Panel data

Start with the expanded example for each command. Open the variants for options and alternative R approaches. Read the qualifications before adapting the code.

hausman

Hausman comparison of fixed and random effects

Stata

hausman fixed random

R plm

plm::phtest(fe, re)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Requires same regressors and compatible covariance assumptions.

Stata reference · R reference

xtcloglog

Random-intercept panel cloglog

Stata

xtcloglog binary x, re

R lme4

lme4::glmer(binary ~ x + (1 | id), data = data, family = stats::binomial(link = "cloglog"), nAGQ = 12)

Differs from Stata Uses normal panel random intercepts and 12-point adaptive quadrature, matching Stata intpoints(12). Quadrature adaptation and convergence criteria remain package-specific.

Stata reference · R reference

xtdpd

One-step difference GMM for a dynamic panel

Stata

xtdpd y L.y x, div(x) dgmmiv(y, lagrange(2 3))

R plm

{ plm <- plm::plm; plm::pgmm(y ~ lag(y, 1) + x | lag(y, 2:3), data = pd, effect = "individual", model = "onestep", transformation = "d", collapse = TRUE) }

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Supports the standard difference-GMM subset, with deliberately collapsed instruments. Stata xtdpd supports richer error processes; align instruments, initial weight matrices and finite-sample corrections before comparing.

Stata reference · R reference

xtdpdsys

System GMM with level and differenced equations

Stata

xtdpdsys y x, maxldep(3)

R plm

{ plm <- plm::plm; plm::pgmm(y ~ lag(y, 1) + x | lag(y, 2:3), data = pd, effect = "individual", model = "onestep", transformation = "ld", collapse = TRUE) }

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Additional level moment conditions require stronger initial-condition assumptions. Instrument collapsing, instrument ranges and weighting differ unless explicitly matched; no blanket parity with xtdpdsys defaults.

Stata reference · R reference

xtgee

binomial GEE with exchangeable working correlation

Stata

xtgee binary x, family( binomial ) corr( exchangeable )

R geepack

geepack::geeglm(binary ~ x, id = id, waves = time, data = data, family = stats::binomial(), corstr = "exchangeable")

Differs from Stata Keep rows sorted in contiguous panels. geeglm defaults to sandwich robust standard errors (std.err=“san.se”), corresponding to xtgee, vce(robust); Stata defaults to conventional model-based standard errors. Scale estimation, small-sample adjustments and correlation estimation may still differ.

Stata reference · R reference

binomial GEE with independence working correlation

Stata

xtgee binary x, family( binomial ) corr( independence )

R geepack

geepack::geeglm(binary ~ x, id = id, waves = time, data = data, family = stats::binomial(), corstr = "independence")

Differs from Stata Keep rows sorted in contiguous panels. geeglm defaults to sandwich robust standard errors (std.err=“san.se”), corresponding to xtgee, vce(robust); Stata defaults to conventional model-based standard errors. Scale estimation, small-sample adjustments and correlation estimation may still differ.

Stata reference · R reference

gaussian GEE with ar1 working correlation

Stata

xtgee y x, family( gaussian ) corr( ar1 )

R geepack

geepack::geeglm(y ~ x, id = id, waves = time, data = data, family = stats::gaussian(), corstr = "ar1")

Differs from Stata Keep rows sorted in contiguous panels. geeglm defaults to sandwich robust standard errors (std.err=“san.se”), corresponding to xtgee, vce(robust); Stata defaults to conventional model-based standard errors. Scale estimation, small-sample adjustments and correlation estimation may still differ.

Stata reference · R reference

gaussian GEE with exchangeable working correlation

Stata

xtgee y x, family( gaussian ) corr( exchangeable )

R geepack

geepack::geeglm(y ~ x, id = id, waves = time, data = data, family = stats::gaussian(), corstr = "exchangeable")

Differs from Stata Keep rows sorted in contiguous panels. geeglm defaults to sandwich robust standard errors (std.err=“san.se”), corresponding to xtgee, vce(robust); Stata defaults to conventional model-based standard errors. Scale estimation, small-sample adjustments and correlation estimation may still differ.

Stata reference · R reference

gaussian GEE with independence working correlation

Stata

xtgee y x, family( gaussian ) corr( independence )

R geepack

geepack::geeglm(y ~ x, id = id, waves = time, data = data, family = stats::gaussian(), corstr = "independence")

Differs from Stata Keep rows sorted in contiguous panels. geeglm defaults to sandwich robust standard errors (std.err=“san.se”), corresponding to xtgee, vce(robust); Stata defaults to conventional model-based standard errors. Scale estimation, small-sample adjustments and correlation estimation may still differ.

