Marginal Empirical Likelihood and Independence Feature Screening

• Activity Number: 347
• Type: Topic Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Chengyong Tang* and Yichao Wu and Jinyuan Chang
• Companies: Temple University and North Carolina State University and The University of Melbourne
• Keywords: Empirical likelihood ; high-dimensional data analysis ; nonparametric and semi- parametric models ; sure independence screening

Abstract

We consider independence feature screening technique for identifying contributing explanatory variables in broad high-dimensional nonparametric and semi-parametric regression analysis. Without requiring any parametric form of the underlying data model, our approach accommodates a wide spectrum of nonparametric and semiparametric model families. To surely capture the local contributions of explanatory variables, our approach constructs empirical likelihood in conjunction with marginal nonparametric regressions. Facilitated by empirical likelihood, our approach provides a unique perspective by directly assessing the strength of evidence from observed data for the local contribution of each explanatory variable. Since our approach actually requires no estimation, it is advantageous in scenarios such as the single-index models where even specification and identification of a marginal model is an issue. Theoretical analysis shows that the proposed feature screening approach can handle data dimensionality growing exponentially with the sample size. By extensive theoretical illustrations and empirical examples, we show that the local independence screening approach performs promisingly.