Evaluation Of Confidence Interval Methodology For The Occupational Compensation Survey Program

Christina L. Harpenau, Joan Coleman, and Mark Lincoln


Previous work by Dorfman and Valliant (1993) and Casady, Dorfman, and Wang (1994) on the Occupational Compensation Survey Program (OCSP) indicated that current confidence interval methodology tended to provide coverage which was less than the stated level for estimated mean wages. In the latter paper, alternative methodologies were developed to address this problem. A study was undertaken to evaluate these proposed methodologies. A population of establishments was created using available sample data from the OCSP. This population was constructed to have the properties of a "typical" MSA in terms of size, variety of occupations, size and number of establishments. This population served as the basis for a series of simulation studies to evaluate the current and alternate confidence interval methodologies. In addition, the study included the evaluation of alternative methods of collapsing strata for purposes of variance estimation. Based on the results of this study, procedures for estimating the precision of estimate mean wages in the OCSP will be revised as necessary.