中文

Causal inference

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

boottest

Null-imposed wild cluster bootstrap using rademacher weights

Stata

boottest treat, cluster(group) reps(199) weighttype(rademacher) seed(128)

R Package:fwildclusterboot

fwildclusterboot::boottest(fit, param = "treat", clustid = "group", B = 199, type = "rademacher", impose_null = TRUE, conf_int = FALSE, engine = "R-lean")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Small replication count is a runnable example, not an analysis recommendation. Align bootstrap variant, clustering and finite-sample correction. Installed fwildclusterboot emits a one-time seeding-change warning; both R and dqrng generators are explicitly seeded. fwildclusterboot 0.13 changed RNG seeding; seed-for-seed agreement with older versions is not expected.

Stata reference · R reference

Null-imposed wild cluster bootstrap using webb weights

Stata

boottest treat, cluster(group) reps(199) weighttype(webb) seed(128)

R Package:fwildclusterboot

fwildclusterboot::boottest(fit, param = "treat", clustid = "group", B = 199, type = "webb", impose_null = TRUE, conf_int = FALSE, engine = "R-lean")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Small replication count is a runnable example, not an analysis recommendation. Align bootstrap variant, clustering and finite-sample correction. Installed fwildclusterboot emits a one-time seeding-change warning; both R and dqrng generators are explicitly seeded. fwildclusterboot 0.13 changed RNG seeding; seed-for-seed agreement with older versions is not expected.

Stata reference · R reference

cem

Coarsened exact matching on a categorical covariate

Stata

cem category, treatment(treat)

R Package:MatchIt

MatchIt::matchit(treat ~ category, data = transform(data, category = as.integer(category) - 1L), method = "cem", estimand = "ATT", cutpoints = list(category = 0))

Fixed workflow; adapt the named data columns explicitly. Matching preprocessing only; categorical levels need no continuous cutpoints. For continuous variables, explicitly match cutpoints and estimand rather than relying on different software defaults.

Stata reference · R reference

did_imputation

Average imputation treatment effect over treated observations

Stata

did_imputation y id time cohort

R Package:stats

{ baseline <- stats::lm(y ~ factor(id) + factor(time) + x, data = data, subset = treated == 0); effects <- data$y - stats::predict(baseline, newdata = data); mean(effects[data$treated == 1]) }

Fixed workflow; adapt the named data columns explicitly. Point-estimation workflow only; untreated-outcome model, weighting and supported treated observations must be chosen explicitly.

Stata reference · R reference

Untreated-outcome imputation effects by event time

Stata

did_imputation y id time cohort, allhorizons

R Package:stats

{ untreated <- data$treated == 0; baseline <- stats::lm(y ~ factor(id) + factor(time) + x, data = data, subset = untreated); effects <- data$y - stats::predict(baseline, newdata = data); keep <- data$treated == 1; stats::aggregate(effects[keep], list(horizon = data$time[keep] - data$cohort[keep]), mean) }

Fixed workflow; adapt the named data columns explicitly. Point-estimation illustration with unit/time effects learned only from untreated observations. Never-treated Stata cohort should be missing, whereas this R fixture uses an out-of-range sentinel. Does not reproduce did_imputation’s influence-function variance, autosample or pretrend tests.

Stata reference · R reference

drdid

Two-period panel doubly robust DiD ATT point estimate

Stata

drdid y x, ivar(id) time(time) treatment(treated) dripw

R Package:stats

{ ps <- stats::fitted(stats::glm(treated ~ x, data = data, family = stats::binomial())); m0 <- stats::predict(stats::lm(dy ~ x, data = data, subset = treated == 0), data); weighted.mean(data$dy - m0, data$treated) - weighted.mean(data$dy - m0, (1 - data$treated) * ps / (1 - ps)) }

Fixed workflow; adapt the named data columns explicitly. Wide-data point-estimation workflow for the dripw variant: logistic propensity plus control-group outcome-change OLS. Stata expects long data. No influence-function variance or improved drimp estimator.

