The 2013 Wave of the Panel Study of Income Dynamics (PSID) included both centralized (Lab) and decentralized (Field) telephone interviewing. Although there were no systematic differences in the sample assigned to each group, the average length of the interviews conducted centrally was significantly, unexpectedly longer than those conducted by Field interviewers from their home offices. On a study where respondent burden can impact current and future wave participation (and on-going participation of all family members), this was of great concern. However, it was unclear how interview length might correlate with interview quality and even whether the longer timings of Lab interviews was a sign of higher or lower quality or, perhaps, of differences in respondent characteristics.

Production managers used a variety of paradata tools to investigate and interpret this phenomenon. A small percentage of both centralized and decentralized interviews was digitally recorded. The outcomes of Lab and Field interview evaluations were assessed. In addition, Blaise audit trail data were parsed into an online analytical processing (OLAP) cube and the resulting dataset was used to explore the relationship between various factors and interview length, focusing on differences in interviewer behavior across the Telephone Lab and Field. Use of the audit trail data allowed quantification of differences such as number of asked questions, number of backups, frequency of remark entry, item timings and item missing data rates.

This presentation will discuss the use of paradata (specifically audit trail data ) to explore an unanticipated production phenomenon. Examples provided will show how this analysis was used during production to redirect efforts, post-production to inform future project decisions, and in preparation for the 2015 data collection.