关于R 3.6.3（macOS Catalina）运行Zelig及替代方案的技术问询

Great question—let’s break this down step by step for your macOS Catalina 10.15.4 + R 3.6.3 setup:

The latest CRAN releases of Zelig (v5.1.7 and above) are built for R 4.0+, so they won’t work natively with your R 3.6.3 install out of the box.

You could try installing an older, compatible version of Zelig (like v5.1.1) using:

library(devtools)
install_version("Zelig", version = "5.1.1")

But heads up: this might require tracking down older versions of dependent packages (e.g., ggplot2, dplyr) that also support R 3.6.3, which can be tedious. On Catalina, you’ll also need Xcode 11.x or its command-line tools to compile these older packages, so ensure those are set up first.

That said, if you want a smoother workflow, switching to MatchIt or the Matching package is far more reliable for your setup.

Both packages work seamlessly with R 3.6.3 on Catalina and make it straightforward to compute treatment effects after nearest-neighbor PSM.

Using MatchIt #

MatchIt is user-friendly and integrates well with standard tidy workflows:

• First, run nearest-neighbor propensity score matching:

library(MatchIt)
# Replace with your treatment variable, covariates, and dataset
match_obj <- matchit(treat ~ cov1 + cov2 + cov3, 
                     data = your_dataset, 
                     method = "nearest", 
                     ratio = 1) # 1:1 matching

• Extract the matched dataset (optional but useful for further analysis):

matched_data <- match.data(match_obj)

• Compute ATT: The summary() function directly returns the ATT estimate and its standard error:

summary(match_obj, standardize = FALSE)

Look for the Estimate value under the "Average Treatment Effect on the Treated (ATT)" section.

• Compute ATE: Nearest-neighbor matching is optimized for ATT, but you can estimate ATE using propensity score weighting instead:

weighted_match <- matchit(treat ~ cov1 + cov2 + cov3, 
                          data = your_dataset, 
                          method = "weighting", 
                          estimand = "ATE")
summary(weighted_match, standardize = FALSE)

The Estimate under "Average Treatment Effect (ATE)" is your result.

Using the Matching Package #

The Matching package (often called Match for short) lets you directly specify whether you want ATT or ATE during matching:

• First, estimate propensity scores and run matching:

library(Matching)
# Step 1: Fit a propensity score model
ps_model <- glm(treat ~ cov1 + cov2 + cov3, 
                data = your_dataset, 
                family = binomial)
ps_scores <- predict(ps_model, type = "response")

# Step 2: Calculate ATT with nearest-neighbor matching
att_results <- Match(Y = your_dataset$outcome_variable, 
                     Tr = your_dataset$treat, 
                     X = ps_scores, 
                     estimand = "ATT", 
                     method = "nearest")
summary(att_results)

# Step 3: Calculate ATE with nearest-neighbor matching
ate_results <- Match(Y = your_dataset$outcome_variable, 
                     Tr = your_dataset$treat, 
                     X = ps_scores, 
                     estimand = "ATE", 
                     method = "nearest")
summary(ate_results)

The summary() output will include the treatment effect estimate, standard error, and p-value for both metrics.

内容的提问来源于stack exchange，提问作者Tanu