Stata reference · R reference

Panel outcome-regression DiD ATT point estimate

Stata

drdid y x, ivar(id) time(time) treatment(treated) reg

R Package:stats

{ m0 <- stats::predict(stats::lm(dy ~ x, data = data, subset = treated == 0), data); mean((data$dy - m0)[data$treated == 1]) }

Fixed workflow; adapt the named data columns explicitly. Two-period wide-data regression-adjustment workflow; requires conditional parallel trends and no anticipation. No standard errors.

Stata reference · R reference

Stabilized inverse-probability-weighted DiD ATT

Stata

drdid y x, ivar(id) time(time) treatment(treated) stdipw

R Package:stats

{ ps <- stats::fitted(stats::glm(treated ~ x, data = data, family = stats::binomial())); mean(data$dy[data$treated == 1]) - weighted.mean(data$dy, (1 - data$treated) * ps / (1 - ps)) }

Fixed workflow; adapt the named data columns explicitly. Two-period wide-data point estimate; no influence-function variance and no repeated-cross-section implementation.

Stata reference · R reference

ebalance

Entropy balancing treated/control covariate moments through order 1

Stata

ebalance treat x z, targets(1)

R Package:WeightIt

WeightIt::weightit(treat ~ x + z, data = data, method = "ebal", estimand = "ATT", moments = 1)

Fixed workflow; adapt the named data columns explicitly. Preprocessing weights targeting the treated population; normalization, higher moments and optimization tolerances must match. No outcome standard errors.

Stata reference · R reference

Entropy balancing treated/control covariate moments through order 2

Stata

ebalance treat x z, targets(2)

R Package:WeightIt

WeightIt::weightit(treat ~ x + z, data = data, method = "ebal", estimand = "ATT", moments = 2)

Fixed workflow; adapt the named data columns explicitly. Preprocessing weights targeting the treated population; normalization, higher moments and optimization tolerances must match. No outcome standard errors.

Stata reference · R reference

estat

IV weak-instrument and endogeneity diagnostics

Stata

estat endogenous

R Package:AER

summary(AER::ivreg(y ~ x + w | z + w, data = data), diagnostics = TRUE)$diagnostics

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Wu-Hausman tests depend on model assumptions; exactly identified models have no overidentification test.

Stata reference · R reference

1 more examples and alternatives

First-stage instrument relevance regression

Stata

estat firststage

R Package:stats

summary(stats::lm(x ~ z + w, data = data))

Fixed workflow; adapt the named data columns explicitly. This is the full first-stage regression; a joint excluded-instrument F test may require a nested comparison.

Stata reference · R reference

eventstudyinteract

Aggregate cohort-interacted post-treatment effects

Stata

eventstudyinteract y lead3 lead2 lag0 lag1 lag2, absorb(id time) cohort(cohort) control_cohort(never)

R Package:fixest

{ fit <- fixest::feols(y ~ sunab(cohort, time, ref.p = -1) | id + time, data = data, vcov = ~id, nthreads = 1); stats::aggregate(fit, agg = "ATT") }

Fixed workflow; adapt the named data columns explicitly. ATT aggregation is an R postestimation step; match event-time/cohort weights rather than averaging coefficients unweighted.

Stata reference · R reference

Sun-Abraham cohort-interacted event-study estimation

Stata

eventstudyinteract y lead3 lead2 lag0 lag1 lag2, absorb(id time) cohort(cohort) control_cohort(never) covariates(x) vce(cluster id)

R Package:fixest

fixest::feols(y ~ sunab(cohort, time, ref.p = -1) + x | id + time, data = data, vcov = ~id, nthreads = 1)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Uses never-treated reference cohorts and -1 event-time reference. Match supported horizons, cohort-share aggregation and covariance conventions to eventstudyinteract.

