437 – Adaptive Survey Designs: Reflecting on the Past, Describing the Present, and Considering the Possibilities of the Future
Limiting the Risk of Nonresponse Bias by Using Regression Diagnostics as a Guide to Data Collection
James Wagner
University of Michigan
Survey data collections have traditionally been judged by the response rate. In terms of improving the response rate, every case has the same value. An optimal approach to maximizing the response rate would be to always attempt to interview the "easiest" remaining case. However, this strategy might lead to interviewing very similar cases. In this way, as a guiding indicator, the response rate might distort the survey data. The ultimate goal is to control nonresponse bias, but this bias is never observed in practice. Researchers are investigating proxy indicators that may be used to control the risk of nonresponse bias. Unfortunately, the field of survey methodology knows very little about how other indicators might perform as guides to data collection. One possible set of indicators are drawn from regression diagnostics. These diagnostics can be used to identify influential data points. Exploring the covariate space near these influential points may reduce our uncertainty about these regions. Under this approach, not all cases have the same value. We are currently running some exploratory experiments that prioritize currently active cases that are similar to observed cases that are