Online Program

Model Based Macro-Editing Approach to State and Area Estimates from the Current Employment Statistics Survey

Keywords: statistical data editing, macro-editing, outlier, robust small area estimation

Estimates of employment from the Current Employment Statistics (CES) survey are published every month for a large number of cells defined at various detailed industrial and geographic levels. Before the estimates are released for publication, they need to be reviewed. The purpose of the review is to isolate cells that may contain erroneously reported and influential records not captured during the editing procedure. The traditional approach to the macro-editing is to compare the current estimates to the historical data and mark any significant deviation as suspicious. However, it may happen that estimates deviate from the historical records for legitimate reasons (for example, due to a changing economic pattern). We propose to use a model based approach leading to a more effective screening. The model considered in this paper is the area-level Fay-Herriot model, where we apply a robust method for estimating the model parameters. A standardized difference between the sample based estimate and the synthetic part of the model predictor is used as the basis for screening. While the general CES policy is to rely on the purely sample based estimates when the sample is moderately large, a possibly useful by-product of the proposed screening procedure is the set of the robust model-based estimates that can be used to replace the direct sample estimates in a limited number of extreme cases.