Agnostic Hypothesis Tests in Political Science Research

• In Event: Causal Inference: Hypothesis Tests and Covariate Balancing

Abstract

Building on Esteves et al. (2016) and Coscrato et al. (2018), in this study we present the agnostic hypothesis test (AHT) as an alternative to the simultaneous hypothesis test commonly conducted in political science research. AHT has at least three advantages over simultaneous tests. First, it is the only test to satisfy four conditions of logical consistency: monotonicity, invertibility, union, and intersection consonances. Second, AHT also provides researchers with three choices instead of the two choices offered by traditional Null Hypothesis Significance Test: accepting the null hypothesis, rejecting the null hypothesis, and remaining agnostic with respect to the hypothesis. Third, AHT allows optimal control of type I and type II errors simultaneously. By discussing and presenting examples of AHT in political science, we show that AHT can improve the conduction of hypothesis test and statistical inference in political science research.

Authors

• Guilherme Jardim Duarte, Princeton University

• Thiago Nascimento da Silva, Universität Mannheim