Tools for Planning & Analyzing Randomized Controlled Trials & A/B Tests

This repository contains the materials for the EDM 2024 half day tutorial "Tools for Planning & Analyzing Randomized Controlled Trials & A/B Tests" on 7/14/24. This tutorial is organized by Adam Sales, Johann Gagnon-Bartsch, Duy Pham, Charlotte Mann, and Jaylin Lowe.

The tutorial focuses on how to use the current version of the dRCT package in R.

Contents

The contents are organized as follows:

• data/: processed data files and documentation
  • README.md: data dictionaries
  • A/B example data:
    • abTestExample.csv
    • auxiliaryLogData.csv
    • rctLogData.csv
    • yhatNN.csv
  • Field test example data (AEIS) outputted by 00-data-setup.Rmd:
    • rct_aux_dat.Rdata
    • rct_schools.Rdata
    • aux_dat_small.csv: Smaller version of rct_aux_dat.Rdata with some covariates removed for the Shiny app demo.
• input/: input data files
  • MS_data_public.Rdata: AEIS data from manncz/aeis-aux-rct/
  • var_names.csv: Cross-walk for cleaning variable names in AEIS data
  • subset_var_names.csv: Names of covariates to be used in creating the smaller version of the auxiliary data for the Shiny app demo
• scripts/: all tutorial scripts to be run in order
  • _installPackages.R: contains code to install all packages used in tutorial scripts
  • 00-data-setup.Rmd
  • 01-explore-aeis-data.Rmd
  • 02-effect-est.Rmd
  • 03-integrate-aux.Rmd
  • 04-effect-estABtest.Rmd
  • 04-effect-estABtestSolution.Rmd
  • 05-heterogeneousEffects.Rmd
  • compiled_tutorials/: compiled versions of all tutorial scripts
• temp/: intermediate data artifacts
  • auxpred.Rdata: auxiliary predictions to be used in 03-integrate-aux.Rmd

Use

Option 1 -- Run Everything:

1. Run 00-data-setup.Rmd
2. Now the scripts can be run in any order, although we will work through the scripts in order for the workshop

Option 2 -- Skip Data Build:

1. Skip 00-data-setup.Rmd and can can still run the scripts in any order because the outputs are already saved in the data/ directory

Option 3 -- Run Nothing:

1. Read through the compiled scripts in the scripts/compiled_tutorials/ directory

Real-Data Examples

We work through two real-data examples in this workshop - a educational school-level field experiment (AEIS data) and educational A/B tests (ASSISTments).

AEIS Data and Synthetic RCT

In scripts 00-03, we work through an example of analyzing a school-level field experiment using data provided by the Texas Education Agency called the Academic Excellence Indicator System (AEIS). Data documentation can be found on the AEIS website here and here.

We use an already processed version of the AEIS data for this tutorial (MS_data_public.Rdata). A detailed description of our data processing can be found on at github.com/manncz/aeis-aux-rct.

Inspired by the Cognitive Tutor Algebra I Study (Pane et. al, 2014), we construct a synthetic RCT with the middle schools included in the Texas AEIS data for this workshop.

ASSISTments Data

You then have the chance to implement what you learned in 04-effect-estABtest.Rmd, with data from real educational A/B tests.

References

Johann A. Gagnon-Bartsch, Adam C. Sales, Edward Wu, Anthony F. Botelho, John A. Erickson, Luke W. Miratrix, and Neil T. Heffernan. Precise unbiased estimation in randomized experiments using auxiliary observational data. Journal of Causal Inference, 11(1):20220011, August 2023. URL: https://www.degruyter.com/document/doi/10.1515/jci-2022-0011/html.

Kosuke Imai. Variance identification and efficiency analysis in randomized experiments under the matched-pair design. Statistics in Medicine, 27(24):4857–4873, October 2008. URL: https://onlinelibrary.wiley.com/doi/10.1002/sim.3337.

Kosuke Imai and Zhichao Jiang. experiment: R Package for Designing and Analyzing Randomized Experiments. April 2022. URL https://cran.r-project.org/web/packages/experiment/index.html.

Jaylin Lowe, Charlotte Mann, Jiaying Wang, Adam Sales and Johann Gagnon-Bartsch. Power Calculations for Randomized Controlled Trials with Auxiliary Observational Data EDM2024.

Charlotte Z. Mann, Adam C. Sales, and Johann A. Gagnon-Bartsch. A General Framework for Design-Based Treatment Effect Estimation in Paired Cluster-Randomized Experiments. Preprint. July 2024. URL: https://arxiv.org/abs/2407.01765.

John F. Pane, Beth Ann Griffin, Daniel F. McCaffrey, and Rita Karam. Effectiveness of Cognitive Tutor Algebra I at Scale. Educational Evaluation and Policy Analysis, 36(2): 127–144, June 2014. URL: https://doi.org/10.3102/0162373713507480.

Duy Pham, Kirk Vanacore, Adam Sales and Johann Gagnon-Bartsch. LOOL: Towards Personalization with Flexible & Robust Estimation of Heterogeneous Treatment Effects EDM2024.

Texas Education Agency. Academic Excellence Indicator System. 2020. URL: https://rptsvr1.tea.texas.gov/perfreport/aeis/index.html. Accessed on 2/12/2024.

Edward Wu and Johann A. Gagnon-Bartsch. The LOOP Estimator: Adjusting for Covariates in Randomized Experiments. Evaluation Review, 42(4):458–488, August 2018. URL: https://doi.org/10.1177/0193841X18808003.

Edward Wu and Johann A. Gagnon-Bartsch. Design-Based Covariate Adjustments in Paired Experiments. Journal of Educational and Behavioral Statistics, 46(1):109–132, February 2021. URL: https://doi.org/10.3102/1076998620941469.

Edward Wu, Adam C. Sales, Charlotte Z. Mann, and Johann A. Gagnon-Bartsch. dRCT, December 2023. URL: https://github.com/manncz/dRCT.