Small-area Estimation with Autocorrelated Observations and Stochastic Benchmark Constraints

• Activity Number: 26
• Type: Contributed
• Date/Time: Sunday, August 8, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Richard B. Tiller*+ and Danny Pfeffermann
• Companies: Bureau of Labor Statistics and Hebrew University and University of Southampton

Abstract

In order to reduce the variances of State labor force estimates derived from the U.S. Current Population Survey (CPS), the Bureau of Labor Statistics (BLS) uses state-space models that are fitted to each of the direct CPS series, independently between the states. The models combine a model for the true population values with a model for the sampling errors. At the end of each calendar year, the model-dependent state estimates are benchmarked to the corresponding CPS annual average. This approach has the disadvantages of no real time benchmarking and instability in the benchmarking process because even the annual CPS averages are subject to relatively large sampling errors due to the high correlations between the monthly estimators. This paper investigates a new approach to benchmarking that constrains the separate monthly model-dependent state estimates in groups of states, (or nationwide) to sum to the corresponding aggregate CPS estimates in real time. The use of this approach requires joint modeling of the direct estimators in several states and adding the sampling errors and the benchmark constraints to the observation equations.