Generalized Variance Functions for the Current Employment Statistics Survey

1. INTRODUCTION The Bureau of Labor Statistics' (BLS) Current Employmem Statistics (CES) Survey gathers data monthly from over 380,000 nonagricultural business establishments allocated across all States, for the purpose of estimating total employment, employment of women and production workers, hours, and earnings. Estimates are made for over 1,500 industry cells, complementing the demographic detail provided by estimates of employment from the Current Population Survey (CPS). Monthly estimates of level and over-the-month change in employment are of primary importance to the users of these data. In addition to national estimates of level and change for these 1500 cells, state estimates for many of these cells are required. Unfortunately, many estimation cells within individual states, suffer sample sizes that are extremely small. It is not necessarily the case that variances of such cell estimates are always going to be large; there is a complex relationship between cell variances of employment estimates and cell characteristics other than sample size. This paper documents the derivation of these relationships and their application to variance estimation using generalized variance functions (GVFs). The estimate of over-the-month change for a basic estimating cell is the quotient of total employees for the current month in the matched sample (all sample establishments that reported employment for both current and previous month) over total employees for the previous month in the matched sample. This quotient is called the link (between current and previous month). The estimate of level for the current month is the product of this link and the estimate of level for the previous month. This estimate of level is called the hnk relative estimator. It is closely related to both the ratio and the regression estimator. looked at the general problem of estimating the variances of ratio and regression estimators. Their findings lead to some of