Abstract

Among survey researchers, there is the hope that paradata or auxiliary variables describing interviewer observations and other measurements about the data collection process, would be suitable candidates for performing nonresponse adjustments to survey estimates. Such paradata are available for both respondents and nonrespondents, can be tailored to key survey outcome variables and capture information about the survey data collection process. Specifically, interviewer observations have been shown to have weak-to-moderate relationships with response propensity , establishment of contact, and key survey variables. This study aims to generalize the results of West (2010b) by embedding a randomized experiment in a new wave of data collection from a nationally representative area probability sample of households in Germany (the PASS survey). The experiment is designed to test whether providing interviewers with predictive information for variables that they are attempting to observe will improve the accuracy of their observations on two key auxiliary variables: household income, and whether anyone in a sampled household is currently receiving unemployment benefits. Specifically, a random subsample of addresses will have predictive information on income and receipt of unemployment benefits provided to interviewers on their address lists, and a random (control) subsample of addresses will not have this information provided. Given that these observations can be validated using respondent reports and administrative data (for unemployment benefit recipient cases), comparisons of the accuracy of interviewer observations in the two random subsamples of addresses will be performed, in addition to estimation of interviewer variance in both the accuracy of the observations and the effect of the predictive intervention on accuracy.