@jmwooldridge.bsky.social
Econometrics professor and author. Dogs = 2, cats >= 10.
@jmwooldridge.bsky.socialDon't tempt us!
@jmwooldridge.bsky.socialNice! 😂
@jmwooldridge.bsky.socialGreece with Scott, Pedro, and me? What’s more fun than that?
@jmwooldridge.bsky.socialI've stolen money once or twice living in a crypto paradise.
@jmwooldridge.bsky.socialSeems like I should've been able to answer at least the first question in a few lines, but it didn't turn out that way ....
@jmwooldridge.bsky.socialWhen your poster bombs at ASSA.
@jmwooldridge.bsky.socialI've been teaching this approach for awhile, and it's nice to finally have written it up for a special issue of Econometric Reviews. It's under review there.
@jmwooldridge.bsky.socialI think an important feature is that, when applying the assumptions in the binary treatment case, they collapse to the same ones I used in my earlier TWFE-TW Mundlak paper. In other words, the only extra assumptions are imposed because of the nonbinary nature of the treatment.
@jmwooldridge.bsky.socialIn principle, the treatment can vary from its initial level, and even turn off. But the selection restriction becomes less plausible. I also show how to do an event study (leads and lags) version of the estimator, and how to incorporate heterogeneous trends when CPT is violated.
@jmwooldridge.bsky.socialCallaway, Goodman-Bacon, Sant'Anna (2026). Using flexible TWFE allows adding covariates, which allows both violation of the usual PT and allows for selection on X -- like CG-BS'A strong PT conditional on X. Putting in flexible functions of continuous treatments is easy.
@jmwooldridge.bsky.socialIn addition to additivity in the unit and time FEs -- the usual (conditional) parallel trends assumption -- a selection restriction is needed on the heterogeneous slopes. In the T = 2 case without covariates, it essentially reduces to the "strong" PT assumption in
@jmwooldridge.bsky.socialThe treatment can be continuous, discrete, multiple levels, and so on. Estimation is to take the binary time-varying treatment, W_it, from my extended TWFE approach and replace W_it with any other treatment variable. The resulting equation can be estimated by TWFE.
@jmwooldridge.bsky.socialIn case anyone is interested, here is a new working paper on non-binary treatments with staggered interventions. Essentially, I show under what assumptions it is valid to estimate cohort x time dose-response functions.
@jmwooldridge.bsky.socialHi All. Perhaps something useful from me. I'd like to announce that drlate, a Stata module for doubly robust estimation of the local average treatment effect (LATE) and the local average treatment effect on the treated (LATT), is now available in SSC.
