A central challenge in causal inference from observational data is adjusting for confounding without imposing strong parametric assumptions. My research on matching and weighting addresses this challenge across a range of settings. Early work introduced matching as nonparametric preprocessing for parametric causal inference, showing that preprocessing data with methods such as those implemented in the MatchIt software package can substantially reduce model dependence and improve the robustness of causal estimates, and clarified common misunderstandings about the relationship between experimental and observational approaches. A parallel line of work developed the Covariate Balancing Propensity Score (CBPS), which estimates propensity scores by directly optimizing covariate balance—yielding more robust inverse probability weights for marginal structural models, continuous treatments, high-dimensional settings, and with optimal theoretical guarantees. A related contribution generalized the propensity score to arbitrary treatment regimes, with extensions to non-binary treatments. For panel and time-series cross-sectional data, I showed that unit fixed effects regression models implicitly perform a form of matching, developed analogous results for two-way fixed effects models, and built the PanelMatch framework to make these connections explicit. Additional work addressed matched-pairs designs in randomized experiments—including cluster-randomized trials and sensitivity analysis for missing outcomes—and applied these ideas to evaluate large-scale social programs. Recent work extends these methods to spatio-temporal settings where the treatment is a point process in space and time, with extensions to heterogeneous treatment effects and the geocausal package for spillover and carryover effects. Further ongoing work develops covariate balancing estimators for continuous treatments that exploit the continuity of treatment dose, low-rank covariate balancing estimators under network interference, and distributionally robust causal estimators that remain valid under departures from the assumed propensity score model.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Imai, Kosuke and Marc Ratkovic
(2014).
`` Covariate Balancing Propensity Score.''
Journal of the Royal Statistical Society, Series B (Statistical Methodology), Vol. 76, No. 1 (January), pp. 243-263.
|
|
King, Gary, Emmanuela Gakidou, Kosuke Imai, Jason Lakin, Ryan T. Moore, Clayton Nall, Nirmala Ravishankar, Manett Vargas, Martha María Téllez-Rojo, Juan Eugenio Hernández Ávila, Mauricio Hernández Ávila, and Héctor Hernández Llamas
(2009).
`` Public Policy for the Poor? A Randomised Assessment of the Mexican Universal Health Insurance Programme.''
(with a comment) The Lancet, Vol. 373, No. 9673 (April), pp. 1447-1454.
|
|
|
|
|
|
|
|