Conferences

Nonresponse is a large and growing problem in surveys. As a result, nonresponse bias is a serious challenge to the accuracy and validity of survey research. One approach to nonresponse bias is to reduce it by reducing nonresponse. However, it has become increasingly difficult to minimize nonresponse and increasingly costly to do so. A second approach is to detect and then adjust for non-response bias. But this approach is handicapped by the problem that non-respondents do not participate in surveys and thus one gains no information about them from interviews. This naturally makes non-response bias hard to both detect and adjust for. Promising techniques to reduce this problem are to make more use of contextual and paradata. Contextual data are aggregate-level information that is known about the area or sampling stratum from which all cases, respondents and nonrespondents, are Dracn. Since this information known for all cases it can be used to measure and adjust for nonresponse bias. Paradata are information that is collected as part of the process of conducting a survey. As such, it is also known for all cases, respondents and nonrespondents. Paradata can consist of both information collected only as part of the normal administration of the survey (e.g. record of calls) and special observational data that interviewers record about all sampled cases, respondents and nonrespondents (e.g. condition of the dwelling, household characteristics). Since the contextual and paradata are known for all cases regardless of response status, they can be used to detect and adjust for non-response bias.