Stata reference · R reference

gencohort

Create first-treatment cohorts including never-treated units

Stata

gencohort cohort, group(id) time(time) treat(treated)

R Package:stats

transform(data, cohort = ave(ifelse(treated == 1, time, Inf), id, FUN = function(v) { first <- min(v); rep(if(is.infinite(first)) 0 else first, length(v)) }))

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Uses zero as a never-treated category for cohort-based DiD. An absorbing treatment history and unique unit/time records are prerequisites; imputation estimators may instead require missing never-treatment times.

Stata reference · R reference

ivreghdfe

Instrumental variables with two absorbed effects

Stata

ivreghdfe y z (x = instrument), absorb(id time)

R Package:fixest

fixest::feols(y ~ z | id + time | x ~ instrument, data = data, vcov = "iid", nthreads = 1)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata command. Match singleton handling, covariance correction and absorbed-effect degrees of freedom before comparing inferential output.

Stata reference · R reference

Cluster-robust absorbed-effects IV

Stata

ivreghdfe y z (x = instrument), absorb(id time) cluster(id)

R Package:fixest

fixest::feols(y ~ z | id + time | x ~ instrument, data = data, vcov = ~id, nthreads = 1)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata command. Match singleton handling, covariance correction and absorbed-effect degrees of freedom before comparing inferential output.

Stata reference · R reference

Heteroskedasticity-robust absorbed-effects IV

Stata

ivreghdfe y z (x = instrument), absorb(id time) robust

R Package:fixest

fixest::feols(y ~ z | id + time | x ~ instrument, data = data, vcov = "hetero", nthreads = 1)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata command. Match singleton handling, covariance correction and absorbed-effect degrees of freedom before comparing inferential output.

Stata reference · R reference

Instrumental variables with individual fixed effects

Stata

ivreghdfe y z (x = instrument), absorb(id)

R Package:fixest

fixest::feols(y ~ z | id | x ~ instrument, data = data, vcov = "iid", nthreads = 1)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata command. Match singleton handling, covariance correction and absorbed-effect degrees of freedom before comparing inferential output.

Stata reference · R reference

ivregress

Formula-based instrumental-variable regression

Stata

ivregress 2sls y w (x = z)

R Package:AER

AER::ivreg(y ~ x + w | z + w, data = data)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Separate structural regressors and instruments with the formula bar.

Stata reference · R reference

5 more examples and alternatives

Cluster-robust two-stage least squares

Stata

ivregress 2sls y w (x = z), vce(cluster cluster)

R Package:estimatr

estimatr::iv_robust(y ~ x + w | z + w, data = data, clusters = cluster, se_type = "stata")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Small-cluster reference degrees of freedom are explicit.

Stata reference · R reference

Two-stage least squares with HC2 covariance

Stata

ivregress 2sls y w (x = z), vce(robust)

R Package:estimatr

estimatr::iv_robust(y ~ x + w | z + w, data = data, se_type = "HC2")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. HC corrections differ; estimatr stata aliases HC1 for unclustered IV inference.

Stata reference · R reference

Two-stage least squares with HC3 covariance

Stata

ivregress 2sls y w (x = z), vce(robust)

R Package:estimatr

estimatr::iv_robust(y ~ x + w | z + w, data = data, se_type = "HC3")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. HC corrections differ; estimatr stata aliases HC1 for unclustered IV inference.

Stata reference · R reference

Two-stage least squares with HC1 covariance

Stata

ivregress 2sls y w (x = z), vce(robust) small

R Package:estimatr

estimatr::iv_robust(y ~ x + w | z + w, data = data, se_type = "HC1")

Differs from Stata Stata small requests the N/(N-K) HC1 correction and t-based inference. estimatr se_type=“stata” aliases HC1 for unclustered IV. Without small, ivregress vce(robust) uses HC0 and normal reference inference, so it is a different default.

Stata reference · R reference

Two-stage least squares with stata covariance

Stata

ivregress 2sls y w (x = z), vce(robust) small

R Package:estimatr

estimatr::iv_robust(y ~ x + w | z + w, data = data, se_type = "stata")

Differs from Stata Stata small requests the N/(N-K) HC1 correction and t-based inference. estimatr se_type=“stata” aliases HC1 for unclustered IV. Without small, ivregress vce(robust) uses HC0 and normal reference inference, so it is a different default.

