UCLA

This thesis contribute towards the understanding of labor economics and applied econometrics; the thesis is made up of three chapters.

The first chapter explores the causal effect of parents’ social capital on the intergenerational occupational inertia in addition to individuals’ labor market outcomes. A new data extract was constructed by re-weighting and combining the Panel Study of Income Dynamics (PSID) and the Survey of Income and Program Participation (SIPP) to correct the selection biases induced by children’s endogenous moving behaviors post-graduation. By exploiting the recent technological revolution and the resulting changes in occupational skill compositions measured by Dictionary of Titles (DOT) and its successor O*NET, it became possible to isolate the effect of inherited social capital from inherited human capital through a regression discontinuity design. Besides, a correction of the selection bias induced by the social capital advantage through children’s occupational switching patterns after the first jobs was made. The results indicate that around 30% of individuals choose the same occupation as their parents for their first job; such people rely more on their parents’ social connections in job hunting. Also, they enjoy a positive wage premium of about 5% of the percentile ranks of annual labor income for entry-level jobs but this positive effect fades away in the long-run.

The second chapter studies the estimation and inference of nonlinear econometric model when the economic variables are contained in different datasets. We show that the unknown structural parameters of interest can be possibly uniquely identified if there are some common conditioning variables in different datasets. The identification result is constructive, which enables us to estimate the unknown parameters based on a simple minimum distance (MD) estimator. We study the asymptotic properties of the MD estimator and provide inference procedure. A simple model specification test on the key identification conditions is also provided.

The third chapter provides an application example of the method developed in the second chapter. It is a long-standing problem in the empirical research that the economic variables are contained in different datasets. One well-accepted solution to this problem is the imputation method, which serves as a crucial step in the seminal work, Blundell, Pistaferri, and Preston (2008) studied the dynamic relationship between consumption and income, with consumption data from CEX and income data from PSID. In this chapter, we first prove that the imputation method is biased because they are significantly different from those based on true data, which is the newly available PSID from 1999 which includes both consumption and income data. Furthermore, we investigate the finite sample performance of our new method with this new PSID data and show that our method delivers comparable results with those based on the true data. We conclude that the imputation gives largely biased estimation compared to the real data results and the new estimator developed in Chapter 2 performs better.

The three chapters share the same interests in the long-lasting question that how we can deal with the situation in which the economic variables or the study population is contained in different datasets. The first chapter starts off from the simplest scenario that the data set

is complete in terms of variables but biased in terms of representativeness. The other two chapters deal with the other more difficult and more usual case that the data set is incomplete in terms of economic variables. We not only contribute methodologically by providing

a new estimator but also implement the method in an important application case and discuss the implications.