如何从面板模型的coeftest()中获取完整聚类调整协方差矩阵

First, let's break down what's happening in your workflow: when you pass vcov = vcovHC(panelmodel, type = "sss") and cluster = "id" to coeftest, the function takes the heteroskedasticity-robust HC-SSS matrix and adjusts it for clustering on your id variable. To get this final cluster-adjusted covariance matrix for your后续 calculations, you have two simple, reliable options:

Option 1: Extract directly from the coeftest result #

The coeftest object stores the exact covariance matrix it uses to compute adjusted standard errors. You can pull it out easily using the vcov() function after saving the coeftest output:

# Load required packages
library("plm")
library("lmtest")

# Your original model setup
Data <- data.frame(id = c(rep("a",50), rep("b", 50)), y = rnorm(100), x = c(rnorm(50), rnorm(50, sd = 5)), z = c(rnorm(50), rnorm(50, sd = 3)))
panelmodel <- plm(y ~ x + z, data = Data, effect = "individual", model = "within", index = "id")

# Run coeftest and save the result
ct_result <- coeftest(panelmodel, vcov = vcovHC(panelmodel, type = "sss"), cluster = "id")

# Extract the cluster-adjusted covariance matrix
clustered_hcov_matrix <- vcov(ct_result)

# Inspect the matrix
print(clustered_hcov_matrix)

Option 2: Generate the matrix directly with vcovHC (no coeftest needed) #

You don’t have to go through coeftest to create this matrix. The vcovHC() function in plm has a built-in cluster parameter that lets you specify clustering on individual groups (your id variable). This generates the same cluster-adjusted HC-SSS covariance matrix directly:

# Generate the cluster-adjusted HC-SSS matrix directly
clustered_hcov_matrix <- vcovHC(panelmodel, type = "sss", cluster = "group")

# Note: "group" refers to the individual effect dimension (your `id` variable)
# You can also explicitly use `cluster = "id"` for clarity if preferred

This works because cluster = "group" targets the individual/group level of your panel (matching your effect = "individual" specification). The output will be identical to the matrix used in your coeftest call.

Verify consistency #

To confirm both methods produce the same matrix, run this check:

all.equal(vcov(ct_result), vcovHC(panelmodel, type = "sss", cluster = "group"))

It should return TRUE, confirming the matrices are identical.

内容的提问来源于stack exchange，提问作者Jakob