“Indexes and multiple hypothesis testing”

Abstract

Multiple hypothesis testing issues have appeared in the economics field over the last decade, providing a broad palette of methods designed to address this problem. The social sciences, in general, have greatly benefited from these advancements. Issues related to testing multiple hypotheses with a single treatment variable have been growing in economics over the last decades. However, other methods for constructing indices were in use before Anderson’s. Until today, researchers have not provided, or at least discussed, a structured set of ground rules to properly use these methods. In this thesis, we generate a statistical framework, primarily in the context of program evaluation, to assess the performance of different indexing techniques currently employed in the social sciences literature. Specifically, we evaluate the index proposed by Anderson (2008), the index produced from Principal Component Analysis and finally, a simple sum of the standardized variables index. We find that the way such indices are generated can lead to important differences in decisions related to rejecting or not rejecting null hypotheses of the significance of grouped outcomes