Using Propensity Score Weighting to Reduce Selection Bias in Large-Scale Data Sets

Abstract

Data sets from large-scale longitudinal surveys involving young children and families have become available for secondary analysis by researchers in a variety of fields. Researchers in early intervention have conducted secondary analyses of such data sets to explore relationships between nonmalleable and malleable factors and child outcomes, and to address issues of measurement. Survey data have been used to a lesser extent to examine plausible causal relationships between variables, perhaps due to the increased likelihood of selection bias that results with nonexperimental data. In this article, we use National Early Intervention Longitudinal Study data to demonstrate the use of inverse probability of treatment weighting, a quasi-experimental methodology based on propensity scores that can be used to reduce selection bias and examine plausible causal relationships. We discuss the advantages and disadvantages of this approach, and implications for its use in early intervention research.