AN EMPIRICAL COMPARISON OF SOME METHODS IN THE STATISTICAL ANALYSIS OF DATA ON THE DURATION OF UNEMPLOYMENT AND EMPLOYMENT

The dissertation presents an empirical comparison and evaluation of some methods in the statistical analysis of data on the duration of unemployment, and to a lesser extent, employment. The data used in the study are continuous four year labor market histories of the male participants in the Denver Income Maintenance Experiment.;We review the relevant predictions of the search model and conclude that providing the economic environment is stationary, the search model predicts a rising hazard rate from unemployment.;We discuss the issues and methods involved in the continuous time analysis of duration data in the presence of unmeasured heterogeneity, and present some statistical models based on random effect and fixed effect specifications. The statistical issues involved in selecting the various samples used in estimating these models are considered in detail. We also present a rigorous statistical examination of a random effect model.;In the empirical results, we find evidence supporting a rising hazard rate from unemployment and a falling hazard from employment. We also find evidence supporting the specification of a gamma distribution for both employment and unemployment durations, and we find substantial differences in the point estimates and efficiency of our random effect and fixed effect models for unemployment duration.;Perhaps the most basic problems in the statistical analysis and interpretation of labor turnover data are due to the presence of unmeasured heterogeneity. Economic theory ofter makes predictions about the time dependence of the hazard rate from employment and unemployment; however, the presence of unmeasured heterogeneity tends to bias the slope of the estimated hazard, so that the empirically estimated hazard, when no heterogeneity correction is attempted, is unlikely to reflect that of the individuals in the sample and hence is not generally useful in evaluating economic theory.