Had no idea that the best data science podcast is back 😀
podcasts.apple.com/us/podcast/l...
What?? Very exciting!
Battery life?
A solution could be to legislate authority away from that post. But this is a complicated coordination problem.
The regime rules through institutions, but over the years of Khamenei’s leadership the institutions were organized in a way that concentrated authority in his office. Now there is trepidation about appointing someone to a post with such concentrated authority.
Good analysis of the institutional conundrum facing the Iranian leadership after Khamenei’s death: www.project-syndicate.org/commentary/k...
If you are looking for a way to do something meaningful amidst the current events, I’ll suggest again for you to just sit and watch “It Was Just an Accident.”
Yes 🙌
Re the age of AI issues, we do a lot of pen and paper work in class now. I want to see that students can at least get started on mapping out formal analyses using potential outcomes, DAGs, etc and probability/asymptotic operations.
You’re right re Rosenbaum. I tend to stay in the Neymanian framework. That said I particularly appreciate Rosenbaum’s insights in evidence factors and nested testing.
Yes that is very much my inferential foundation.
In all seriousness though, I’m rarely (not never, but rarely) convinced that identification strategies like IV, DID, or RD get us *all the way* to our target quantities. So modeling with covariates comes in to play more than Mostly Harmless teaches.
Excellent. Let’s see if we can arrange a talk. I’ll reach out when we are closer to when I’ll be discussing this.
I had also done extended DID lectures, following the frenzy of papers. I will condense that too, getting quickly to the counterfactual estimation perspective and trajectory balancing, and open up space for g computation. Aside from IV, we are nearing full “post Mostly Harmless.”
Update to my Quant 2 course: I used to do an Angrist and Pischke style discussion of regression models and treatment effects. Dumped that (or actually made it a recitation/review), brought in double machine learning instead. cyrussamii.com?page_id=4190
“Affective polarization” is a lot of syllables when the word “hate” is right there.
For anyone looking for postdocs for their students, over the next few months LSE “fellow” positions will start being listed. Research + teaching but 2-3 year runway.
The new Global School of Sustainability has 5 new ones posted now! jobs.lse.ac.uk/Vacancies/W/...
Seems it is still going strong so if you email the folks at SIWPS they’d probably have it to share: www.siwps.org/programs/sum...
Excited to see this out. The paper was a huge lift. If you’d be surprised that citizen empowerment *increases* the share that pay taxes in a place like Kinshasa, check out the details in Soeren’s thread.
This will be chapter 1 in my regression textbook.
Folks, some news. No, not that kind of news – what do you think this is, LinkedIn? It's this:
1. The applied causal graphs workshop deadline is 28th Feb. so get your abstracts in and hang out with us in Potsdam this May. Form and description is below
2. @dagophile.bsky.social is giving a keynote 🥳
Really, instead of Gauss-Markov, students should be taught the finite-sample, uniform-risk optimal sieve expansion depth under Holder smoothness. That would actually be useful.
What Will It Take to Rebuild the Government in Post-Maduro Venezuela?
I wrote about three interrelated governance challenges: 1) a fragmented security apparatus, 2) the strong presence organized criminal groups, and 3) the vulnerability of the oil sector. 👇
www.lawfaremedia.org/article/what...
It Was *Just An Accident (or closer to the original Farsi, Simply An Accident)
Finally got to watch “It Was An Accident.” A lot of the psychology of the situation in Iran is right there, very well portrayed. The movie makes you feel both hopeful and utterly defeated.
But yeah just throwing a series of derivations up there or bombarding people with code without providing qualitative insight is not so valuable, especially when the machine can do it better.
But math and programming are also good!
On this point I love this paper by Acevedo et al., which shows that a fat tailed effect distribution implies that optimal research programs should not try to hunt for small effects: www.journals.uchicago.edu/doi/abs/10.1...
Absolutely—experiments *in the real world* count much more now.
And on a personal note was great to be back in the Great Lakes region after ~10 years to enjoy a cold one:
