The goal of propensity is to calculate propensity scores and weights for a wide variety of research questions.
propensity is under very early development.
You can install the development version of propensity from GitHub with:
# install.packages("devtools")
devtools::install_github("r-causal/propensity")Currently, propensity supports calculating propensity score weights for binary exposures.
There are 5 common types of weights that you might wish to use in a propensity score setting:
-
Average treatment effect (ATE): used when the target population is the entire population of interest. An example question this answers is “Should a marketing campaign be rolled out to all eligible people?”
-
Average treatment effect among the treated (ATT): used when the target population is the exposed/treated population. An example question this answers is “Should we stop our marketing campaign to those currently receiving it?”
-
Average treatment effect among the unexposed (ATU): used when the target population is the unexposed/untreated/control population. An example question this answers is “Should we send our marketing campaign to those not currently receiving it?”
-
Average treatment effect among the evenly matchable (ATM): used when the target population is those deemed “evenly matchable” by some distance metric. An example question this answers is “Should we send our marketing campaign to those of with similar demographic characteristics?”
-
Average treatment effect among the overlap (ATO): used when the target population is the same as the ATM setting, however the ATO weights are slightly attenuated with improved variance properties.
Each of these weights can optionally be “stabilized” to prevent extreme weights that lead to wide confidence intervals.
The below example shows how we would generate ATE and ATO weights for 4 participants (1 exposed and 3 unexposed) with known probabilities of being exposed (propensity scores).
library(propensity)
propensity_scores <- c(.1, .3, .4, .3)
x <- c(0, 0, 1, 0)
# Average treatment effect (ATE) weights
wt_ate(propensity_scores, .exposure = x)
#> [1] 1.111111 1.428571 2.500000 1.428571
# Stabilized ATE weights
wt_ate(propensity_scores, .exposure = x, stabilize = TRUE)
#> [1] 0.2777778 0.3571429 0.6250000 0.3571429
# Average treatment effect in the overlap (ATO) weights
wt_ato(propensity_scores, .exposure = x)
#> [1] 0.1 0.3 0.6 0.3