Comparative observational studies are used to estimate the effect of treatments, large-scale policy interventions and environmental exposures when controlled randomized studies are unethical, infeasible, or during the early stages of research for the purpose of generating hypotheses. The goal of this project is to produce user-friendly software that yields improved designs for concurrent-cohort observational studies that reduce bias due to measured covariates. The methods that will be implemented include multivariate matching and subclassification using propensity scores and the Mahalanobis metric. Tables, graphical displays, and diagnostic reports will be automatically created so users can quickly assess several designs to select the one most appropriate for their data. These displays will incorporate formulas predicting the optimal performance of the methods developed during the past decade. The design selection process does not utilize response variables, thereby avoiding the severe bias that can occur when an investigator evaluates several analyses but reports only those conforming with their prior beliefs. The software will support importing data from major software packages (e.g., SAS, SPSS), and the exporting of data and results to these same programs. This product will be a significant step towards the wider availability of the rapidly developing methods for causal inference based on observational comparisons.
This project will produce a user-friendly Windows-based software product for use in the design of observational evaluations of the effects of medical procedures, environmental exposures, policy interventiopns in areas such as education and job training, and outcome research studies involving costs/service utilization and quality of life. It will be useful to academic, government, non-profit research organizations and pharmaceutical companies. It will be marketed and distributed by Statistical Solutions, Ltd.
