Abstract:
|
Many surveys employ complicated sampling and estimation designs. Complex designs lead to complex variance estimation systems requiring significant investments of resources over long time periods to develop, implement, and run. The Census 2000 Accuracy and Coverage Evaluation (A.C.E.) Survey with a complex multi-stage design, imputation, and post-stratified dual-system estimation is an example. The A.C.E. design resulted in a complex variance estimation system using the stratified Jackknife approach. The development of this system involved external consultants, several years' effort for several persons, and an intense programming effort. A simplified variance system was also programmed. This system used the final weights, ignored the intermediate stages of sampling, ignored the variance of the imputations, and estimated a simple Jackknife variance. Total design and programming time could be measured in days for one person. Run time was also significantly reduced. The results of the simplified program for estimates of housing unit coverage are compared to those from the full variance estimation system.
|