Stata reference · R reference

kmatch

Entropy balancing treated/control covariate moments through order 1

Stata

kmatch eb treat x z (y), att targets(1)

R Package:WeightIt

WeightIt::weightit(treat ~ x + z, data = data, method = "ebal", estimand = "ATT", moments = 1)

Fixed workflow; adapt the named data columns explicitly. Preprocessing weights targeting the treated population; normalization, higher moments and optimization tolerances must match. No outcome standard errors.

Stata reference · R reference

Entropy balancing treated/control covariate moments through order 2

Stata

kmatch eb treat x z (y), att targets(2)

R Package:WeightIt

WeightIt::weightit(treat ~ x + z, data = data, method = "ebal", estimand = "ATT", moments = 2)

Fixed workflow; adapt the named data columns explicitly. Preprocessing weights targeting the treated population; normalization, higher moments and optimization tolerances must match. No outcome standard errors.

Stata reference · R reference

Inverse-probability weighting for the treated population

Stata

kmatch ipw treat x z (y), att

R Package:WeightIt

WeightIt::weightit(treat ~ x + z, data = data, method = "glm", estimand = "ATT")

Fixed workflow; adapt the named data columns explicitly. Weights only; outcome contrasts and propensity-estimation variance require a separate estimator.

Stata reference · R reference

Multivariate Mahalanobis matching

Stata

kmatch md treat x z (y), att nn(1)

R Package:MatchIt

MatchIt::matchit(treat ~ x + z, data = data, method = "nearest", distance = "mahalanobis", replace = TRUE, estimand = "ATT")

Fixed workflow; adapt the named data columns explicitly. Explicit nearest-neighbor alternative to kmatch’s kernel defaults; matching preprocessing only.

Stata reference · R reference

Propensity-score nearest-neighbor matching

Stata

kmatch ps treat x z (y), att nn(1)

R Package:MatchIt

MatchIt::matchit(treat ~ x + z, data = data, method = "nearest", distance = "glm", replace = TRUE, estimand = "ATT")

Fixed workflow; adapt the named data columns explicitly. Explicit nearest-neighbor alternative to kmatch’s kernel defaults; matching preprocessing only.

Stata reference · R reference

1 more examples and alternatives

Exact matching on categorical covariates

Stata

kmatch em treat category (y), att

R Package:MatchIt

MatchIt::matchit(treat ~ category, data = data, method = "exact", estimand = "ATT")

Fixed workflow; adapt the named data columns explicitly. Exact-match preprocessing; estimand and stratum normalization are explicit.

Stata reference · R reference

psmatch2

Matched ATT point contrast after propensity-score preprocessing

Stata

psmatch2 treat x z, outcome(y) logit neighbor(1)

R Package:MatchIt

{ match <- MatchIt::matchit(treat ~ x + z, data = data, method = "nearest", distance = "glm", replace = TRUE, estimand = "ATT"); matched <- MatchIt::match_data(match); with(matched, weighted.mean(y[treat == 1], weights[treat == 1]) - weighted.mean(y[treat == 0], weights[treat == 0])) }

Fixed workflow; adapt the named data columns explicitly. Point contrast only; variance is not estimated, and matching ties/order may differ.

Stata reference · R reference

Propensity-score nearest matching with replacement

Stata

psmatch2 treat x z, outcome(y) logit neighbor(1) 

R Package:MatchIt

MatchIt::matchit(treat ~ x + z, data = data, method = "nearest", distance = "glm", link = "logit", replace = TRUE, estimand = "ATT")

Fixed workflow; adapt the named data columns explicitly. Returns matching preprocessing, not psmatch2’s outcome estimate or variance. Replacement, tie-breaking and match order must be explicit.

