Low-Discrepancy, Pseudorandom Experimental Designs

Abstract

We will improve the design of randomized experiments and randomized controlled trials to account for knowledge of subject covariates. The proposed designs will result in more accurate assessment of experimental eects when they are predicted well by the subject covariates, without much loss of accuracy when the eects are unrelated to those covariates.New algorithms will be developed to enable experimenters to sample from thesedesigns, and analyses will be conducted to help experimenters compute the accuracy of the resulting estimators. Our approach to experimental design builds on recent developments in algorithmic discrepancy theory to accommodate moderate and high dimensional covariate information.

Document Details

Document Type
DoD Grant Award
Publication Date
May 08, 2020
Source ID
N000142012335

Entities

People

  • Dan Spielman

Organizations

  • Office of Naval Research
  • United States Navy
  • Yale University

Tags

Readers

  • Distributed Systems and Data Platform Development
  • Regression Analysis.