Horvitz-Thompson estimator for RCTs, with Joel Middleton
Htestimate calculates unbiased estimates of treatment effects from randomized trials when the random assignment is correlated across units, using the Horvitz-Thompson estimator (Särndal et al. 2003, section 2.8). Standard approaches to RCT evaluation (difference in means and regression) are generally biased under clustered randomization (Middleton 2008) or under rerandomization (Morgan & Rubin 2012), for example. In addition to the treatment effect the package produces a standard error and p-value of that effect estimate. Differences in outcome totals rather than means can also be produced. Any number of experimental arms/conditions are allowed.
This package is currently under active development so bug reports and feature requests are encouraged.
Install directly from github using devtools:
install.packages("devtools") # If not already installed.devtools::install_github("ck37/htestimate")library(htestimate)
R packages: dplyr
# Example using data from RI package.y <- c(8,6,2,0,3,1,1,1,2,2,0,1,0,2,2,4,1,1)Z <- c(1,1,0,0,1,1,0,0,1,1,1,1,0,0,1,1,0,0)# Generate 10,000 random permutations of the assignment vector.perms = ri::genperms(Z, maxiter=10000)# Estimate the probability of assignment for each unit and assignment level.prob_matrix = createProbMatrix(perms)# Estimate the treatment effect using Horvitz-Thompson.htestimate(y, Z, contrasts = c(-1, 1), prob_matrix = prob_matrix)
Aronow, P. M., & Middleton, J. A. (2013). A class of unbiased estimators of the average treatment effect in randomized experiments. Journal of Causal Inference, 1(1), 135-154.
Middleton, J. A. (2008). Bias of the regression estimator for experiments using clustered random assignment. Statistics & Probability Letters, 78(16), 2654-2659.
Morgan, K. L., & Rubin, D. B. (2012). Rerandomization to improve covariate balance in experiments. The Annals of Statistics, 40(2), 1263-1282.
Särndal, C. E., Swensson, B., & Wretman, J. (2003). Model assisted survey sampling. Springer Science & Business Media.