Stata reference · R reference

Propensity-score matching with an absolute caliper

Stata

psmatch2 treat x z, outcome(y) logit neighbor(1) caliper(.1)

R Package:MatchIt

MatchIt::matchit(treat ~ x + z, data = data, method = "nearest", distance = "glm", link = "logit", replace = TRUE, caliper = .1, std.caliper = FALSE, estimand = "ATT")

Fixed workflow; adapt the named data columns explicitly. Absolute probability-scale caliper; MatchIt defaults to standardized calipers unless explicitly disabled. Outcome estimation remains separate.

Stata reference · R reference

Propensity-score nearest matching without replacement

Stata

psmatch2 treat x z, outcome(y) logit neighbor(1) noreplacement

R Package:MatchIt

MatchIt::matchit(treat ~ x + z, data = data, method = "nearest", distance = "glm", link = "logit", replace = FALSE, estimand = "ATT")

Fixed workflow; adapt the named data columns explicitly. Returns matching preprocessing, not psmatch2’s outcome estimate or variance. Replacement, tie-breaking and match order must be explicit.

Stata reference · R reference

Mahalanobis matching with replacement

Stata

psmatch2 treat, outcome(y) mahalanobis(x z) neighbor(1)

R Package:MatchIt

MatchIt::matchit(treat ~ x + z, data = data, method = "nearest", distance = "mahalanobis", replace = TRUE, estimand = "ATT")

Fixed workflow; adapt the named data columns explicitly. Matching preprocessing; covariance normalization and ties may differ from psmatch2.

Stata reference · R reference

pstest

Standardized covariate balance before and after matching

Stata

pstest x z, both

R Package:cobalt

cobalt::bal.tab(match, un = TRUE)

Fixed workflow; adapt the named data columns explicitly. Companion command supplied with psmatch2. Choose the same standardization denominator; cobalt’s balance table does not reproduce every pstest t-test or omnibus summary.

Stata reference · R reference

rdbwdensity

Bandwidth selection for density discontinuity

Stata

rdbwdensity x

R Package:rddensity

rddensity::rdbwdensity(x)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Bandwidths optimize density-estimation criteria, not treatment-effect estimation.

Stata reference · R reference

rdbwselect

RD bandwidth selection: cerrd

Stata

rdbwselect y x, bwselect(cerrd)

R Package:rdrobust

rdrobust::rdbwselect(y, x, bwselect = "cerrd")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Official rdrobust companion command; bandwidth selectors target different criteria.

Stata reference · R reference

RD bandwidth selection: certwo

Stata

rdbwselect y x, bwselect(certwo)

R Package:rdrobust

rdrobust::rdbwselect(y, x, bwselect = "certwo")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Official rdrobust companion command; bandwidth selectors target different criteria.

Stata reference · R reference

RD bandwidth selection: mserd

Stata

rdbwselect y x, bwselect(mserd)

R Package:rdrobust

rdrobust::rdbwselect(y, x, bwselect = "mserd")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Official rdrobust companion command; bandwidth selectors target different criteria.

Stata reference · R reference

RD bandwidth selection: msetwo

Stata

rdbwselect y x, bwselect(msetwo)

R Package:rdrobust

rdrobust::rdbwselect(y, x, bwselect = "msetwo")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Official rdrobust companion command; bandwidth selectors target different criteria.

Stata reference · R reference

rddensity

Density continuity at the treatment threshold

Stata

rddensity x

R Package:rddensity

rddensity::rddensity(x)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Tests density continuity, not the RD identification assumptions as a whole.

Stata reference · R reference

Density continuity with fixed bandwidths

Stata

rddensity x, h(.4 .4)

R Package:rddensity

rddensity::rddensity(x, h = c(.4, .4))

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Tests density continuity, not the RD identification assumptions as a whole.

Stata reference · R reference

Cubic local-polynomial density discontinuity

Stata

rddensity x, p(3)

R Package:rddensity

rddensity::rddensity(x, p = 3)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Tests density continuity, not the RD identification assumptions as a whole.

