489 – New Modeling Approaches for Time Series Analysis
Methods for Producing Consistent Control Totals for Benchmarking in Survey Sampling
Ismael Flores Cervantes
Westat
Estimates from probability samples may not match known totals of populations due to sampling variation, undercoverage, or nonresponse. In this situation, special weighting adjustments such as raking or poststratification are implemented so the sums of weights match known totals. This process improves the face-validity of the survey because the total estimates reproduce the "true" or accepted total population generally produced by official statistical agencies. The totals, which are used as benchmarks in weighting, are sometimes generated combining different sources and in some situations they do not match the eligible population of the survey. In this paper we take a heuristic approach and propose methods for computing control totals that are consistent and represent the population of interest while maintaining the face validity of the estimates.