In an observational study or nonrandomized experiment, treated and control groups formed without random assignment are compared in an effort to estimate the effects of a treatment. Observational studies are common in most fields that study the effects of policies, programs, or hazards on people. In an observational study, treated and control groups may have differed prior to treatment in ways that are relevant for the outcomes under study, and these pretreatment differences may be mistaken for an effect of the treatment. If the groups differed in terms of relevant pretreatment variables or covariates that were observed and recorded, then there is an overt bias. If the groups differed in terms of covariates that were not recorded, then there is a hidden bias. Controlling overt biases and addressing hidden biases are central concerns in an observational study. The proposed work concerns improvements in both research design and analysis to reduce and address the biases that arise in observational studies. The work is of methodological relevance to the diverse fields that study the effects of interventions on human populations, including epidemiology, psychology, economics, and educational research.
