Computerized-Response Audience Polling (CRAP)

A number of survey designs deviate from the parameters of a scientific probability design, with significant consequences for how the results can be characterized. Computerized-response audience polling (CRAP) is an example of such a design. In this kind of poll, a sample of telephone numbers is typically purchased and loaded into a computer for automatic dialing. The questionnaire is produced through computer software that employs the digitized voice of someone assumed to be known to many of those who are sampled, such as the voice of a newscaster from a client television station. After an introduction, the computerized voice goes through the questionnaire one item at a time, and the respondent uses the key pad on a touchtone phone to enter responses to each question asked, as in an interactive voice response (IVR) system.

A major problem with CRAP polls is that the methodology does not allow for specific respondent selection, meaning that the basic premise of probability sampling, namely that each respondent has a known, nonzero probability of selection, is violated. Interviews are conducted with whoever answers the phone, and there is no guarantee that the person answering is eligible by age or other personal characteristics. Although information can be gathered about the household composition, there is no random selection of a designated respondent from the household. The computer can dial a large set of telephone numbers in a short period of time, working through a purchased sample quickly but producing a low contact or cooperation rate as a result. There also is no attempt to recontact a household to obtain an interview with a designated respondent who is not at home at the time of the first call. Because of these considerations, it is inappropriate to calculate a margin of error around any estimates produced from such a poll.

This method shares many characteristics with self-selected listener opinion polls (SLOP) and other designs that employ volunteer samples. A true response rate cannot be calculated, although a version of a cooperation rate can. The data can be collected rapidly and at low cost. Although post-stratification weighting can be applied to the resulting set of respondents, it is difficult to interpret its meaning when information about respondent selection is missing.