Dealing with Overt and Hidden Biases Due to Selection: Evidence from an Intergenerational Longitudinal Study

• In Event: Using Quasi-Experimental Research Designs to Isolate Environmental Influences: Advancements in Life-Course Research

Abstract

In evaluation research, we are often interested in the effect of some treatment on an outcome of interest, such as crime or substance use. Unfortunately, many of the commonly considered ‘treatments’ in criminology, such as delinquent peer groups or abusive homes, do not lend themselves easily to social experiments out of either practical or ethical concerns. This limitation makes it highly important to find suitable ways to control for nonrandom selection into treatment to identify a treatment effect. The aim of this paper is to show how selection effects can be controlled for within observational data using the Rochester Intergenerational Study, the intergenerational component of the Rochester Youth Development Study. Specifically, these data feature a large number of observable covariates which can be eliminated as confounders using matching techniques including with propensity scores. We will also present statistics which may help quantify the potential limitations due to so-called hidden biases or unmeasured confounders.

Authors

• Emily L. Walter, University of Maryland

• Thomas Loughran, University of Maryland