Stata reference · R reference

4 more examples and alternatives

binomial GEE with ar1 working correlation

Stata

xtgee binary x, family( binomial ) corr( ar1 )

R geepack

geepack::geeglm(binary ~ x, id = id, waves = time, data = data, family = stats::binomial(), corstr = "ar1")

Differs from Stata Keep rows sorted in contiguous panels. geeglm defaults to sandwich robust standard errors (std.err=“san.se”), corresponding to xtgee, vce(robust); Stata defaults to conventional model-based standard errors. Scale estimation, small-sample adjustments and correlation estimation may still differ.

Stata reference · R reference

poisson GEE with ar1 working correlation

Stata

xtgee count x, family( poisson ) corr( ar1 )

R geepack

geepack::geeglm(count ~ x, id = id, waves = time, data = data, family = stats::poisson(), corstr = "ar1")

Differs from Stata Keep rows sorted in contiguous panels. geeglm defaults to sandwich robust standard errors (std.err=“san.se”), corresponding to xtgee, vce(robust); Stata defaults to conventional model-based standard errors. Scale estimation, small-sample adjustments and correlation estimation may still differ.

Stata reference · R reference

poisson GEE with exchangeable working correlation

Stata

xtgee count x, family( poisson ) corr( exchangeable )

R geepack

geepack::geeglm(count ~ x, id = id, waves = time, data = data, family = stats::poisson(), corstr = "exchangeable")

Differs from Stata Keep rows sorted in contiguous panels. geeglm defaults to sandwich robust standard errors (std.err=“san.se”), corresponding to xtgee, vce(robust); Stata defaults to conventional model-based standard errors. Scale estimation, small-sample adjustments and correlation estimation may still differ.

Stata reference · R reference

poisson GEE with independence working correlation

Stata

xtgee count x, family( poisson ) corr( independence )

R geepack

geepack::geeglm(count ~ x, id = id, waves = time, data = data, family = stats::poisson(), corstr = "independence")

Differs from Stata Keep rows sorted in contiguous panels. geeglm defaults to sandwich robust standard errors (std.err=“san.se”), corresponding to xtgee, vce(robust); Stata defaults to conventional model-based standard errors. Scale estimation, small-sample adjustments and correlation estimation may still differ.

Stata reference · R reference

xtgls

Group-heteroskedastic panel GLS by maximum likelihood

Stata

xtgls y x z, panels(heteroskedastic) igls

R nlme

nlme::gls(y ~ x + z, data = data, weights = nlme::varIdent(form = ~1 | id), method = "ML")

Fixed workflow; adapt the named data columns explicitly. Common slopes with independent residuals and group-specific residual variances; likelihood, variance normalization and convergence criteria can differ.

Stata reference · R reference

xthtaylor

Hausman-Taylor model with time-varying and invariant endogenous covariates

Stata

xthtaylor y x1 x2 a b, endog(x2 b) constant(a b)

R plm

plm::plm(y ~ x1 + x2 + a + b | x1 + a | x2, data = data, index = c("id", "time"), random.method = "ht", model = "random", inst.method = "baltagi")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. The three-part R formula explicitly separates exogenous covariates and time-varying endogenous covariates. Identification of invariant endogenous effects requires sufficient exogenous time-varying instruments; these classifications need substantive justification.

Stata reference · R reference

xtivreg

Fixed-effects instrumental-variable regression

Stata

xtivreg y z (x = instrument), fe

R plm

plm::plm(y ~ x + z | instrument + z, data = pd, model = "within")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Instrument validity must be justified substantively; this fixture only verifies execution.

Stata reference · R reference

xtlogit

Conditional fixed-effects logistic regression

Stata

xtlogit binary x, fe

R survival

{ coxph <- survival::coxph; survival::clogit(binary ~ x + survival::strata(id), data = data) }

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Conditions out group intercepts; groups without outcome variation contribute no coefficient information.

Stata reference · R reference

xtpcse

Beck-Katz panel-corrected covariance

Stata

xtpcse y x z, correlation(independent)

R plm

{ fit <- plm::plm(y ~ x + z, data = pd, model = "pooling"); lmtest::coeftest(fit, vcov. = plm::vcovBK(fit, type = "HC0", cluster = "time")) }

Differs from Stata vcovBK(cluster=“time”) allows contemporaneous cross-panel covariance, as required for PCSE. Balancedness, pairwise missing-data treatment, finite-sample normalization and reference tests can still differ from xtpcse.

