This R package estimates the statistical parameter E(E(Y|A=1,B)-E(Y|A=0,B)|A=a)
using targeted maximum likelihood estimation (TMLE). If A
is a treatment, then this is sort of like a Treatment Effect, Conditional on one of the treatments, hence the name tmlecte
. Under various causal models, this statistical parameter has interpretations of things like
- the affect of treatment among the treated (ATT)
- the natural direct effect (NDE), and
- the natural direct effect among the untreated.
The methods in this package will probably be added to the tmle package on CRAN in the future.
#Installation
This package can be installed directly from github using the devtools
package:
install.packages("devtools") #if you don't have it already
library(devtools)
install_github("lendle/tmlecte")
Alternatively you could download the source or clone it with git, then R CMD BUILD
and R CMD INSTALL
it.
#Usage
Once installed, the package is loaded with library(tmlecte)
. Documentation is available in R's built in documentation system (?tmlecte
). Some examples are provided in the documentation.
#References
Lendle, S. D., Subbaraman, M. S., and van der Laan, M. J. (2013). Identification and efficient estimation of the natural direct effect among the untreated. Biometrics, 69(2), 310-317.
Van der Laan, M. J., and Rose, S. (2011). Targeted learning: causal inference for observational and experimental data. Springer.