iiinova — where your digital self keeps evolving

David Slichter @davidslichter.bsky.social · 12h

1 ≥ Pr(quitter|uses partial identification) ≥ 0

David Slichter @davidslichter.bsky.social · 02/10/2026

Hugo Jales is one of the most interesting, and most fun, people in economics. I really enjoyed listening to his conversation with Greg Caetano. www.youtube.com/watch?v=YGSs...

youtube.com

What Do Academic Incentives Actually Reward? | With Hugo Jales

YouTube video by It’s all chaotic! — Seeking Nuance

010

David Slichter @davidslichter.bsky.social · 23/09/2026

Please pretend I am posting this anonymously. I recently accompanied my children to such a place and secretly wore regular socks. So far the authorities have not found me.

031

David Slichter @davidslichter.bsky.social · 23/09/2026

If you enjoy the idea of being employed, you might be interested in our Assistant Professor in Applied Microeconomics position: www.aeaweb.org/joe/listing.... We're open to both advanced APs and fresh PhDs.

aeaweb.org

American Economic Association: JOE Listings - August 1, 2026 - January 31, 2027

021

Reposted by David Slichter

Institute for Replication @i4replication.bsky.social · 14/09/2026

It only took us 5 years, but I4R finally has a newsletter. 🎉 We’ll use it to share new studies, Replication Games, research, events, podcasts, and other updates from the Institute for Replication. Subscribe here: www.i4replication.org/newsletter

075

David Slichter @davidslichter.bsky.social · 12/09/2026

True, though groups of humans collectively do have the power to kill off other species, as evidenced by the fact that we've done it a number of times.

110

Reposted by David Slichter

Todd Pugatch @toddpugatch.bsky.social · 08/09/2026

1st @aefpweb.bsky.social EdDev mtg of the academic year, this F Sep 11 11am US eastern! Presenters: -Pepi Pandiloski, "Nation Formation in the Wild Wild East" -Innocent Akampurira, "The Effect of School Feeding on Education Outcomes" Contact me for Zoom link. Join group (free) at AEFP. Plz RT!

011

Reposted by David Slichter

John V. Kane @uptonorwell.bsky.social · 31/08/2026

Compared to "significant effects," null results lead citizens to: --View study as less important --Ask for less info --Be less willing to share it --Think journalists would be less likely to cover the study --See the study as being lower quality & researchers as less competent (😬)

284

David Slichter @davidslichter.bsky.social · 01/09/2026

How is it simultaneously possible that Ariely is still getting 4000 new citations per year, and that nobody is surprised that this paper turns out to be fabricated?

010

David Slichter @davidslichter.bsky.social · 19/08/2026

What's the a priori plausible magnitude of effect that you think people should update away from, what's the a priori plausible prior you think they should update towards, and why do you think these magnitudes are a priori plausible?

000

David Slichter @davidslichter.bsky.social · 19/08/2026

Yes, I agree with their theoretical best guess about the effect. Where I disagree with them is that the tone of the paper is nonetheless "you should update towards the view that there's no effect on labor market outcomes" when the power issue actually means we should just not update at all.

100

David Slichter @davidslichter.bsky.social · 19/08/2026

Hard to determine precision of the employment results without a joint hypothesis test or evidence about serial correlation in the outcome across time horizons, but the fact that results aren't significant except when implausibly large suggests we should do little updating.

000

David Slichter @davidslichter.bsky.social · 19/08/2026

But a large magnitude point estimate is a predictable consequence of having lots of sampling error. The Bayes factor comparing "the effect is a 1% increase in earnings" to "the point estimate is right" is 0.7. Among plausible priors for earnings effects, there is basically no updating.

100

David Slichter @davidslichter.bsky.social · 18/08/2026

Make the effect three times larger (because magic) and you would still only have Pr(reject at 5% level) = 0.069. So, I'd bet on a null result well more than 90% of the time!

110

David Slichter @davidslichter.bsky.social · 18/08/2026

I'm not surprised, there's no statistical power! The intervention probably increased attainment by on the order of 0.1 years, which might raise earnings by 1%. Meanwhile, the SE of the effect is 7.5% of control mean earnings. With that effect size, Pr(reject null at 5% level) = 0.052.

210

David Slichter @davidslichter.bsky.social · 02/08/2026

Maybe, but also "come up with an interesting question" hardly seems like some kind of insurmountably difficult task for LLMs, given what they can already do.

