Xiaoxia Shi - University of Wisconsin - Madison

"Simple Inequalities Testing in the Presence of Linear Nuisance Parameters"

Abstract

In this paper we build on the conditional chi-squared test of Cox and Shi (2021, Restud, CS21) for subvector inference in moment inequality models. We significantly expand the scope of the test by removing a key restriction that CS21 imposes on the way the nuisance parameter enters the model. As a result, the test can be used for specification testing as well as for inference on a linear function of the model parameter in models that can be written as multiple inequalities/equalities restrictions. Such models include linear interval outcome regression, non-parametric instrumental variable regression with discrete variables and shape restrictions, inference for the optimal value of linear programs, as well as panel data multinomial choice models. In addition, we prove a new analytical result that greatly simply the calculation of the data-depend degree of freedom for the chi-squared critical value. This not only makes the test easier to program but also improves the speed of the test dozens to hundreds fold.

Additional information:

  • Speaker: Xiaoxia Shi
  • Time: Thursday, 19.05.2022, 16:00 - 17:00
  • Location: Online via Zoom
  • Further links:
  • Organizer: Statistics Group
  • Contact: