A Bayesian machine learning approach to direct and spillover effects in cluster-randomized trials with noncompliance

Key points

• To flexibly model outcomes and treatment receipt while accounting for within-cluster dependence, we combine Bayesian additive regression trees with cluster-level random effects.1
• We augment the outcome regression with a peer-treatment clever covariate, defined as the predicted mean proportion of treated peers from a preliminary receipt model, to mitigate regularization-induced distortion of the effect decomposition.1
• Within a Bayesian g-computation framework, we estimate cluster-average intent-to-treat effects and cluster-average local average treatment effects and examine sensitivity to the cross-world restrictions underlying their decomposition.1
• We find that the clever covariate improves the decomposition of the total effect into direct and spillover components under moderate or strong spillover, with similar gains in smaller clusters.1

Sources (1)

• [1]A Bayesian machine learning approach to direct and spillover effects in cluster-randomized trials with noncompliance
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 12:20 AM

  We develop a Bayesian machine learning approach for estimating direct and spillover effects in cluster-randomized trials with treatment noncompliance.

  To flexibly model outcomes and treatment receipt while accounting for within-cluster dependence, we combine Bayesian additive regression trees with cluster-level random effects.

Extractive summary: sentences quoted from the sources.