Jeffrey Wooldridge

They can’t even be charted.

I am excited to be hiring two post-doctoral positions at UCLA--please share with your networks. www.aeaweb.org/joe/listing.....

Update from KSE YouTube video by Tymofiy Mylovanov

#EconSky: Tymofiy Mylovanov from the Kyiv School of Economics is a simply a hero.

While Trump is busy allying with Putin, Tymofiy is out in the cold working for his students and his country.

I will match all donations to KSE, up to $1K total (link below).

www.youtube.com/shorts/RgDXI...

I'll likely be submitting the paper (when finished) to a special issue of Econometric Reviews in honor of James MacKinnon this summer.

It's a natural extension from the binary case. Allow effects to vary by cohort and calendar time, then aggregate. Include covariates flexibly; estimating the resulting equation by TWFE and cluster. You can see the heterogeneous trends and moderating effects, too.

Hi Laura. Here's the link to my shared Dropbox. It's slides and some Stata code from a keynote. The setting is not quite as general, but easy to extend. I like transparent regression methods. On slide 27 you can see the estimating equation.

www.dropbox.com/scl/fo/dt2zn...

Today a student raised the bar for what’s required to visit me outside of my office hours.

Well, as the title suggests, you’re after the long run effects. 😬

In case anyone is interested, here is my latest contribution to DiD. In this case, nonlinear models with repeated cross sections. A (short) paper will be ready soon. It's nothing path breaking but shows the obvious extensions from the panel data case works.

www.dropbox.com/scl/fo/lne2y...

Which is then followed by, "What are the alternatives?" For DiD, we now know there's a set of assumptions -- no anticipation and conditional parallel trends -- under which TWFE estimation of a flexible equation consistently estimates well-defined causal effects. So, what consistently beats it?

TWFE is not perfect, of course. But when selection is largely based on time-constant unobservables, it can still be very effective. We have to use a flexible enough model, but that’s on us — not on the estimator. We can allow lots of heterogeneity. Plus, we sometimes can combine with IV.

I have to figure out where this fits into my own New Year’s resolution. 🤔

Registration closes at 5 pm EST today!

Still time to sign up! Register on or before Dec 8. The group of attendees is shaping up nicely, and I'm looking forward to the online interactions.

@msuecon.bsky.social
@imbernomics.bsky.social

When to use negative binomial and Poisson regression When would one use a negative binomial regression and when would one use Poisson regression with respect to the mean and variance?

And here is a much better answer (from
@noahgreifer.bsky.social al, citing @jmwooldridge.bsky.social) than my mumbling in response to the Poisson vs negbin question.

stats.stackexchange.com/questions/65...

Hi! I’m Mary and I’m on the #EconJobMarket this year.

Extreme heat doesn’t just affect students, it affects the people teaching them.

JMP 🧵:

I was in New Orleans recently at a conference and I tried to reach out to Big Freedia. Apparently, we're not as tight as I thought. 😬

The times are not currently listed on the website: 10 am to 4 pm EST on both days. After each 90 minute lecture, a 30 minute Q&A.

Hope to see you then!

The next online installment of ESTIMATE: The Reduced Form is coming on Dec 11-12. I've continued to unify and expand regression-based methods to apply to exit, non-binary treatments, DDD, discrete outcomes, and more.

All proceeds to the MSU economics PhD program.

econ.msu.edu/academics/es...

It was on my mind because the batteries in my laser pointer died just before a presentation. 😂

Did you find any AAA batteries?

🚨 We are hiring! ‼️‼️

tenured track assistant or associate prof in economics at university of Melbourne @unimelb.bsky.social

welcome applications from all fields (but especially from econometrics)

econjobmarket.org/positions?sh...

Does it help if you call it a “folder”?

I’m still trying to get students to stop sending me papers called “draft.docx.”

So five different estimators when we use MLE weights collapse to one estimator using IPT weights. IPWRA and normalized AIPW work pretty well with MLE weights but differ from each other and the other estimators. The IPT-based estimator is hard to beat especially when the mean and PS are both wrong.

ssc install teffects2

The syntax is essentially the same as teffects and allows MLE or IPT logit PS estimation. Where appropriate, we allow normalized or unnormalized weights (with MLE for IPW and AIPW). Preference for normalized. Standard errors account for all estimation uncertainty.

Weights for IPW and AIPW are automatically normalized. Holds for ATE and ATT. With MLE-based weights, the three estimators are all different, and the IPW and AIPW weights are not automatically normalized. We have an accompanying Stata command, teffects2.

Thanks Paul. It’s nice to have a paper that’s both elegant (if I may say so) and practically useful. The conclusion is that using a particular covariate balancing PS estimator — inverse probability tilting — renders IPW, AIPW, and IPWRA all numerically identical with a linear conditional mean.

This puzzled me and seems like a kind of appropriation. Mundlak was squarely in the frequentist/FE camp.

Oh I think you know I’ve always been a barbarian.