Two-stage Semiparametric Estimators for Limited Dependent Variables and its Applications

• Main Author: Choi, Jin-Young
• Format: Others
• Language: English
• Published: Boston College 2014
• Subjects: limited dependent variable; panel; sample selection; semiparametrics; tiwns; US presidential election
• Online Access: http://hdl.handle.net/2345/bc-ir:103547

Thesis advisor: Arthur Lewbel === This thesis proposes two semiparametric estimators; one for heavily censored panel models and another one for binary-outcome sample selection models. The first chapter proposes a new panel data estimator, and applies it to investigate whether the key assumption underlying most twin studies is valid. Roughly, the assumption is that differences in twins' outcomes can on average be attributed to differences in observed treatments, possibly after conditioning on observable covariates. The empirical results here cast doubt on this assumption, by showing that a particular outcome, survival, varies by birth order, even after conditioning on health-at-birth characteristics. The proposed panel data estimator is the first one in the literature that simultaneously handles having an unknown error distribution, fixed effects, fixed T, fixed censoring point, and heavy (greater than 50%) censoring. These features are all required to adequately deal with the limitations of available census data on twins. The proposed estimator also allows for coefficients that vary by t, and for a censoring point that is an unknown but deterministic function of regressors. The second chapter proposes a new semiparametric estimator for binary-outcome selection models that does not impose any distributional assumption, nor specify the selection equation. The estimator, however, requires a special regressor satisfying a support restriction in the outcome equation and a variable satisfying the exclusion/inclusion restriction; the former should be continuous whereas the latter can be discrete. The estimators of Klein et al. (2011) and Escanciano et al. (2012) require optimization, but our estimator for the outcome equation has a closed-form expression with no need for any optimization (but the selection equation estimation may still need an optimization). We apply MLE and the proposed estimator to US presidential election data in 2008 and 2012 where Barack Obama won to see to what extent racism mattered; we use a prejudice variable as a measure of racism. Putting our empirical findings in advance, there is evidence that the white Democrats voted less for Obama due to prejudice, whereas the white Republicans acted in a more muted fashion (i.e., almost no change in voting due to racism) or voted more for Obama to escape the stigma of racism. We also found evidence of "own-race favor" by blacks. === Thesis (PhD) — Boston College, 2014. === Submitted to: Boston College. Graduate School of Arts and Sciences. === Discipline: Economics.