RFA is an R package for implementing random forest adjustment (RFA). RFA is a regression adjustment approach that partials out variation in a response and explanatory variable of interest given a set of covariates using random forests. The latest version of the package relies on ranger, which is a fast implementation of random forests. To learn more about the method, download my latest working paper here. For a more comprehensive summary of how to implement RFA in R, see the RFA vignette.
To install and attach the latest version of RFA, enter:
devtools::install_github("milesdwilliams15/RFA")
library(RFA)
RFA relies on ranger to implement random forests, and estimatr to perform linear regression on the random forest adjusted explanatory variable and response.
For a generic dataset, dataset, that contains vectors of some response variable y, an explanatory variable of interest z, and a set of confounding covariates x1, x2, and x3, random forest adjusted estimates are obtained by entering:
rfa(
y ~ z,
covariates = ~ x1 + x2 + x3,
data = dataset
)
The function returns a list consisting of an estimatr::lm_robust object, the computed random forest regressions for the response and explanatory variable (ranger::ranger objects), and the dataset used to generate random forest adjusted estimates with the partialized versions of the response and explanatory variable appended.