Design-Based Treatment Effect Estimation in Paired Cluster-Randomized Experiments

This repository contains the code used to complete the analyses discussed in "A General Framework for Design-Based Treatment Effect Estimation in Paired Cluster-Randomized Experiments."

Contents:

• analysis/

  • 00-helper-functions.R
    • functions for point and variance estimators in the simulations that are not regression-based estimators
  • 00-sim-helpers.R
    • functions to run simulations
  • 01-sim-all-est.Rmd
    • running simulations under all data settings for 20 and 200 pairs and with cluster size correlated with control outcomes or not
    • outputs: sims_m20_n150/*, sims_m200_n150/*, sims_m20_n150_neff/*, sims_m200_n150_neff/*
  • 02-cta-eda.Rmd
    • inputs: MS_HSdata.Rdata (from https://github.com/manncz/aeis-aux-rct)
    • outputs: cta-dat-clean.Rdata
    • clean TEA data for CTAI study schools and conduct EDA
  • 03-cta-ri.Rmd
    • inputs: cta-dat-clean.Rdata
    • outputs: cta_ri/*, cta_plot/*
    • runs randomization inference simulations with CTAI data and cleans simulation data for figures
  • 04-figures.Rmd
    • inputs: sims_m20_n150/*,sims_m200_n150/*, sims_m20_n150_neff/*,sims_m200_n150_neff/*, cta-dat-clean.Rdata, cta_plot/*
    • outputs: all figures
  • 05-cta-apx.Rmd
    • inputs: cta-dat-clean.Rdata, cta_plot/*
    • outputs: 05-cta-apx.html
    • explores irregular variance estimation with real data
  • temp/
    • sims_m20_n150/
    • sims_m200_n150/
    • sims_m20_n150_neff/
    • sims_m200_n150_neff/
    • sims_m200_n150_neff/
    • cta_plot/
    • cta-dat-clean.Rdata/

• figures/

  • all paper figures

Notes:

• Simulations were run in the University of Michigan Great Lakes High Performance Computing environment.
• Simulations for 20 pairs were run across 36 cores with .75 GB of memory per core. They took approximately 13 hours or 17-21:43 in CPU walltime.
• Simulations for 200 pairs were were run across 36 cores with 1 GB of memory per core. They took approximately 2 days and 18 hours or 90-23:22 in CPU walltime.
• Users will not be able to run scripts 02, 03, or 05 as is due to data use agreements for the CTA-I trial.