Time Trend Analysis with Complex Survey Data

• Activity Number: 459 - Time-Trend Analysis with Complex Survey Data
• Type: Topic Contributed
• Date/Time: Wednesday, August 2, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Morgan Earp* and Daniell Toth and Polly Phipps and Charlotte Oslund
• Companies: U.S. Bureau of Labor Statistics and US Bureau of Labor Statistics and Bureau of Labor Statistics and Bureau of Labor Statistics
• Keywords: Bureau of Labor Statistics ; Job Openings and Labor Turnover Survey ; Regression Trees ; Response Rates ; Panel Survey ; Longitudinal Survey

Abstract

This paper examines nonresponse in a longitudinal establishment survey by using regression trees to model the effect of time in survey on data collection response propensities. We built regression trees using the R Package "Recursive Partitioning for Modeling Survey Data (RPMS)," which fits a linear model to the data conditionally on variables selected through recursive partitioning. The RPSM tree model provides estimates for the average response rate (across time), the intercept (initial response rate), and the slope (the effect of time in survey) for each end node. Using the RPSM package, we modeled the effect of time in survey on response rates in the Job Openings and Labor Turnover Survey (JOLTS) of the Bureau of Labor Statistics, so we could identify establishment characteristics and subgroups of establishments that were least likely to respond throughout the data collection process and thus determine which establishments required more follow-up over the course of the survey to ensure that they continue responding.