#econtwitter: Want to produce event-study graph like this one, using estimators robust to heterogeneous treatment effects, across units or over time? You may want to check our new did_multipleGT #Stata package. "ssc install did_multipleGT" to install package and help file. 1/n 
did_multipleGT can be used in DID designs with multiple groups and periods, where all units in same group and period have same treatment (sharp designs). E.g.: treatment is county- or state-level variable. Treatment does not have to be binary. 2/n
By default, did_multipleGT computes Wald-TC estimator of instantaneous treatment effect at the time when switchers switch introduced in Section 3.3 of papers.ssrn.com/sol3/papers.cf… 3/n
If placebo(#) option specified, did_multipleGT also computes placebo estimators, to assess plausibility of parallel (//) trends assumption underlying Wald-TC: were switchers and non switchers on // trends before switch? (see Section 3.3 of papers.ssrn.com/sol3/papers.cf…) 4/n
Finally, in staggered adoption designs where groups' treatment is weakly increasing with time, did_multipleGT computes Wald-TC estimators of dynamic treatement effects among switchers (see Section 5.2 of papers.ssrn.com/sol3/papers.cf…), if dynamic(#) option specified 5/n
Ex: "did_multipleGT Y G T D, placebo(2) dynamic(1) breps(50) cluster(G)". Y outcome, G group (e.g. county), T time, D treatment. Computes 2 placebos, instantaneous effect when switchers switch, and dynamic effect 1 period after switch. Ses estimated by clustered bootstrap 6/n
did_multipleGT stores everything it estimates into eclass objects. Type "ereturn list" to see results, did_multipleGT does not return results table. If breps option specified, did_multipleGT returns graph with estimated effects and placebos, and their 95% confidence intervals 7/n
did_multipleGT uses the fuzzydid package, so "ssc install fuzzydid" before using did_multipleGT. did_multipleGT may take some time to run when controls added to the estimation, and when many dynamic effects estimated. Also, interval between consecutive time periods should = 1 8/n
This is a beta version, please please please report any bug/issue or any option you would like us to add. And please please please send along the nice event-study graphs you will produce using the command! 9/n
Metrics background to conclude: we show in papers.ssrn.com/sol3/papers.cf… that static twoway FE regression not robust to heterogeneous treatment effect, and Abraham and Sun (economics.mit.edu/files/14964) show that same applies to event-study regression with lags and leads of treatment 10/n
Estimators computed by did_multipleGT solve this problem. Other estimators solving this problem include those proposed by Abraham and Sun and Callaway and @pedrohcgs (papers.ssrn.com/sol3/papers.cf…), we discuss connection between estimators in Section 5.2 of papers.ssrn.com/sol3/papers.cf… n/n
