{smcl} {* *! version 0.1.0 03nov2026}{...} {vieweralsosee "staggeredgmm" "help staggeredgmm"}{...} {viewerjumpto "Postestimation commands" "staggeredgmm_postestimation##description"}{...} {viewerjumpto "estat jtest" "staggeredgmm_postestimation##jtest"}{...} {viewerjumpto "estat aggregate" "staggeredgmm_postestimation##aggregate"}{...} {viewerjumpto "Examples" "staggeredgmm_postestimation##examples"}{...} {viewerjumpto "Stored results" "staggeredgmm_postestimation##results"}{...} {title:Title} {phang} {bf:staggeredgmm postestimation} {hline 2} Postestimation tools for staggeredgmm {marker description}{...} {title:Postestimation commands} {pstd} The following postestimation commands are of special interest after {cmd:staggeredgmm}: {synoptset 20}{...} {p2coldent:Command}Description{p_end} {synoptline} {synopt:{helpb staggeredgmm_postestimation##jtest:estat jtest}}specification test of unconditional parallel trends and no anticipation{p_end} {synopt:{helpb staggeredgmm_postestimation##aggregate:estat aggregate}}aggregate ATTs, optionally posted to {cmd:e(b)} for {cmd:esttab} or {cmd:outreg2}{p_end} {synoptline} {pstd} The following standard postestimation commands are also available: {synoptset 20}{...} {p2coldent:Command}Description{p_end} {synoptline} {synopt:{helpb estat vce}}variance-covariance matrix of the estimators{p_end} {synopt:{helpb estimates}}cataloging estimation results{p_end} {synopt:{helpb lincom}}point estimates, standard errors, testing, and inference for linear combinations of coefficients{p_end} {synopt:{helpb nlcom}}point estimates, standard errors, testing, and inference for nonlinear combinations of coefficients{p_end} {synopt:{helpb test}}Wald tests of simple and composite linear hypotheses{p_end} {synoptline} {marker jtest}{...} {title:Syntax for estat jtest} {p 8 17 2} {cmd:estat} {cmd:jtest} [{cmd:,} {opt full} {opt w:indow(#)}] {synoptset 20}{...} {synopthdr} {synoptline} {synopt:{opt full}}use every pre-treatment placebo moment{p_end} {synopt:{opt w:indow(#)}}width of the local pre-window, in periods; default {cmd:3}; ignored with {opt full}{p_end} {synoptline} {title:Description of estat jtest} {pstd} {cmd:estat jtest} is a serial-correlation robust joint test of the identifying assumptions, following Section 4.4 of Arora and Bijani (2026). It is built on the never-treated event-study basis: fixing a base pre-period t0, each treated cohort g and period t != t0 defines {p 8 8 2} D(g,t) = (Ybar(g,t) - Ybar(g,t0)) - (Ybar(inf,t) - Ybar(inf,t0)), {pstd} where inf denotes the never-treated group. Those with t >= g identify the cohort-by-time effects. Those with t < g are placebo restrictions whose expectation is zero only under parallel trends and no anticipation. Collecting J placebo moments into p with covariance Omega_p implied by the fitted serial-correlation model, the statistic p' inv(Omega_p) p is asymptotically chi-squared with J degrees of freedom under the null. Because the placebo moments do not involve the treatment effects, no degrees of freedom are subtracted for estimated parameters. If Omega_p is rank deficient, a truncated inverse is used and the degrees of freedom are reduced accordingly, with a warning. {pstd} The base period is t0 = g_min - 1, the last period before any cohort adopts, so that every moment has an untreated reference. When the only treated cohort adopts in the second observed period (for example, in a two-period panel), the base period is its only pre-treatment period, so no placebo restriction exists and {cmd:estat jtest} exits with an error. {pstd} A large number of restrictions degrades the chi-squared approximation in small panels. By default the test therefore uses only the placebo moments within {opt window()} periods before each cohort's adoption, which holds its nominal size across sample sizes at essentially no cost in power, since a differential trend or an anticipation effect is already visible just before treatment. {opt full} uses every placebo moment and is well sized only when the number of units is comfortably large. {pstd} The test requires a never-treated control group. {pstd} {bf:Important: the test is unconditional.