Direct Adjustment in Combination With Robust or Nonlinear Regression: Software and Methods for RDDs, RCTs and Matched Observational Studies

Co-Principal Investigators: Bowers, Jacob; Errickson, Josh; Sales, Adam

The research team will develop the tools for three broad areas of covariate use: 1) Modeling outcome-covariate associations and estimating effects in distinct regression fits, such as when using non-linear and robust regression methods in analyses that target average treatment effects; 2) Complex research designs, including blocked or clustered observations, regression discontinuity designs, and matched observational studies; 3) Scenarios in which the number of available covariates is large compared with the number of cases, including most studies combining school-level assignment with data from a state's student longitudinal data system. They will test and demonstrate the methods using Monte Carlo simulations and real datasets. They will conduct user-testing for the open-source R software package with faculty and students at the research team's universities and then in increasingly broader settings, such as conference workshops, as it develops and evolves in response to user feedback. The research team will demonstrate use with primary effectiveness analyses of state programs in Michigan, and secondary analysis of student data from a randomized trial conducted in Maine.

Publications:

Lycurgus, T., & Hansen, B. B. (2021). An Aggregation Scheme for Increased Power. arXiv preprint arXiv:2107.13070.

Lycurgus, T., Hansen, B. B., & White, M. (2023). Conjuring Power from a Theory of Change: The PWRD Method for Trials with Anticipated Variation in Effects. Journal of Research on Educational Effectiveness, 16(4), 707–733.