Estimating effects?
ronikobrosly opened this issue · 6 comments
Hi, I love this implementation because it allows for continuous predictors/exposures to be analyzed! It seems like the R package "multiPIM" doesn't allow for that. So thanks!
This isn't so much an issue on the repo, but: do you know of a R implementation of a similar approach (involving TMLE) that allows for estimation of effect between a continuous exposure and a binary outcome? Like in the case of the example you provided in markdown:
Create test dataset.
set.seed(1)
N <- 200
num_normal <- 7
X <- as.data.frame(matrix(rnorm(N * num_normal), N, num_normal))
Y <- rbinom(N, 1, plogis(.2X[, 1] + .1X[, 2] - .2X[, 3] + .1X[, 3]X[, 4] - .2abs(X[, 4])))
Add some missing data to X so we can test imputation.
for (i in 1:10) X[sample(nrow(X), 1), sample(ncol(X), 1)] <- NA
It seems like this package only accepts binary exposures: https://cran.r-project.org/web/packages/tmle/tmle.pdf
Do you have any suggestions?
I'd check out https://github.com/osofr/stremr which says that it supports continuous exposures. I haven't had a chance to try it out myself yet though.
@ronikobrosly Just curious, did stremr work or did you find something else?
Thanks for recommending stremr, but I struggled with it. I ended up using the approach on page 28 of this paper: https://www.jstatsoft.org/article/view/v051i13/v51i13.pdf
Basically, the idea is you create multiple exposure categories and find the average treatment effect.
Ah gotcha. Well if you have time/motivation you might post an issue to stremr, I bet Oleg would try to improve it if possible.
I certainly have the motivation! I'll post that shortly.
Thanks Chris.