dcme
Double-Complier Mediation Effects with Instrumental Variables
Installation
You can install the development version of dcme from GitHub with:
# install.packages("devtools")
devtools::install_github("nt-williams/dcme")Example
library(dcme)
#> Loading required package: mlr3superlearner
#> Loading required package: mlr3learners
#> Loading required package: mlr3
n <- 1000
w_1 <- rbinom(n, 1, prob = 0.6)
w_2 <- rbinom(n, 1, prob = 0.3)
w <- cbind(w_1, w_2,
rbinom(n, 1, prob = pmin(0.2 + (w_1 + w_2) / 3, 1)))
a <- rbinom(n, 1, prob = 0.5)
l <- rbinom(n, 1, plogis(rowMeans(-log(2) + w + 3*a - 1)))
z <- rbinom(n, 1, plogis(rowMeans(-log(1.1) * w) + 3*a - l))
m <- rbinom(n, 1, plogis(rowSums(-log(3) * w[,-3]) + 2*l + z - 1))
y <- rbinom(n, 1, plogis(rowSums(-log(5)*w) + z + m + .3))
colnames(w) <- paste("W", seq_len(ncol(w)), sep = "")
tmp <- as.data.frame(cbind(W = w, A = a, Z = z, L = l, M = m, Y = y))
dcme(tmp, paste0("W", 1:3), "A", "L", "Z", "M", "Y", "binomial", 1)
#> ℹ n effective = 1000. Setting cross-validation folds as 10
#> ℹ n effective = 1000. Setting cross-validation folds as 10
#> ℹ n effective = 1000. Setting cross-validation folds as 10
#> ℹ n effective = 1000. Setting cross-validation folds as 10
#> ℹ n effective = 1000. Setting cross-validation folds as 10
#> Estimand psi 95% CI
#> 1 TIIDE 0.133 0.047, 0.218
#> 2 TIIIE 0.012 -0.021, 0.046
#> 3 JFS 0.354 0.216, 0.492
#> 4 CIDE 0.374 -0.058, 0.128
#> 5 CIIE 0.035 0.081, 0.668