Utilizing a Logistic Regression Approach for Weighting Adjustment in a Longitudinal Data Set

• Activity Number: 607
• Type: Topic Contributed
• Date/Time: Thursday, August 2, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Michael Sinclair*+ and Wan-Ying Chang and Julia Batishev and Michael Yang
• Companies: NORC and National Science Foundation and NORC and NORC
• Keywords: Propensity Models ; Nonresponse ; Paradata

Abstract

NORC under contract with the National Science Foundation conducted a research study to develop new methods of nonresponse weight adjustments for the 2008 integrated National and International Survey of Doctorate Recipients (SDR). Primarily this research was designed to build a logistic regression or propensity adjustment procedure and to contrast it to the current weighting class methodology. Given we expected a propensity based procedure would offer greater flexibility in the use of available covariates, we hoped the transition would enable us to leverage new information in the form of prior survey response patterns for the panel cases and level of effort or paradata to reduce the likelihood for nonresponse bias. This talk will provide an overview of the prior weighting methodology, the new methods explored and related considerations. Comparisons will be conducted by examining the model diagnostics and the distribution of the weight adjustments between a baseline model using the weighting class cells to the new models tested which utilize a combination of main and interaction effects from the weighting class factors with the new paradata and historical panel data.