Stata reference · R reference

xtpoisson

Random-intercept panel Poisson

Stata

xtpoisson count x, re normal

R lme4

lme4::glmer(count ~ x + (1 | id), data = data, family = stats::poisson(link = "log"), nAGQ = 12)

Differs from Stata Uses normal panel random intercepts and 12-point adaptive quadrature, matching the Stata integration-point default. xtpoisson requires re normal for this distribution; its plain re default is gamma. Quadrature adaptation and convergence criteria remain package-specific.

Stata reference · R reference

xtprobit

Random-intercept panel probit

Stata

xtprobit binary x, re

R lme4

lme4::glmer(binary ~ x + (1 | id), data = data, family = stats::binomial(link = "probit"), nAGQ = 12)

Differs from Stata Uses normal panel random intercepts and 12-point adaptive quadrature, matching Stata intpoints(12). Quadrature adaptation and convergence criteria remain package-specific.

Stata reference · R reference

xtrc

Random-coefficient panel model workflow

Stata

xtrc y x z

R nlme

nlme::lme(y ~ x + z, random = ~ x + z | id, data = data, method = "ML")

All three coefficients (intercept, x and z) vary by panel. nlme fits a Gaussian mixed model by ML; xtrc uses Swamy random-coefficient GLS, so the estimators and covariance estimates differ.

Stata reference · R reference

xtreg

Individual and time fixed effects

Stata

xtreg y x z i.time, fe

R plm

plm::plm(y ~ x + z, data = pd, model = "within", effect = "twoways")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. First-difference and pooled rows are explicit benchmarks; match estimator and effects.

Stata reference · R reference

11 more examples and alternatives

Between-unit regression

Stata

xtreg y x z, be

R plm

plm::plm(y ~ x + z, data = pd, model = "between", effect = "individual")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. First-difference and pooled rows are explicit benchmarks; match estimator and effects.

Stata reference · R reference

Individual fixed effects

Stata

xtreg y x z, fe

R plm

plm::plm(y ~ x + z, data = pd, model = "within", effect = "individual")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. First-difference and pooled rows are explicit benchmarks; match estimator and effects.

Stata reference · R reference

Fit a within-panel fixed-effects regression

Stata

xtreg y x z, fe

R plm

plm::plm(y ~ x + z, data = pd, model = "within", effect = "individual")

Both commands fit individual fixed effects using the within transformation. plm reports the transformed-model coefficient vector without Stata reported intercept; residual degrees of freedom and covariance scaling must be aligned separately.

Stata reference · R reference

Cluster-robust panel standard errors

Stata

xtreg y x z, fe vce(cluster id)

R plm

{ V <- plm::vcovHC(fit, method = "arellano", type = "HC1", cluster = "group"); lmtest::coeftest(fit, vcov. = V) }

Differs from Stata Stata xtreg, fe vce(robust) equals vce(cluster id). plm HC1 uses N/(N-K) with K counting within-model coefficients and coeftest uses residual degrees of freedom; Stata also applies its cluster correction and uses G-1 reference degrees of freedom. These standard errors and p-values are not exact Stata finite-sample parity.

Stata reference · R reference

Cluster-robust within-panel standard errors

Stata

xtreg y x z, fe vce(robust)

R plm

{ V <- plm::vcovHC(fit, method = "arellano", type = "HC1", cluster = "group"); lmtest::coeftest(fit, vcov. = V) }

Differs from Stata Stata xtreg, fe vce(robust) equals vce(cluster id). plm HC1 uses N/(N-K) with K counting within-model coefficients and coeftest uses residual degrees of freedom; Stata also applies its cluster correction and uses G-1 reference degrees of freedom. These standard errors and p-values are not exact Stata finite-sample parity.

Stata reference · R reference

Pooled panel regression benchmark

Stata

xtreg y x z, pa corr(independent)

R plm

plm::plm(y ~ x + z, data = pd, model = "pooling", effect = "individual")

Under independent working correlation, population-averaged Gaussian GEE and pooled OLS have the same coefficients. xtreg, pa defaults to conventional model-based GEE covariance (not robust unless requested); pooled plm uses OLS residual scaling, so reported standard errors can differ.

