Dependent Interviewing

Dependent interviewing is a method of scripting computer-assisted survey questionnaires, in which information about each respondent known prior to the interview is used to determine question routing and wording. This method of personalizing questionnaires can be used to reduce respondent burden and measurement error. The prior information can be incorporated reactively, for in-interview edit checks, or proactively, to remind respondents of previous answers.

Dependent interviewing exploits the potential of scripting computer-assisted questionnaires such that each interview is automatically tailored to the respondent's situation. This can be done using routing instructions and text fills, such that both the selection of questions and their wording are adapted to the respondent's situation. Both routing and text fills are usually based on responses to earlier questions in the questionnaire. Dependent interviewing in addition draws on information known to the survey organization about the respondent prior to the interview. In panel surveys, where dependent interviewing is mainly used, this information stems from previous waves of data collections. For each panel wave, prior survey responses are exported and stored together with identifying information (such as name, address, and date of birth) used by interviewers to locate sample members eligible for the round of interviewing.

The previous information can be incorporated into the questionnaire script to reduce respondent burden and measurement error. In panel surveys, a set of core questions are repeated at every interview. For respondents whose situation has not changed between [Page 187]interviews, it can be frustrating and lengthen the interview unnecessarily to have to answer the same questions repeatedly. With dependent interviewing, information from previous waves can be used to verify whether a respondent's situation has changed. If not, and if the responses given in the previous interview still accurately reflect the respondent's situation, the questionnaire script can automatically route the respondent around unnecessary redundant questions. Responses from previous waves can then be filled in for the current wave. For open-ended questions such as those regarding occupation, this not only reduces the length of the interview, but also of coding time.

In general, the purpose of asking the same questions at different points in time is to generate data that can be used to investigate individual-level change. Estimates of change from panel surveys, however, tend to be biased. This is because responses about the reference period reported in one interview tend to be internally consistent but are not necessarily consistent with responses given in earlier interviews. These longitudinal inconsistencies can be due to respondent errors (such as simple variation in the way the respondent understands a question or describes her or his situation, recall errors, or estimation strategies used to compute responses), or interviewer errors, coding errors, or processing errors.

A consequence of these inconsistencies is the phenomenon called the “seam effect.” Dependent interviewing can be used to remind respondents of previous responses or for edit checks to verify whether apparent changes are true. The hope is that this will reduce response variance, improve respondent recall, and catch interviewer errors. Routing around redundant open-ended questions and imputing codes from previous waves further increases longitudinal consistency. Dependent interviewing has been shown to effectively reduce, although not completely eliminate, seam effects.

...