Seasonal Adjustment of CPS Time Series with Large Survey Errors

• Activity Number: 407
• Type: Contributed
• Date/Time: Thursday, August 15, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Business & Economics Statistics Section*
• Author(s): Richard Tiller*+
• Affiliation(s): U.S. Bureau of Labor Statistics
• Keywords: periodic surveys ; small area estimation ; state-space model

Abstract

BLS directly seasonally adjusts 182 time series generated from the Current Population Survey (CPS). While the CPS produces highly reliable estimates for national aggregates, many of the detailed demographic series are based on small samples. When substantial survey error (SE) exists in a CPS series, conventional methods of seasonal adjustment produce a trend, seasonal, and irregular decomposition that is very different from the classical decomposition. Much of the correlated survey error is absorbed into the trend, which produces spurious long-run fluctuations. SE also tends to cause seasonal patterns to look less stable than they really are. We consider a signal-plus-noise model that combines a model of the population values and its unobserved components (trend, seasonal, and irregular) with a model of the survey errors. SE is treated as an additional unobserved component of the time series, with the special advantage that its variance-covariance structure is objectively identified by design information. We compare our estimates with official estimates for employed and unemployed black male teenagers and female teenagers employed in agriculture.