thinkCausal: Practical Tools for Understanding and Implementing Causal Inference Methods

thinkCausal: Practical Tools for Understanding and Implementing Causal Inference Methods is sponsored by Institute of Education Sciences (IES). Developed a software package, thinkCausal, to assist education researchers in understanding and implementing causal inference methods.

thinkCausal: Practical Tools for Understanding and Implementing Causal Inference Methods | IES thinkCausal: Practical Tools for Understanding and Implementing Causal Inference Methods Statistical and Research Methodology in Education (07/01/2020 - 06/30/2024) Methodological Innovation The purpose of this grant was to develop a highly scaffolded multi-purpose causal inference software package, thinkCausal, with the Bayesian Additive Regression Trees (BART) predictive algorithm as a foundation.

This new tool scaffolds education researchers through the data analytic process, from uploading data all the way through to graphical and tabular displays of results. The software also has interactive educational components to provide researchers with the opportunity to gain a deeper understanding of the methods and underlying assumptions just at the time that they need it.

thinkCausal will allow education researchers from varied backgrounds to access sophisticated machine learning algorithms and better understand causal inference. The performance of the thinkCausal tool was evaluated relative to other choices for estimating causal effects in observational studies.

This randomized experiment demonstrated that users were more likely to obtain accurate treatment effect estimates and uncertainty intervals, and did so in less time, as compared to users employing other methodological options. The core BART algorithms were also extended to accommodate multilevel data structures, resulting in a new R package, stan4bart.

The performance of stan4bart was evaluated using simulations and found to be superior to standard alternatives. The project performed randomized experiments to reveal how typical students understand language used to describe research findings. Results suggest that many students interpret many findings causally even when the language is intended to be non-causal.

However, specific language choices and contexts did have an impact on the level of causal attribution. People and institutions involved Co-principal investigator Products and publications Dorie, V. , Perrett, G.

, Hill, J. L. , & Goodrich, B.

(2022). Stan and BART for Causal Inference: Estimating Heterogeneous Treatment Effects Using the Power of Stan and the Flexibility of Machine Learning . Entropy, 24 (12), 1782.

Hill, J. , Perrett, G. & Dorie, V.

(2023). Machine Learning for Causal Inference. In J.

R. Zubizarreta, E. A Stuart, D.

S. Small, & P. R Rosenbaum (Eds.)

, Handbook of Multivariate Matching and Weighting for Causal Inference (pp. 416-443). Chapman & Hall/CRC: Boca Raton, FL [ERIC Accession number: ED660568] Hill, J.

, Perrett, G. , Hancock, S. , Bergner, Y.

, & Win, L. (2024). Causal Language and Statistics Instruction: A randomized experiment.

Statistics Education Research Journal, 23 (1). [ERIC Accession number: ED660558] Additional project information Find available citations in ERIC for this award here . NYU/Columbia Postdoctoral Training Program Sensitivity Analysis—If We're Wrong, How Far Are We from Being Right?

What, When, and for Whom? Principled Estimation of Effect Heterogeneity Across Multiple Treatments, Outcomes, and Groups Questions about this project? To answer additional questions about this project or provide feedback, please contact the program officer.

Questions about this project? To answer additional questions about this project or provide feedback, please contact the program officer. Summer Research Training Institute on Cluster-Rand...

Data Science for Education (DS4EDU) Designing Effective Surveys to Support Continuous ...