Stata reference · R reference

rdrobust

Sharp local-linear RD

Stata

rdrobust y x

R Package:rdrobust

rdrobust::rdrobust(y, x)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata rdrobust command. Match version, bandwidth, kernel and variance settings; no finite-fixture cross-software parity claim.

Stata reference · R reference

10 more examples and alternatives

Coverage-error-optimal RD bandwidth

Stata

rdrobust y x, bwselect(cerrd)

R Package:rdrobust

rdrobust::rdrobust(y, x, bwselect = "cerrd")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata rdrobust command. Match version, bandwidth, kernel and variance settings; no finite-fixture cross-software parity claim.

Stata reference · R reference

Separate MSE-optimal bandwidths by side

Stata

rdrobust y x, bwselect(msetwo)

R Package:rdrobust

rdrobust::rdrobust(y, x, bwselect = "msetwo")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata rdrobust command. Match version, bandwidth, kernel and variance settings; no finite-fixture cross-software parity claim.

Stata reference · R reference

Cluster-robust RD inference

Stata

rdrobust y x, cluster(cluster)

R Package:rdrobust

rdrobust::rdrobust(y, x, cluster = cluster)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata rdrobust command. Match version, bandwidth, kernel and variance settings; no finite-fixture cross-software parity claim.

Stata reference · R reference

Covariate-adjusted RD

Stata

rdrobust y x, covs(z)

R Package:rdrobust

rdrobust::rdrobust(y, x, covs = z)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata rdrobust command. Match version, bandwidth, kernel and variance settings; no finite-fixture cross-software parity claim.

Stata reference · R reference

Fuzzy regression discontinuity

Stata

rdrobust y x, fuzzy(treat)

R Package:rdrobust

rdrobust::rdrobust(y, x, fuzzy = treat)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Requires an exclusion restriction, a treatment-probability discontinuity and a suitable monotonicity assumption.

Stata reference · R reference

RD with fixed symmetric bandwidth

Stata

rdrobust y x, h(.4)

R Package:rdrobust

rdrobust::rdrobust(y, x, h = .4)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata rdrobust command. Match version, bandwidth, kernel and variance settings; no finite-fixture cross-software parity claim.

Stata reference · R reference

RD with Epanechnikov kernel

Stata

rdrobust y x, kernel(epanechnikov)

R Package:rdrobust

rdrobust::rdrobust(y, x, kernel = "epanechnikov")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata rdrobust command. Match version, bandwidth, kernel and variance settings; no finite-fixture cross-software parity claim.

Stata reference · R reference

RD with uniform kernel

Stata

rdrobust y x, kernel(uniform)

R Package:rdrobust

rdrobust::rdrobust(y, x, kernel = "uniform")

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata rdrobust command. Match version, bandwidth, kernel and variance settings; no finite-fixture cross-software parity claim.

Stata reference · R reference

RD with ninety-percent confidence intervals

Stata

rdrobust y x, level(90)

R Package:rdrobust

rdrobust::rdrobust(y, x, level = 90)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata rdrobust command. Match version, bandwidth, kernel and variance settings; no finite-fixture cross-software parity claim.

Stata reference · R reference

Local-quadratic RD

Stata

rdrobust y x, p(2)

R Package:rdrobust

rdrobust::rdrobust(y, x, p = 2)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. User-written Stata rdrobust command. Match version, bandwidth, kernel and variance settings; no finite-fixture cross-software parity claim.

Stata reference · R reference

tebalance

Covariate balance before and after weighting

Stata

tebalance summarize

R Package:cobalt

cobalt::bal.tab(w, un = TRUE)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Standardization denominator and binary-variable scaling must be aligned.

