Making a function for matching on multiple dependent variables, purrr - r

Question

I want to estimate the matched treatment effect using the Matching package on multiple dependent variables.
For just a single dependent variable, I can run the below which returns what I want:
library(carData)
library(purrr)
library(tidyverse)
library(Matching)
matching_df <- Mroz %>%
mutate(wc = case_when(wc == "yes" ~ "TRUE",
wc == "no" ~ "FALSE")) %>%
drop_na(k5, k618, age, wc, hc, lfp)
matching_df$wc <- as.logical(matching_df$wc)
ps1 <- glm(wc ~ k5 + k618 + age + hc,
family = binomial, data = matching_df)
pscore <- ps1$fitted.values
matching_df <- cbind(matching_df, pscore)
Y <- matching_df$lfp
Tr <- as.logical(matching_df$wc)
psm1 <- Matching::Match(
Y = Y,
Tr = Tr,
X = pscore,
estimand = "ATT",
M = 1,
replace = TRUE,
caliper = 0.05,
version = "fast")
summary(psm1)
Estimate... 0.17479
SE......... 0.044963
T-stat..... 3.8873
p.val...... 0.00010135
Original number of observations.............. 753
Original number of treated obs............... 212
Matched number of observations............... 207
Matched number of observations (unweighted). 1074
Caliper (SDs)........................................ 0.05
Number of obs dropped by 'exact' or 'caliper' 5
But when I try and make a function using purrr:map_dfr so I can repeat this operation for multiple dependent variables, it returns an error. This is my attempt at the function:
vars <- c("lfp", "lwg", "inc")
names(vars) <- vars
matching_fcn <- function(.x){
matching_df <- Mroz %>%
mutate(wc = case_when(wc == "yes" ~ "TRUE",
wc == "no" ~ "FALSE")) %>%
drop_na(k5, k618, age, wc, hc, .x)
matching_df$wc <- as.logical(matching_df$wc)
ps1 <- glm(wc ~ k5 + k618 + age + hc,
family = binomial, data = matching_df)
pscore <- ps1$fitted.values
matching_df <- cbind(matching_df, pscore)
Y <- matching_df$.x
Tr <- as.logical(matching_df$wc)
psm1 <- Matching::Match(
Y = Y,
Tr = Tr,
X = pscore,
estimand = "ATT",
M = 1,
replace = TRUE,
caliper = 0.05,
version = "fast")
summary(psm1)
}
purrr::map_dfr(
.x = all_of(vars),
.f = matching_fcn)
Error: All columns in a tibble must be vectors.
x Column `lfp` is a `summary.Match` object.
x Column `lwg` is a `summary.Match` object.
x Column `inc` is a `summary.Match` object.
Run `rlang::last_error()` to see where the error occurred.
Ultimately, I would like a tibble which includes the name of the dependent variable in one column, then the estimate, se, T-stat, and p.val that are returned by the Matching::Match function in other columns

Answer 1

The summary(psm1) can't be put into a tibble. So choose some values of psm1 and make your own. Further, drop_na is no good idea and will bias your results.
library(Matching)
vars <- c("dependent_var_1", "dependent_var_2", "dependent_var_3")
names(vars) <- vars
matching_fcn <- function(.x){
# matching_df <- matching_df %>%
# drop_na(covar_1, covar_2, covar_3, covar_4, covar_5, covar_6, covar_7, treat_1, .x)
ps1 <- glm(treat_1 ~ covar_1 + covar_2 + covar_3 + covar_4 + covar_5 + covar_6 + covar_7,
family = binomial, data = matching_df)
pscore <- ps1$fitted.values
matching_df <- cbind(matching_df, pscore)
Y <- matching_df[[.x]]
Tr <- matching_df$treat_1
psm1 <- Matching::Match(
Y = Y,
Tr = Tr,
X = pscore,
estimand = "ATT",
M = 1,
replace = TRUE,
caliper = 0.05,
version = "fast")
p <- 1 - pnorm(abs(psm1$est.noadj/psm1$se.standard))
with(psm1, tibble(dv=.x, est=est.noadj, se=se.standard, p=p, ndrops=ndrops))
}
Usage and result
library(dplyr)
library(tidyr)
purrr::map_df(
.x = tidyselect::all_of(vars),
.f = matching_fcn)
# # A tibble: 3 × 5
# dv est se p ndrops
# <chr> <dbl> <dbl> <dbl> <dbl>
# 1 dependent_var_1 0.652 0.231 0.00238 8
# 2 dependent_var_2 -0.216 0.188 0.125 8
# 3 dependent_var_3 -0.506 0.249 0.0210 8
Data
v <- c('covar_1', 'covar_2', 'covar_3', 'covar_4', 'covar_5', 'covar_6',
'covar_7', 'treat_1', 'dependent_var_1', 'dependent_var_2', 'dependent_var_3')
set.seed(830595665)
matching_df <- data.frame(matrix(rnorm(100*length(v)), 100, length(v), dimnames=list(c(), v)))
matching_df$treat_1 <- +(matching_df$treat_1 > 0)