Taking Selection Seriously in Correlational Studies of Child Development: A Call for Sensitivity Analyses

Correlational studies have played a major role in building our cumulative knowledge on child development. Yet as a result, we often have difficulty making causal inferences. The concern is selection effects: When children have not been randomly assigned to conditions, pre-existing biological, psychological, or social factors may bias correlations. In this article, we draw attention to sensitivity analyses, statistical techniques for estimating the robustness or fragility of results in light of potential selection effects. We highlight the coefficient of proportionality method recently developed by Oster (2019), which does not require assumptions about the number of omitted selection variables. The coefficient of proportionality provides an indication of how large the impact of unobserved selection factors would need to be-relative to observed covariates-to nullify a result. We offer two empirical examples to demonstrate the value of this method compared with other approaches used by child development researchers.