Stata reference · R reference

Random-effects panel regression

Stata

xtreg y x z, re

R plm

plm::plm(y ~ x + z, data = pd, model = "random", effect = "individual")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. First-difference and pooled rows are explicit benchmarks; match estimator and effects.

Stata reference · R reference

Random effects with amemiya variance components

Stata

xtreg y x z, re

R plm

plm::plm(y ~ x + z, data = pd, model = "random", random.method = "amemiya")

Amemiya is an R alternative variance-component estimator with no corresponding xtreg option. Stata xtreg, re uses its default estimator or sa for Swamy-Arora; the Stata line does not select Amemiya and numerical parity is not claimed.

Stata reference · R reference

Random effects with nerlove variance components

Stata

xtreg y x z, re

R plm

plm::plm(y ~ x + z, data = pd, model = "random", random.method = "nerlove")

Nerlove is an R alternative variance-component estimator with no corresponding xtreg option. Stata xtreg, re uses its default estimator or sa for Swamy-Arora; the Stata line does not select Nerlove and numerical parity is not claimed.

Stata reference · R reference

Random effects with swar variance components

Stata

xtreg y x z, re

R plm

plm::plm(y ~ x + z, data = pd, model = "random", random.method = "swar")

Fixed workflow; adapt the named data columns explicitly. These are named alternative variance-component estimators; only matching methods justify comparing estimates.

Stata reference · R reference

Random effects with walhus variance components

Stata

xtreg y x z, re

R plm

plm::plm(y ~ x + z, data = pd, model = "random", random.method = "walhus")

Walhus is an R alternative variance-component estimator with no corresponding xtreg option. Stata xtreg, re uses its default estimator or sa for Swamy-Arora; the Stata line does not select Walhus and numerical parity is not claimed.

Stata reference · R reference

xtregar

Random intercept with AR(1) residual correlation

Stata

xtregar y x z, re

R nlme

nlme::lme(y ~ x + z, random = ~1 | id, correlation = nlme::corAR1(form = ~time | id), data = data, method = "ML")

Fixed workflow; adapt the named data columns explicitly. Likelihood-based mixed-model workflow differs from xtregar’s estimator.

Stata reference · R reference

xtscc

Driscoll-Kraay covariance

Stata

xtscc y x z, fe lag(2)

R plm

{ V <- plm::vcovSCC(fit, type = "HC1", maxlag = 2); lmtest::coeftest(fit, vcov. = V) }

Differs from Stata Requires the community command xtscc (ssc install xtscc); xtreg has no vce(dkraay) option. Both use a within estimator with lag-2 Driscoll-Kraay covariance; finite-sample scaling and reference degrees of freedom differ.

Stata reference · R reference

xttest0

Breusch-Pagan test of individual random effects

Stata

xtreg y x z, re
xttest0

R plm

plm::plmtest(plm::plm(y ~ x + z, data = pd, model = "pooling"), type = "bp", effect = "individual")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Stata’s boundary-adjusted mixture distribution can differ from plm’s chi-square reference.

Stata reference · R reference

xtunitroot

Panel unit-root test: madwu

Stata

xtunitroot fisher y, dfuller lags(1)

R plm

plm::purtest(pd$y, test = "madwu", exo = "intercept", lags = 1, pmax = 1)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Hadri reverses the null to stationarity; deterministic terms and lag rules must match.

Stata reference · R reference

Panel unit-root test: hadri

Stata

xtunitroot hadri y, robust

R plm

plm::purtest(pd$y, test = "hadri", exo = "intercept", Hcons = TRUE)

Differs from Stata The null is panel stationarity. Hcons=TRUE selects the heteroskedasticity-robust Hadri statistic, corresponding to Stata robust; plm ignores lag selections for Hadri and provides no Bartlett HAC kernel correction.

Stata reference · R reference

Panel unit-root test: ips

Stata

xtunitroot ips y, lags(1)

R plm

plm::purtest(pd$y, test = "ips", exo = "intercept", lags = 1, pmax = 1)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Hadri reverses the null to stationarity; deterministic terms and lag rules must match.

Stata reference · R reference

Panel unit-root test: levinlin

Stata

xtunitroot llc y, lags(1)

R plm

plm::purtest(pd$y, test = "levinlin", exo = "intercept", lags = 1, pmax = 1)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Hadri reverses the null to stationarity; deterministic terms and lag rules must match.

Stata reference · R reference