Experiment in Kenya.” Demography 40: 247–68. Molm, L. D., N. Takahashi, and G. Peterson. 2000. “Risk and Trust in Social Exchange: An Experimental Test of a Classical Proposition.” American Journal of Sociology 105:1396–427. Morel-Seytoux, Sylvie. 2000. Review of Gender Issues in the USAID/Kenya Integrated Strategic Plan (ISP) 2001–2005. Washington, DC: Women in Development Technical Assistance Project. Oppenheim, A. N. 1992. Questionnaire Design, Interviewing and Attitude Measurement. London, England: Pinter Publishers. Oyugis. 1994. Unpublished KDICP field notes by Susan Watkins (KDICPJournal-HH1-A-Dec1994.pdf). Available at http://kenya.pop. upenn.edu/level%203/Kenya/level3_kenya_ fieldnotes. Oyugis. 2000. Unpublished KDICP field notes by Susan Watkins (KDICPJournal-HH3-K3.pdf). Available at http://kenya.pop.upenn.edu/ level%203/Kenya/level3_kenya_fieldnotes. Plummer, Mary L., Daniel Wight, David A. Ross, Rebecca Balira, Alessandra Anemona, Jim Todd, Zachayo Salamba, Angela I. N. Obasi, Heiner Grosskurth, John Changalunga and Richard J. Hayes. 2004. “Asking Semi-Literate Adolescents About Sexual Behaviour: The Validity of Assisted Self-Completion Questionnaire (ASCQ) Data in Rural Tanzania.” Tropical Medicine and International Health 9:737–54. Ross, David A. and J. Patrick Vaughan. 1986. “Health Interview Surveys in Developing Countries: A Methodological Review.” Studies in Family Planning 17:78–94. Rudolph, L. and S. H. Rudolph. 1958. “Surveys in India: Field Experience in Madras State.” Public Opinion Quarterly 33:235–44. Schaeffer, Nora C. 1980. “Evaluating Race-ofInterviewer Effects in a National Survey.” Sociological Methods and Research 8:400–19. ———. 1991. “Conversation With a Purpose—or Conversation? Interaction in the Standardized Interview.” Pp. 367–91 in Measurement Errors in Surveys, edited by Paul B. Biemer, Robert M. Groves, Lars E. Lyberg, Nancy A. Mathiowetz, Seymour Sudman, and Paul B. Biemer. New York: Wiley and Sons. Schaeffer, Nora C. and Stanley Presser. 2003. “The Science of Asking Questions.” Annual Review of Sociology 29:65–88. Schober, Michael and Frederick G. Conrad. 1997. “Does Conversational Interviewing Reduce Survey Measurement Error.” Public Opinion Quarterly 61:576–602. Simmel, Georg. 1950a. “The Stranger.” Pp. 402–8 in The Sociology of Georg Simmel, translated and edited by Kurt H. Wolf. New York: Free Press. ———. 1950b. “Knowledge, Truth, and Falsehood

in Human Relations.” Pp. 307–16 in The Sociology of Georg Simmel, translated and edited by Kurt H. Wolf. New York: Free Press. Singer, Eleanor, Dawn R. Von Thurn, and Esther R. Miller. 1995. “Confidentiality Assurances and Response: A Quantitative Review of the Experimental Literature.” Public Opinion Quarterly 59:66–77. Stone, Linda and J. Gabriel Campbell. 1984. “The Use and Misuse of Surveys in International Development: An Experiment from Nepal.” Human Organization 43:27–37. Stycos, J. M. 1960. “Sample Surveys for Social Science in Underdeveloped Areas.” Pp. 375–88 in Human Organization Research, edited by R. N. Adams and J. Preiss. Homewood, IL: The Dorsey Press. Suchman, Lucy and Brigitte Jordan. 1990. “Interactional Troubles in Face-to-Face Survey Interviews.” Journal of American Statistical Association 85:232–53. Sudman, Seymour and Norman M. Bradburn. 1974. Response Effects in Surveys: A Review and Synthesis. Chicago, IL: Aldine Publishing Company. Sudman, Seymour, Norman M. Bradburn, Ed Blair, and Carol Stocking. 1977. “Modest Expectations: The Effects of Interviewers’ Prior Expectations on Responses.” Sociological Methods and Research 6:171–82. Taietz, Philip. 1962. “Conflicting Group Norms and the “Third” Person in the Interview.” American Journal of Sociology 68:97–104. Tönnies, Ferdinand. 1961. Custom: An Essay on Social Codes. Chicago, IL: Henry Regnery Company. Tourangeau, Roger. 1990. “Comment on Suchman and Jordan.” Journal of American Statistical Association 85:250–51. Tourangeau, Roger, Lance J. Rips, and Kenneth Rasinski. 2000. The Psychology of Survey Response. Cambridge, England, Cambridge University Press. United Nations, Department of Economic and Social Affairs. 1961. The Mysore Population Study. New York: United Nations. van Tilberg, Theo. 1998. “Interviewer Effects in the Measurement of Personal Network Size: A Nonexperimental Study.” Sociological Methods and Research 26(3):300-328. Wakula South. 1994. Unpublished KDICP field notes by Susan Watkins (KDICPJournal-HH1-BJan1995.pdf). Available at http://kenya. pop.upenn.edu/level%203/Kenya/level3_kenya_ fieldnotes. Wakula South. 1996. Unpublished KDICP field notes by Susan Watkins (KJournal-HH2-Dec96.pdf) Available from author. Ware, Helen. 1977. Language Problems in

#2989-ASR 71:6 filename:71607-weinreb—p.1039 THE LIMITATIONS OF STRANGER-INTERVIEWERS IN RURAL KENYA —– 1039

Demographic Field Work in Africa: The Case of the Cameroon Fertility Survey. London, England: International Statistical Institute & World Fertility Survey, Scientific Report No. 2. Warren, Carol A. B. 1988. Gender Issues in Field Research. Qualitative Research Methods Series. Thousand Oaks, CA: Sage Publications. Watkins, Susan C. 2000. “Local and Foreign Models of Reproduction in Nyanza Province, Kenya, 1930–98.” Population and Development Review 26:725–59. Weinreb, Alexander A. 2000. Integrating Respondents, the Community, and the State in the Analysis of Contraceptive Use. Ph.D. dissertation, Population Studies Center, University of Pennsylvania, Philadelphia, PA.

———. 2001a. “A State-Centered, Political Capital Approach to the Explanation of Demographic Differences: With Special Reference to Regional/Ethnic Inequality in Sub-Saharan Africa.” In Proceedings of the General Conference of the International Union for the Scientific Study of Population (IUSSP), August 18–24, Salvador, Brazil. ———. 2001b. “First Politics, Then Culture: Accounting for Ethnic Differences in Demographic Behavior in Kenya.” Population and Development Review 27:437–67. Zeller, Richard A. and Edward G. Carmines. 1980. Measurement in the Social Sciences: The Link Between Theory and Data. New York: Cambridge University Press.