Assessing potential errors in level-of-effort paradata using GPS data

Surveys are a critical resource for social, economic, and health research. The ability to effciently collect these data and develop accurate post-survey adjustments depends upon reliable data about effort required to recruit sampled units. Level-of-effort paradata are data generated by interviewers during the process of collecting data in surveys. These data are often used as predictors in nonresponse adjustment models or to guide data collection efforts. However, recent research has found that these data may include measurement errors, which would lead to inaccurate decisions in the field or reduced effectiveness for adjustment purposes (Biemer, Chen, & Wang, 2013; West, 2013). In order to assess whether errors occur in level-of-effort paradata for in-person surveys, we introduce a new source of data - Global Positioning System (GPS) data generated by smartphones carried by interviewers. We examine the quality of the GPS data. We also link the GPS data with the interviewer-reported call records in order to identify potential errors in the call records. Specifically, we examine the question of whether there may be missing call records. Given the lack of a gold standard, we perform a sensitivity analysis under various assumptions to see how these would change our conclusions.