Comparison of Variance Estimation Methods for the National Compensation Survey

Steven P. Paben


A key use of the National Compensation Survey (NCS) is for the estimation of hourly mean wages for different localities. In NCS, hourly mean wages are calculated as the ratio of total annual pay to the total number of annual hours worked. In this study, an artificial Metropolitan Statistical Area (MSA) was created by combining NCS data from 16 different localities to serve as a sampling frame for 100 simulated samples. Then, the variance of the 100 sample estimates was compared to variance estimates obtained with the linearized Taylor Series method of variance estimation and three different methods of replication, balanced repeated replication (BRR), a Jackknife method, and Fay's method. To compare the variance estimates of the different methods a number of evaluative statistics were calculated, including bias, stability, and 95% confidence interval coverage.