Interviewer-Related Error

Preventing Interviewer-Related Error

Prevention focuses on three basic strategies: (1) reducing or eliminating human intervention between respondent and data capture, (2) engineering error-proof questionnaires and associated data collection tools, and (3) standardizing interviewer behaviors to minimize error.

In their review of interviewer-related error, Robert M. Groves and his colleagues note that the very presence of an interviewer has been shown to bias responses. Employing computerized, scanned, or voice-response self-administration avoids both the costs and errors associated with employing human interviewers. Sloppy data resulting from respondent-related error, the bugaboo of self-administration, can be attacked through programming that mandates response and requires clarification of contradictory information.

This approach, however, has its own drawbacks. Notable among them are higher front-end costs and lead time, limitations on complexity, and lower response rates. Cybernetic approaches are expensive for all but the simplest of questionnaires; they require intensive programming, pretesting, and debugging to meet the requirement that they be far more error-proof than is required when a trained interviewer is involved. Even for simple questionnaires, the value added by the presence of an interviewer to motivate engagement and probe for focused and detailed answers compensates for and usually exceeds the error contributed. Complex enumerations and life history matrices are approached with trepidation in the absence of trained interviewers. Finally, it is far easier for respondents to opt out of higher-burden self-[Page 379]administered surveys than to avoid or disappoint a pleasant, yet determined and persistent interviewer.

Because both interviewer-administered and self-administered data collection have strengths and weaknesses, in surveys where the elimination of interviewers entirely is not prudent or possible, questions known to be affected by interviewer characteristics or limits to their capabilities can be switched to self-administration. Barbara Mensch and Denise Kandel found in their reanalysis of data from a longitudinal study that young respondents having the same interviewer over multiple data collection waves significantly underreported drug use. Their conclusion was that the development of “over-rapport” with the interviewer heightened self-censorship. Because that very rapport was responsible for the panel's extraordinary retention (ongoing response) rate, the solution was to maintain continuity of interviewer assignments but move the sensitive questions to self-administration.

...