Sampling Weights Truncation in Household Sample Surveys Using Mean Square Error Procedure

Abstract

The sampling weights truncation aims at adjusting the weights applied to survey data to reduce variances. In most household-based surveys, the disproportional sample selection in stratified sampling using Probability Proportional to Size (PPS) introduce differing selection probabilities. The inverse of the such probabilities constitute sampling weights which are further adjusted for non-response to generate unbiased estimates. However, variations and adjustments may introduce extreme weights and increase in variances. Although sampling weight truncation minimize variances it also introduces bias on estimates. This study used a Mean Square Error (MSE) procedure to examine the effects of weights adjustments on eight selected variables from 2005-06 Kenya Integrated Household Budget Survey (KIHBS) in Nairobi district. The Taylor series linearization method of variance estimation was used to cater for complexity of the survey. Using 70th to 99th percentiles cut-off points of weight variable, the procedure identified the 85th percentile as the optimal level of truncation. Analysis showed a substantial reduction of variances with minimal biases in estimates.