Stata reference · R reference

teffects

Augmented inverse-probability ATE point estimate

Stata

teffects aipw (y x z) (treat x z)

R Package:stats

{ ps <- stats::fitted(stats::glm(treat ~ x + z, data, family = stats::binomial())); m0 <- stats::predict(stats::lm(y ~ x + z, data, subset = treat == 0), data); m1 <- stats::predict(stats::lm(y ~ x + z, data, subset = treat == 1), data); mean(m1 - m0 + data$treat * (data$y - m1) / ps - (1 - data$treat) * (data$y - m0) / (1 - ps)) }

Differs from Stata Fixed workflow; adapt the named data columns explicitly. No standard errors; double robustness still requires overlap and identifying assumptions.

Stata reference · R reference

13 more examples and alternatives

Propensity-score weights targeting ATC

Stata

teffects ipw (y) (treat x z) 

R Package:WeightIt

WeightIt::weightit(treat ~ x + z, data = data, method = "glm", estimand = "ATC")

Fixed workflow; adapt the named data columns explicitly. Computes weights only, not treatment effects or standard errors. ATO and ATC are alternative target populations, not Stata teffects defaults.

Stata reference · R reference

Propensity-score weights targeting ATE

Stata

teffects ipw (y) (treat x z) 

R Package:WeightIt

WeightIt::weightit(treat ~ x + z, data = data, method = "glm", estimand = "ATE")

Fixed workflow; adapt the named data columns explicitly. Computes weights only, not treatment effects or standard errors. ATO and ATC are alternative target populations, not Stata teffects defaults.

Stata reference · R reference

Propensity-score weights targeting ATO

Stata

teffects ipw (y) (treat x z) 

R Package:WeightIt

WeightIt::weightit(treat ~ x + z, data = data, method = "glm", estimand = "ATO")

Fixed workflow; adapt the named data columns explicitly. Computes weights only, not treatment effects or standard errors. ATO and ATC are alternative target populations, not Stata teffects defaults.

Stata reference · R reference

Normalized IPW point effect for ATE

Stata

teffects ipw (y) (treat x z) 

R Package:WeightIt

{ w <- WeightIt::weightit(treat ~ x + z, data = data, method = "glm", estimand = "ATE")$weights; weighted.mean(data$y[data$treat == 1], w[data$treat == 1]) - weighted.mean(data$y[data$treat == 0], w[data$treat == 0]) }

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Point estimate only; outcome-free weighting assumes conditional exchangeability and overlap. Weight-estimation uncertainty is not quantified.

Stata reference · R reference

Propensity-score weights targeting ATT

Stata

teffects ipw (y) (treat x z) , atet

R Package:WeightIt

WeightIt::weightit(treat ~ x + z, data = data, method = "glm", estimand = "ATT")

Fixed workflow; adapt the named data columns explicitly. Computes weights only, not treatment effects or standard errors. ATO and ATC are alternative target populations, not Stata teffects defaults.

Stata reference · R reference

Normalized IPW point effect for ATT

Stata

teffects ipw (y) (treat x z) , atet

R Package:WeightIt

{ w <- WeightIt::weightit(treat ~ x + z, data = data, method = "glm", estimand = "ATT")$weights; weighted.mean(data$y[data$treat == 1], w[data$treat == 1]) - weighted.mean(data$y[data$treat == 0], w[data$treat == 0]) }

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Point estimate only; outcome-free weighting assumes conditional exchangeability and overlap. Weight-estimation uncertainty is not quantified.

Stata reference · R reference

Inverse-probability-weighted outcome-regression ATE

Stata

teffects ipwra (y x z) (treat x z)

R Package:stats

{ ps <- stats::fitted(stats::glm(treat ~ x + z, data, family = stats::binomial())); data$w <- ifelse(data$treat == 1, 1 / ps, 1 / (1 - ps)); m0 <- stats::predict(stats::lm(y ~ x + z, data, subset = treat == 0, weights = w), data); m1 <- stats::predict(stats::lm(y ~ x + z, data, subset = treat == 1, weights = w), data); mean(m1 - m0) }

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Point estimation only; Stata jointly computes the estimating-equation variance.

