Extended Bootstrap Bias Corrections with Application to Multilevel Modeling of Survey Data Under Informative Sampling

• Activity Number: 55
• Type: Topic Contributed
• Date/Time: Sunday, August 2, 2009 : 4:00 PM to 5:50 PM
• Sponsor: Social Statistics Section
• Author(s): Solange T. Correa*+
• Companies: Brazilian Institute of Geography and Statistics
• Keywords: Bias correction ; Bootstrap ; Multilevel models ; Probability weighting ; Informative sampling

Abstract

Data available to analysts are often obtained from complex sample surveys in which population units are selected by stratified multi-stage designs with unequal selection probabilities. Unweighted estimators of the model parameters may be severely biased in such cases if the selection probabilities are related to the outcome values even after conditioning on the model covariates (informative sampling). Probability weighting reduces the bias but does not eliminate it, unless the sample sizes at each level of the model hierarchy are very large. In this paper a general approach for bias correction based on bootstrap is proposed. The method is assessed by simulation study using probability weighted estimators of two-level model parameters when fitting survey data under informative sampling designs. The proposed method showed to be effective in bias reduction in all the cases considered.