Computer-Assisted Personal Interviewing

Computer-assisted personal interviewing (CAPI) and its cousin, computer-assisted telephone interviewing (CATI), dominate survey data collection (see COMPUTER-ASSISTED DATA COLLECTION). There are a large number of software packages available and more coming online. In this entry, we will not attempt to evaluate the alternatives but instead discuss the major technical problems any CAPI system needs to solve [Page 160]and provide the outline of a solution. The advantages of computer-assisted interviewing are so substantial—such as efficient question filters, range checks, and skip pattern checks—that the only candidates for not using computer-assisted techniques are very short, simple questionnaires or surveys that involve few respondents or can use low-wage labor to process data and cannot afford the hardware.

Mounting a major survey requires the coordination of many tasks and details that touch data collection. Devising a CAPI software system to handle a large, complex survey is difficult. Unless the CAPI system integrates well with the other aspects of the survey process, the benefits of computerized data collection can be not only offset but also outweighed by the difficulties created by a poorly integrated effort. CAPI initially promised more accurate data collection through built-in checks that prevent or correct errors and inconsistencies. It also promised faster data delivery times from the end of the field effort to the release of the data. CAPI has met these promises. On the other hand, organizations that paid for surveys hoped that CAPI would reduce their costs. In general, this has not happened. We believe that problems with cost reflect a failure to integrate CAPI software with the rest of the survey process.

In mounting a major CAPI effort, there are six closely linked tasks. First, we must design the interview to meet the needs of the project. Second, we must prepare the software that puts the question and, where appropriate, the set of allowable answers on the computer screen. Third, we prepare any input data that drives the interview to the software. The input data can range from the rudimentary—with little or no data about the respondent being fed into the survey—to large and complex data for longitudinal surveys that contain hundreds or even thousands of data points from previous rounds of the study (which can be referenced during the interview). For example, when collecting event histories, the names and demographic characteristics of household members, names of employers, and dates and types of transitions figure prominently in the input file to a longitudinal survey. Input data can also include sound files for applications in which the computer “reads” the question text to the respondent, a technique used for sensitive questions about sexual practices or substance use when directly reporting to the interviewer might bias the response. The fourth task is to train interviewers to use the software and send them out to collect or “capture” the data. The fifth task is to move the data from the interviewer's portable PC to a central data repository; the sixth task is to prepare data files for analysis from the data in that central repository.

...