README.md

HTEBayes

In this repository, we provide the code for the function and simulation study of the paper "Confounder-Dependent Bayesian Mixture Model: Characterizing Heterogeneity of Causal Effects in Air Pollution Epidemiology" by D. Zorzetto, F.J. Bargagli-Stoffi, A. Canale, and F. Dominici, Biometrics, Volume 80, Issue 2, 2024.

Examples:

Load functions:

source("src/functions_simulations.R")
source("src/model_CDBMM.R")

Information for simulation data:

# sample size
n = 500

#set seed
seed = 1

Parameters for the model:

# iterations
R = 3000
R_burnin = 2000

# number of maximum groups for the two treatment levels
L_0 = 20
L_1 = 20

Example 1 (3 groups)

Generation dataset and estimation CDBMM:

dataset <- simulation_sample_3groups(seed=seed,
                                     eta_0=c(2,4,6),
                                     eta_1=c(0,3,6),
                                     sigma_0=rep(0.3,3),
                                     sigma_1=rep(0.3,3),
                                     n=n)

cdbmm_results <- CDBMM_Gibbs(c=1,
                             data_sample=list(dataset),
                             n=n)

Visualize ITEs and GATEs for discovered groups:

hist(cdbmm_results$tau,
     nclass = 50,
     main = "ITEs")

group_allocation<-paste0(cdbmm_results$partition[,1],"-",cdbmm_results$partition[,1])

sapply(unique(group_allocation), function(g) 
  cdbmm_results$atoms$p_1[as.integer(substr(g, 3, 3))]-
    cdbmm_results$atoms$p_0[as.integer(substr(g, 1, 1))])

Example 2 (5 groups)

Generation dataset and estimation CDBMM:

dataset <- simulation_sample_5cov(seed=seed,
                                  eta_0=c(2,2,3,4.5,6.5),
                                  eta_1=c(0,1,2.5,5,7.5),
                                  sigma_0=rep(0.2,5),
                                  sigma_1=rep(0.2,5),
                                  n=n)

cdbmm_results <- CDBMM_Gibbs(c=1,
                             data_sample=list(dataset),
                             n=n)

Visualize ITEs and GATEs for discovered groups:

hist(cdbmm_results$tau,
     nclass = 50,
     main = "ITEs")

group_allocation<-paste0(cdbmm_results$partition[,1],"-",cdbmm_results$partition[,1])

sapply(unique(group_allocation), function(g) 
  cdbmm_results$atoms$p_1[as.integer(substr(g, 3, 3))]-
    cdbmm_results$atoms$p_0[as.integer(substr(g, 1, 1))])

Cite

@article{zorzetto2024confounder,
  title={Confounder-dependent Bayesian mixture model: Characterizing heterogeneity of causal effects in air pollution epidemiology},
  author={Zorzetto, Dafne and Bargagli-Stoffi, Falco J and Canale, Antonio and Dominici, Francesca},
  journal={Biometrics},
  volume={80},
  number={2},
  year={2024},
  publisher={Oxford University Press}
}