88 – Measuring Nonresponse and Other Nonsampling Errors in Surveys
The Empirical Properties of Survey Quality Indicators Under Nonignorable Nonresponse
Shin-Jung Lee
University of Michigan
Recent developments in survey quality indicators may provide promising alternatives to the response rate. Survey quality indicators evaluate the representativeness of respondents, which reflects the quality of the survey inferences in terms of bias. We investigate the performance of these indicators and their sampling properties under different response rates and non-response mechanisms through simulation experiments. NHANES data from 1999-2008 is used as the �nite population for the simulation experiments. The experiments are conducted under two assumptions. First, all co-variates associated with the true response probability are available and are included in the model. Second, auxiliary variables such as variables from a frame are known at the subject-level for both respondents and non-respondents. A total of 72 scenarios from different combinations of experimental parameters are explored. The results are derived from 10,000 replications for each scenario. We describe the �ndings for each indicator in terms of their ability to anticipate the level of bias that may arise in a speci�c survey analysis.