Semiparametric Limited Dependent Variable Estimators, with Applications

本研究的目的是为有限因变量模型，特别是二元选择、删失和截尾回归模型，构造和应用新的半参数和非参数估计。该项目在具有重要政策含义的经验应用中使用了新的估计器。支付意愿模型被设计用来估计一个人口愿意为公共产品支付的条件分布的均值、中位数和其他特征。删失回归估计量将被用来评估税收和收入政策对耐用品和其他消费类别的需求的影响，在这些类别中，观察到许多“零”。二元选择估计量将被应用于女性劳动力供给决策的建模，考虑到家庭资产收入等测量不佳的协变量。更具体地说，该项目开发了一个根n相容、渐近正态的二元选择模型估计量，它允许一些一般形式的未知异方差，以及有序选择、多项选择和删失回归模型的扩展。二元选择估计器特别适合于研究在评估公共产品时出现的“支付意愿”模型。研究人员继续他在这类应用上的工作，以及使用数据分类方法消除对第一阶段核密度估计的需要的估计器的变体。这一扩展允许在具有固定效应、弱外生回归和未知误差分布的二元选择面板数据模型中估计系数。还将调查估计器的扩展，以处理内生或错误测量的回归变量。这些和面板数据扩展将被应用于劳动力供给决策的实证研究。在此基础上，提出了截尾和截尾回归模型的一组新的非参数估计。在这些模型中，回归函数和误差分布都是非参数辨识和估计的。

The proposed research is to construct and empirically apply new semiparametric and nonparametric estimators for limited dependent variable models, especially binary choice, censored, and truncated regression models. The project employs the new estimators in empirical appli-cations having important policy implications. The willingness-to pay models are designed to estimate means, medians, and other features of the conditional distribution of a population's willingness to pay for public goods. The censored regression estimators will be used to assess the effects of tax and income policies on the demand for durables and other consumption categories in which many "zeros" are observed. The binary choice estimator will be applied to model female labor supply decisions, allowing for badly measured covariates such as household asset income.More specifically, the project develops a root n consistent, asymptotically normal binary choice model estimator that permits some general forms of unknown heteroscedasticity, along with extensions to ordered choice, multinomial choice, and censored re-gression models. The binary choice estimator is particularly well suited for studying the "Willingness-To-Pay" models that arise in valuing public goods. The investigator continues his work on this class of applications, along with variants of the estimator that use a data sorting method to eliminate the need for a first stage kernel density estimator.A panel data extension of the estimator is also developed. This extension permits estimation of coefficients in a binary choice panel data model with fixed effects, weakly exogenous regressors, and unknown error distribution. Extensions of the estimator to handle endogenous or mismeasured regres-sors will also be investigated. These and the panel data extension will be applied to an empirical study of labor supply decisions. Separate from the above, a new set of nonparametric estimators for censored and truncated regression models are proposed. In these models, both the la-tent regression function and the distribution of the errors are nonparametrically identified and estimated.