dc.description.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. | en |