Imputing for Late Reporting in the U.S. Current Employment Statistics Survey

Surveys of economic conditions are often published monthly to provide up-to-date measures of the state of a country’s economy. In establishment surveys, some sample units may not report in time to be included in the current month’s estimates, but eventually do report data. This late reporting can lead to revisions of estimates as more sample data become available. To maintain credibility, it is important that the size of revisions be kept as small as possible. We study this issue using the U.S. Current Employment Statistics (CES) survey. A model-based view of the CES weighted link relative estimator is used to identify potential bias due to model misspecification. An alternative approach, involving imputation for missing data, is used in an attempt to reduce the magnitude of revisions between preliminary and final estimates of employment for a month. The alternative, while not yielding statistically significant improvement in monthly revisions at the industry level, offers the potential for improved estimates for lower level aggregation.