Abstract:Many papers on high-dimensional statistics have proposed methods for variable selection and inference in linear regression models by relying explicitly or implicitly on the assumption that all regressors are exogenous. However, applications abound where endogeneity arises from selection biases, omitted variables, measurement errors, unmeasured confounding and many other challenges common to data collection Fan et al. (2014). The most common cure to endogeneity issues consists in resorting to instrumental variable (IV) inference. The objective of this paper is to present a Bayesian approach to tackling endogeneity in high-dimensional linear IV models. Using a working quasi-likelihood combined with an appropriate sparsity inducing spike-and-slab prior distribution, we develop a semi-parametric method for variable selection in high-dimensional linear models with endogeneous regressors within a quasi-Bayesian framework. We derive some conditions under which the quasi-posterior distribution is well defined and puts most of its probability mass around the true value of the parameter as $p \rightarrow \infty$. We demonstrate through empirical work the fine performance of the proposed approach relative to some other alternatives. We also include include an empirical application that assesses the return on education by revisiting the work of Angrist and Keueger (1991).

Feature Screening in High Dimensional Regression with Endogenous Covariates

Faithful Variable Screening for High-Dimensional Convex Regression

Feature Screening for Nonparametric and Semiparametric Models with Ultrahigh-Dimensional Covariates

Feature Screening for High-Dimensional Variable Selection in Generalized Linear Models

Variable screening in multivariate linear regression with high-dimensional covariates

Conditional Feature Screening for Mean and Variance Functions in Models with Multiple-Index Structure

Feature Screening for Multi-Response Varying Coefficient Models with Ultrahigh Dimensional Predictors

Ultrahigh dimensional variable selection: beyond the linear model

Feature Screening in Ultrahigh-Dimensional Partially Linear Models with Missing Responses at Random

Model-Free Feature Screening for Ultrahigh Dimensional Discriminant Analysis

Efficient Feature Screening for Ultrahigh-Dimensional Varying Coefficient Models

Robust Sure Independence Screening for Ultrahigh Dimensional Models

Adaptive Model-Free Sure Independence Screening

Screen then select: a strategy for correlated predictors in high-dimensional quantile regression

Strong Sure Screening of Ultra-high Dimensional Categorical Data

Bayesian variable selection in linear regression models with instrumental variables

Independent Feature Screening for Ultrahigh-Dimensional Models with Interactions

Feature Screening for Ultrahigh-dimensional Censored Data with Varying Coefficient Single-index Model

Sure Independence Screening Adjusted for Confounding Covariates with Ultrahigh-dimensional Data

Robust Variable Screening for Ultrahigh Dimensional

Feature Screening and Error Variance Estimation for Ultrahigh-Dimensional Linear Model with Measurement Errors