Panel Data Models with plm
In stargazer2: Beautiful Regression Tables for Modern Econometrics Packages

knitr::opts_chunk$set(
  collapse = TRUE,
  comment  = "",
  message  = FALSE,
  warning  = FALSE
)
library(stargazer2)

stargazer2 supports plm model objects natively. It auto-detects the estimator type (FE, RE, Pooled OLS, FD, Between), displays fixed-effect and random-effect indicator rows, and reports the appropriate fit statistics from summary.plm.

Dataset and models

The Grunfeld dataset (10 US manufacturing firms, 1935--1954, balanced panel) is included with plm. We estimate four specifications of an investment equation:

library(plm)
data("Grunfeld", package = "plm")

m_pool <- plm(inv ~ value + capital, Grunfeld,
              index = c("firm", "year"), model = "pooling")

m_fe   <- plm(inv ~ value + capital, Grunfeld,
              index = c("firm", "year"), model = "within")

m_twfe <- plm(inv ~ value + capital, Grunfeld,
              index = c("firm", "year"), model = "within", effect = "twoways")

m_re   <- plm(inv ~ value + capital, Grunfeld,
              index = c("firm", "year"), model = "random")

Default output

A bare call produces a complete table. stargazer2 reads the model type from each plm object and builds indicator rows for the fixed and random effects present across all columns:

stargazer(m_pool, m_fe, m_twfe, m_re, type = "text")

Things to notice:

Model-type row identifies each column as Pooled OLS, FE, or RE.
Firm FE and Year FE rows track which specifications absorb each set of fixed effects. Both show "Yes" only in the two-way FE column.
Firm RE row shows "Yes" for the random-effects column and "No" elsewhere.
The Constant disappears from the FE columns — the within transformation absorbs it.
The RE column's F statistic is a Wald chi-square test and is shown without degrees-of-freedom annotation.

Robust standard errors

plm ships its own vcov functions. Pass them inline through the vcov argument. vcovHC(..., method = "arellano") gives standard errors clustered by the individual unit, the most common choice for FE models:

stargazer(m_fe, m_twfe,
          type = "text",
          vcov = list(vcovHC(m_fe,   method = "arellano"),
                      vcovHC(m_twfe, method = "arellano")))

stargazer2 auto-detects the SE type from the inline call and labels the table note accordingly. For replication of Stata's vce(cluster id) results, sandwich::vcovCL is preferable as it applies the G/(G−1) small-sample correction that Stata uses; plm::vcovHC applies a heteroskedasticity-style n/(n−k) correction instead.

Driscoll-Kraay standard errors

For panels with cross-sectional dependence or long time dimensions, Driscoll-Kraay (spatial HAC) standard errors are a common alternative. plm::vcovSCC implements this estimator:

stargazer(m_fe, m_twfe,
          type = "text",
          vcov = list(vcovSCC(m_fe),
                      vcovSCC(m_twfe)))

Adding formatting options

Once the defaults look right, labels can be added. The LaTeX source below sets a title and cross-reference label, renames the dependent variable and covariates, and assigns custom column headers:

stargazer(m_pool, m_fe, m_twfe, m_re,
          type             = "latex",
          title            = "Investment Equations: Grunfeld Panel Data",
          label            = "tab:grunfeld",
          dep.var.labels   = "Investment",
          covariate.labels = c("Market Value", "Capital Stock"),
          column.labels    = c("Pooled OLS", "FE", "Two-way FE", "RE"))