|
Abstract:
|
As survey response rates drop, researchers are seeking improved methods to predict respondents' propensity to respond and to guide an adaptive design effectively. The Census Bureau created the 2015 Planning Database (PDB) as a research tool that includes selected 2010 Census variables, ACS estimates, and a Low Response Score (LRS) based on response to the 2010 Census at the census block-group and the tract levels. MSG data is a proprietary database of household-level information. Westat selected two surveys that are multistage, area cluster samples of households, designed to produce samples representing the U.S. non-institutionalized population. In both surveys the second level of sampling is area segments comprising Census blocks so the PDB and MSG data are good candidates for modeling segment-level response rates at two stages. We explore whether the PDB or MSG provides useful yield estimates at the household or individual level. The actual results were compared to predictions from the PDB. In our examples, the LRS contained little predictive power. While these results may apply only to these two Westat surveys, these unexpected outcomes may be of interest to other surveys.
|
