Nonresponse Analysis for School Surveys

• Activity Number: 250
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Ronaldo Iachan* and Maria Profiryakova and Kurt Peters
• Companies: ICF International and ICFI and ICFI
• Keywords: Nonresponse ; Adjustments ; recursive partitioning ; propensity models

Abstract

This research is focused on the selection of predictors for non-response analysis and weight adjustments for school survey data. Weighting classes for non-response adjustments are formed using a set of core variables that are correlated with response behavior and with survey outcomes, with the intent of minimizing the potential non-response bias (balanced against acceptable increases in weighting variance). The choice of variables to use when defining weighting classes is determined by measures of variable importance for predicting response. This study compares the use of response propensity models estimated using mixed effects logistic regression and a range of recursive partitioning methods for evaluating variable importance and the resulting bias reduction achieved through non-response weighting adjustments. Mixed effects models are run using SAS Proc Glimmix that allows modeling of random effects (e.g. school, class) and works best for modelling response indicator variables that follow exponential distributions.