| This dissertation consists of three chapters that deal with semiparametric estimation problems in economics. Chapter I presents a random effects panel data model from which inferences on earnings mobility are made using longitudinal data from the Panel Study of Income Dynamics. The model has two error components: an individual effect that does not vary over time and represents unmeasured characteristics of individuals, and a serially correlated transitory effect. The model predicts the probability distribution of an individual's future earnings conditional on the individual's current earnings and observed characteristics. The distributions of the error components are estimated nonparametrically using methods based on deconvolution. This approach contrasts with previous research, which has assumed that the error components belong to parametric families of distributions, usually the normal. The nonparametric estimate of the distribution of the transitory error component has tails that are much heavier than those of the normal distribution. As a consequence, the estimated probabilities of transitions from low to high-earnings states are much lower with the nonparametric estimates than they are when the error components are assumed to be normally distributed. The results of numerical experiments indicate that this result reflects real properties of earnings paths and is not an artifact of the estimation procedure.; The purpose to chapter II is to develop a data based selection rule for optimally choosing the smoothing parameters for a deconvolution estimator similar to that used in chapter I. The rule chooses the parameters to minimize an estimated version of a simple analog to integrated mean squared error.; Chapter III examines the structure of experience and education in the earnings model and discusses the consequences of poorly specifying these variables in measuring the returns to education and analyzing the black/white wage gap. The evidence suggests that the Mincer specification provides a sufficient control in analyzing the wage gap. In measuring the returns to education, however, the Mincer formulation is unable to successfully model an interesting feature of the data discovered with a nonparametric estimator---that returns to a college education are interacted with experience. More general models, however, successfully predict this feature. |