} The placebo moments are unadjusted group means, whether or not the fitted model adjusts for covariates. After a covariate-adjusted fit, {cmd:estat jtest} therefore tests {bf:unconditional} parallel trends and no anticipation; it does {bf:not} test the covariate-conditional parallel trends assumption under which the effects were estimated. If covariates are distributed differently across cohorts, the test can reject although conditional parallel trends hold, and a non-rejection does not validate the conditional assumption. When the model includes covariates, {cmd:estat jtest} prints this warning below the result, and {cmd:r(scope)} is {cmd:unconditional}. {marker aggregate}{...} {title:Syntax for estat aggregate} {p 8 17 2} {cmd:estat} {cmdab:agg:regate} [{cmd:,} {opt post} {opt l:evel(#)}] {synoptset 20}{...} {synopthdr} {synoptline} {synopt:{opt post}}replace {cmd:e(b)} and {cmd:e(V)} with the aggregate ATTs and their covariance{p_end} {synopt:{opt l:evel(#)}}set confidence level; default is {cmd:level(95)}{p_end} {synoptline} {title:Description of estat aggregate} {pstd} {cmd:estat aggregate} reports the two aggregate ATTs, {cmd:att_cw} (the treated-observation ATT) and {cmd:att_ew} (the cohort-equal ATT), as a coefficient table with their joint covariance, which is computed exactly from the covariance of the cohort-by-time effects. When some cohort-by-time effects are not identified, both aggregates are taken over the identified effects only, with the weights renormalised to sum to one, and the output reports the share of the original weight they cover. {pstd} With {opt post}, {cmd:e(b)} and {cmd:e(V)} are replaced by the two aggregates and their covariance, so that {cmd:esttab}, {cmd:outreg2}, {cmd:lincom}, {cmd:test} and {cmd:estimates store} treat them like any other coefficients. All other results of {cmd:staggeredgmm} are kept, {cmd:e(subcmd)} is set to {cmd:aggregate}, and {cmd:estat jtest} still works. To return to the cohort-by-time effects, rerun {cmd:staggeredgmm} or restore stored estimates. {marker examples}{...} {title:Examples} {phang2}{cmd:. sysuse sgmm_sim_panel, clear}{p_end} {phang2}{cmd:. staggeredgmm y, ivar(unit_id) time(year) gvar(cohort)}{p_end} {phang2}{cmd:. estat jtest}{p_end} {phang2}{cmd:. estat jtest, window(5)}{p_end} {phang2}{cmd:. estat jtest, full}{p_end} {phang2}{cmd:. matrix list r(moments)}{p_end} {pstd}Exporting the aggregate ATTs and the cohort-by-time effects{p_end} {phang2}{cmd:. staggeredgmm y, ivar(unit_id) time(year) gvar(cohort)}{p_end} {phang2}{cmd:. estimates store catt}{p_end} {phang2}{cmd:. estat aggregate, post}{p_end} {phang2}{cmd:. estimates store att}{p_end} {phang2}{cmd:. esttab att, se}{p_end} {phang2}{cmd:. esttab catt, se}{p_end} {marker results}{...} {title:Stored results} {pstd} {cmd:estat jtest} stores the following in {cmd:r()}: {synoptset 20 tabbed}{...} {p2col 5 20 24 2: Scalars}{p_end} {synopt:{cmd:r(J)}}test statistic{p_end} {synopt:{cmd:r(df)}}degrees of freedom{p_end} {synopt:{cmd:r(p)}}p-value{p_end} {synopt:{cmd:r(base_period)}}base period t0{p_end} {synopt:{cmd:r(n_moments)}}number of placebo moments{p_end} {synopt:{cmd:r(n_dropped)}}placebo moments dropped for referencing an unobserved group-period{p_end} {synopt:{cmd:r(window)}}width of the local pre-window (local test only){p_end} {p2col 5 20 24 2: Macros}{p_end} {synopt:{cmd:r(type)}}{cmd:local} or {cmd:full}{p_end} {synopt:{cmd:r(scope)}}{cmd:unconditional}: the placebo moments are never covariate-adjusted{p_end} {synopt:{cmd:r(weighting)}}weighting scheme of the fitted model{p_end} {p2col 5 20 24 2: Matrices}{p_end} {synopt:{cmd:r(moments)}}placebo moments: {cmd:g}, {cmd:t}, {cmd:event_time}, {cmd:moment}{p_end} {pstd} {cmd:estat aggregate} stores the following in {cmd:r()}: {synoptset 20 tabbed}{...} {p2col 5 20 24 2: Scalars}{p_end} {synopt:{cmd:r(share_cw)}}share of treated-observation weight on identified effects{p_end} {synopt:{cmd:r(share_ew)}}share of cohort-equal weight on identified effects{p_end} {p2col 5 20 24 2: Matrices}{p_end} {synopt:{cmd:r(b)}}aggregate ATTs, {cmd:att_cw} and {cmd:att_ew}{p_end} {synopt:{cmd:r(V)}}their covariance matrix{p_end} {pstd} With {opt post}, {cmd:e(b)} and {cmd:e(V)} hold the same two aggregates and {cmd:e(subcmd)} is {cmd:aggregate}. {title:Reference} {phang} Arora, P. and R. Bijani. 2026. Estimating treatment effects under staggered timing and non-spherical errors. Available at SSRN, {browse "https://doi.org/10.2139/ssrn.6558759":doi:10.2139/ssrn.6558759}. {p_end}