Evaluating impact: causal inference and experimental design
What works, and for whom?
Why causal inference
Monitoring tells us where we are and where we are heading. The next question is what works, and for whom: did a program or treatment change outcomes, and did it work better for some groups than for others? Here, too, Bayesian approaches can help us draw conclusions from the data at hand, especially in settings where prior information needs to be used or where effects vary across groups.
Design and analysis of experimental data
I am interested in the design and analysis of experimental and quasi-experimental studies, such as randomized trials, encouragement designs and instrumental variable settings, as well as observational data. I am especially interested in estimating effects that vary across people and settings, and in methods that applied researchers can actually use.
Prince BART
Led by Lucas Godoy Garraza, we developed Prince BART, a Bayesian nonparametric approach to causal inference that combines principal stratification with Bayesian Additive Regression Trees (BART) to estimate heterogeneous treatment effects in instrumental variable settings.
We used Prince BART to estimate the effect of family planning on women’s employment in Nigeria and Senegal. The work was part of the Family Planning Impact Consortium and is built into the consortium’s main impact tool.
- Methods paper: L.G. Garraza, I.S. Speizer, L. Alkema (2026). Combining BART and principal stratification to estimate the effect of intermediate variables on primary outcomes, with application to estimating the effect of family planning on employment in Nigeria and Senegal. Statistics and Public Policy. https://doi.org/10.1080/2330443X.2026.2672981
- R package: prince_BART
In ongoing work, supported by a UMass Faculty Research Grant / Healey Endowment Grant, we are developing additional materials and user-friendly software that explain and implement traditional and modern causal inference methods for instrumental variable settings. We are also applying Prince BART to new questions in women’s health, in the U.S. and globally, and we will share the findings and tools through workshops, trainings and publications.
Materials will be added to this page as they become available.
Bayesian methods for clinical trials
I am building on my work in Bayesian modeling and moving into Bayesian methods for clinical trials, including design, borrowing information from prior studies, and adaptive trials.
- I am completing the American Statistical Association’s certificate training in clinical trials.
- In Spring 2027, I will teach Bayesian Methods in Clinical Trials (BIOSTATS 690B), an introduction to Bayesian design and analysis of clinical trials, from using prior information to adaptive designs and master protocols. The course combines methods with practical and regulatory considerations, with implementation in R. See the course description.
Do you have a research question with an instrumental variable or a randomized encouragement design? Are you planning a trial that could benefit from a Bayesian design or analysis, or would you like to join a workshop? Send me an email.