Methodology
2026-04-109 min read

Panel Data in R: fixest vs. plm vs. lfe

Comparing fixest, plm, and lfe for panel data estimation in R. Speed benchmarks, syntax differences, and when to use each.

Sytra Team
Research Engineering Team, Sytra AI

Panel data in R means choosing between three packages: fixest (the modern standard), plm (the textbook choice), and lfe (archived but still used). The choice matters for speed, syntax, and what diagnostics are available.

fixest: The Modern Standard

library(fixest)
 
# Two-way FE with multi-way clustering
model <- feols(y ~ x1 + x2 | firm + year,
data = df, cluster = ~firm + year)
 
# Multiple models at once
models <- feols(y ~ csw(x1, x2, x3) | firm + year,
data = df, cluster = ~firm)
etable(models, tex = TRUE)

fixest is R’s answer to Stata’s reghdfe. It’s the fastest option, supports multi-way FE and clustering, and has the best export tools (etable() produces LaTeX tables directly). The csw() function runs stepwise additions in a single call.

plm: The Textbook Package

library(plm)
 
pdf <- pdata.frame(df, index = c("firm", "year"))
model <- plm(y ~ x1 + x2, data = pdf, model = "within")
 
# Hausman test
re_model <- plm(y ~ x1 + x2, data = pdf, model = "random")
phtest(model, re_model)

plm is what most econometrics textbooks teach. Its advantage: built-in Hausman test, Breusch-Pagan test, and other panel diagnostics. Its disadvantage: slow on large datasets, limited to one-way FE without workarounds, and the pdata.frame requirement is clunky.

Stop fighting with syntax.

Sytra is an AI research assistant built specifically for statistical computing. No more copy-pasting code into ChatGPT.

Get Early Access

lfe: Archived but Alive

library(lfe)
model <- felm(y ~ x1 + x2 | firm + year | 0 | firm, data = df)

lfe was the pre-fixest standard. Its felm() function uses a pipe-delimited syntax: outcome ~ regressors | FE | IV | cluster. It’s archived on CRAN (no longer actively maintained) but still installs and works. If you’re maintaining legacy code, it’s fine. For new projects, use fixest.

Speed Benchmarks

1M observations, 50K firms, 20 years

fixest::feols()
~1.2 seconds
lfe::felm()
~8 seconds
plm::plm(within)
~35 seconds

Decision Tree

  • New project, any size: fixest
  • Need Hausman test: plm
  • Legacy code: lfe (but consider migrating)
  • Speed critical: fixest (not even close)
  • Publication tables: fixest (etable() is best-in-class)
#Panel Data#Fixed Effects#R#Economics

Enjoyed this article?

Related Guides

Methodology9 min read

Panel Data in Stata: xtreg vs. reghdfe vs. areg

When to use xtreg, areg, or reghdfe for panel data in Stata. Includes speed benchmarks, syntax comparison, and common mistakes.

Methodology12 min read

Difference-in-Differences in R: didimputation, did, and fixest

A complete guide to modern DiD estimation in R — using fixest, did, and didimputation — with parallel trends testing and event study plots.

R + AI9 min read

Why ChatGPT Produces Invalid R Code for Statistical Analysis

ChatGPT knows R syntax better than Stata. But it still produces statistically invalid code — wrong standard errors, missing diagnostics, and unvalidated inference.

Methodology9 min read

Multilevel Models in Stata: HLM for Education Research

A practical guide to multilevel/HLM models in Stata using the mixed command. Students nested in schools. Patients nested in hospitals. Here's how.

Methodology11 min read

Matching Estimators in Stata: PSM, CEM, and Modern Alternatives

A guide to matching methods in Stata — propensity score matching, coarsened exact matching, and why you should probably use teffects instead of psmatch2.

View all articles