Stata reference · R reference

Mahalanobis nearest-neighbor preprocessing

Stata

teffects nnmatch (y x z) (treat), atet

R Package:MatchIt

MatchIt::matchit(treat ~ x + z, data = data, method = "nearest", distance = "mahalanobis", replace = TRUE, estimand = "ATT")

Matches with replacement, as teffects does. MatchIt returns preprocessing weights and matches, not the treatment-effect estimate; tie handling, distance conventions, bias correction and Abadie-Imbens standard errors require separate alignment.

Stata reference · R reference

Matching with replacement

Stata

teffects psmatch (y) (treat x z), nneighbor(1) atet

R Package:MatchIt

MatchIt::matchit(treat ~ x + z, data = data, method = "nearest", distance = "glm", ratio = 1, replace = TRUE, estimand = "ATT")

Fixed workflow; adapt the named data columns explicitly. Preprocessing only; MatchIt does not reproduce teffects matching bias correction or Abadie-Imbens standard errors.

Stata reference · R reference

One-to-one nearest propensity-score matching

Stata

teffects psmatch (y) (treat x z), nneighbor(1) atet

R Package:MatchIt

MatchIt::matchit(treat ~ x + z, data = data, method = "nearest", distance = "glm", ratio = 1, replace = TRUE, estimand = "ATT")

Matches with replacement, as teffects does. MatchIt returns preprocessing weights and matches, not the treatment-effect estimate; tie handling, distance conventions, bias correction and Abadie-Imbens standard errors require separate alignment.

Stata reference · R reference

One-to-two nearest propensity-score matching

Stata

teffects psmatch (y) (treat x z), nneighbor(2) atet

R Package:MatchIt

MatchIt::matchit(treat ~ x + z, data = data, method = "nearest", distance = "glm", ratio = 2, replace = TRUE, estimand = "ATT")

Matches with replacement, as teffects does. MatchIt returns preprocessing weights and matches, not the treatment-effect estimate; tie handling, distance conventions, bias correction and Abadie-Imbens standard errors require separate alignment.

Stata reference · R reference

Regression-adjustment average treatment effect

Stata

teffects ra (y x z) (treat)

R Package:stats

{ fit0 <- stats::lm(y ~ x + z, data, subset = treat == 0); fit1 <- stats::lm(y ~ x + z, data, subset = treat == 1); mean(stats::predict(fit1, data) - stats::predict(fit0, data)) }

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Point-estimate g-computation under correctly specified potential-outcome models; no causal identification is established by fitting alone.

Stata reference · R reference

Regression-adjustment effect among treated units

Stata

teffects ra (y x z) (treat), atet

R Package:stats

{ fit0 <- stats::lm(y ~ x + z, data, subset = treat == 0); fit1 <- stats::lm(y ~ x + z, data, subset = treat == 1); treated <- data[data$treat == 1, ]; mean(stats::predict(fit1, treated) - stats::predict(fit0, treated)) }

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Averages counterfactual differences over the treated covariate distribution.

Stata reference · R reference

teoverlap

Propensity-score overlap summary by treatment

Stata

teoverlap

R Package:stats

{ ps <- stats::fitted(stats::glm(treat ~ x + z, data, family = stats::binomial())); lapply(split(ps, data$treat), function(v) stats::quantile(v, c(0, .05, .5, .95, 1))) }

Fixed workflow; adapt the named data columns explicitly. Numeric overlap diagnostic; no density plot or automatic trimming.

Stata reference · R reference

xtdidregress

Panel difference-in-differences with unit and period fixed effects

Stata

xtdidregress (y x) (treated), group(id) time(time)

R Package:fixest

fixest::feols(y ~ treated + x | id + time, data = data, vcov = ~id, nthreads = 1)

Differs from Stata Fixed workflow; adapt the named data columns explicitly. Homogeneous-effect two-group design with parallel trends and no anticipation. Under staggered timing and heterogeneous effects, TWFE is not a replacement for modern cohort-specific estimators; covariance finite-sample conventions can differ.

Stata reference · R reference