Inventing New Interviewing Techniques
Cannell's approach to the interview was rooted in his extensive observation of interviewers while leading field operations in the U.S. Department of Agriculture and at the University of Michigan Survey Research Center. He developed interviewing procedures designed to encourage respondents to stay on the path to accurate reporting. These procedures are based on a theory of the interview. In essence, Cannell argued that the interview involves not only the communication in questions and probes, but also various sorts of "meta-communication" - intentional and unintentional - that inform respondents about how they should undertake their tasks. Shaping the meta-communication to encourage accurate responding and to avoid inadvertent encouragement of error are the aims of interviewing procedures that Cannell and his colleagues developed. These techniques included ones designed to teach respondents what is needed to meet question demands (Instructions and Feedback) and one designed to motivate respondents to expend needed effort (Commitment).
Recognizing that the respondent role is unfamiliar to most people, Instructions were developed as a means to orient respondents to the tasks involved, both for the entire interview and for specific questions. For example, a global instruction could tell respondents that the interview will involve reporting information as accurately as possible. A question- specific instruction could emphasize the need to take time and search memory carefully before responding.
The Feedback technique accompanied Instructions. It was designed to inform respondents how well they were adhering to the expectations for accurate responding that had been conveyed in the Instructions. A respondent who took time prior to answering, for example, would be told that he or she had performed as desired. One who answered quickly, on the other hand, could be asked the question again, with an emphasis on more reflection. This contingent feedback gave respondents a way to judge how well they were doing as the interview progressed.
The Commitment procedure focused on the need to motivate people to take on and maintain the respondent role throughout the interview. The technique involves informing respondents at the beginning of the interview that their role requires effort and candor. They are then asked to commit themselves to responding in line with these requirements. In face-to-face interviews, respondents were asked to sign a statement to this effect. In telephone interviews, an oral promise was sought. Interviews were terminated if prospective respondents would not commit to answering accurately and honestly. The three techniques were scripted into questionnaires so that interviewers could apply them in a standard way. We review findings from studies examining the effect of the techniques on response accuracy in Section 2.2.
Creating Methods of Interview Observation
The theory of interviewing described above rested on novel empirical analyses of interview interactions. Cannell and his colleagues (most notably Floyd J. (Jack) Fowler and Kent Marquis, later Lois Oksenberg) set out to gather accurate descriptions of the interviewing process so as to inform both theory and practice with respect to the sources of error in the interview. Although behavior coding is often cited as initially developed for use in the evaluation of interviewers, the early studies suggest that its focus was primarily on understanding the interaction in the survey process.
Cannell and Fowler (1964) first uncovered evidence of interviewer effects on reporting of health experiences. They observed that reporting was affected by the size of the interviewers' workload and when during the field period they conducted interviews - later interviews elicited less information. In addition, they saw that the interviewers' pace of speech negatively affected reporting (see also Holbrook, et al., Chapter 17, and Olson and Smyth, Chapter 20, this volume). Such observations laid the foundation for later controlled experiments. Two subsequent studies (Cannell, Fowler, and Marquis 1968; Marquis and Cannell 1969) are foundational investigations into the survey interaction process in face-to-face data collection efforts. The first study used third-person observers and reinterviews with respondents a day later to gather their impressions of the interviewer and the interview; the second study used tape recordings of the interviews. In the first effort, despite hypothesizing that interviewers' and respondents' knowledge and attitudes would be highly correlated with the quality of data, the authors found that behavior in the interview was the main variable that correlated with the index of reporting quality. In addition, Cannell, Fowler, and Marquis (1968) found that the behaviors of the interviewer and the respondent were best described as "balanced." More verbal output by one actor was matched by output from the other. This relationship suggested that systematically changing the interviewer's speech output would change the respondent's activity level, thereby increasing both the amount and the quality of reported health information (a hypothesis studied later by Marquis, Cannell, and Laurent 1972).
Tape recording was employed in the second study (Marquis and Cannell 1969). Further, the coding scheme was expanded, including codes to capture information about probes (both the use and an evaluation of the quality of the probes) and the use of feedback by either the interviewer or the respondent. Three relevant findings from this study are: (1) a large proportion of the interviewer and respondent behavior is extraneous interaction, not helpful to the interview process; (2) interviewer feedback was one of the most frequent behaviors by the interviewers and was indiscriminate with respect to the quality of the response provided; (3) the coding scheme identified problems with questions (poor syntax, complex instructions, parenthetical phrases) and respondent difficulty in understanding or retrieving the information of interest. The findings from these studies led Cannell and colleagues to focus on taming communication in the interview in service of measurement goals. Over time, methods for recording and analyzing interview interaction (behavior coding) were applied to other measurement problems, including question pretesting and interviewer evaluation.