000

David Slichter @davidslichter.bsky.social · 22/07/2026

I always think of mean reversion as a reason to not believe parallel trends assumptions! As in, the difference in mean potential outcomes between treated and control is likely to revert in the direction of zero. So controlling for lagged DV and parallel trends often feel more like bounds to me.

120

David Slichter @davidslichter.bsky.social · 15/07/2026

Seems like maybe people should report the pre-treatment gap between treated and control observations plus the degree of mean reversion observed in the control group. That would help readers assess whether results could be driven by mean reversion or not.

110

David Slichter @davidslichter.bsky.social · 15/07/2026

Looking forward to reading it! I've suspected for a while that mean reversion is driving a lot of DiD results.

110

David Slichter @davidslichter.bsky.social · 09/07/2026

But if the CEF isn't linear, then SAT scores contain *more* information than is implied by a linear regression.

010

David Slichter @davidslichter.bsky.social · 08/07/2026

That's not what these graphs show. These are binned scatterplots, i.e., plots of averages within bins. The noisier plots for first-gen and low-income are presumably because these averages are computed from smaller samples.

120

David Slichter @davidslichter.bsky.social · 08/06/2026

Most of my value-added as an advisor is telling this to grad students.

010

David Slichter @davidslichter.bsky.social · 29/05/2026

Macroeconomics papers are egregious "the title is just a list of variables or concepts" offenders. Representative sample: www.aeaweb.org/journals/mac...

aeaweb.org

American Economic Association: American Economic Journal: Macroeconomics Issues

010

David Slichter @davidslichter.bsky.social · 28/05/2026

Thanks for (the considerably more effortful task of) writing it!

010

David Slichter @davidslichter.bsky.social · 27/05/2026

Great paper that I teach in my Econ of Ed class. It's well-known that, while No Excuses charter schools are great for test scores, other charters aren't. But this paper shows that plenty of non-No Excuses charters with zero/negative test score impacts might nonetheless be great schools.

131

Reposted by David Slichter

Jeremy Singer @jeremylsinger.bsky.social · 26/05/2026

NEW: We connected school attendance value-added estimates (for 2022-23 through 2024-25) to statewide survey data on school-based attendance practices. Our goal was to identify effective attendance strategies (e.g., specific practices, organizational systems, staffing, leadership). What did we find?🧵

Identifying Effective Attendance Strategies in Michigan

12711

David Slichter @davidslichter.bsky.social · 20/05/2026

Had a great time at the FLX Econ of Ed conference this year. Special shoutout to @mariabzhu.bsky.social for organizing, and specifically for having name tags where you can actually read people's names!

051

David Slichter @davidslichter.bsky.social · 12/05/2026

Institutions not in the top 200 of the QS rankings include Rochester, Minnesota, and Maryland, among many others.

050

David Slichter @davidslichter.bsky.social · 12/05/2026

New study finds that jogging for as little as one minute per day may help prevent spinal cord injuries

000

David Slichter @davidslichter.bsky.social · 08/05/2026

The coefficient you'll get will be the weighted average of slope of Y wrt X within each group, with weights depending on what fraction of variance in X is within each group.

110

David Slichter @davidslichter.bsky.social · 08/05/2026

But in your setting where your treatment is continuous and your controls is categorical, you just implement a saturated model and E(X|W) is guaranteed to be linear and additively separable, so no between-covariate identification.

100

David Slichter @davidslichter.bsky.social · 08/05/2026

If W were a continuous variable and X were binary, you'd likely have a lot of nonlinearity in E(X|W), so lots of between-covariate identification. Plus then, due to lack of common support, you'd have very little within-covariate identification. So OLS would be driven by between-covariate.

100

David Slichter @davidslichter.bsky.social · 08/05/2026

All concerns about extrapolation, functional form assumptions, and negative weights are to do with between-covariate identification.

110

David Slichter @davidslichter.bsky.social · 08/05/2026

It's straightforward to show (using Frisch-Waugh-Lovell) that the OLS coefficient on X is a weighted average of the coefficients obtained from these two sources of variation, with weights corresponding to what fraction of total variance in [X - linear projection of X on W] comes from each source.

110

David Slichter @davidslichter.bsky.social · 08/05/2026

Between: If E(X|W) is nonlinear, but W is assumed to enter linearly in Y equation, then, if X enters in Y equation, we'd expect a nonlinear relationship between W and Y mirroring nonlinear relationship between W and X. Formally, this is regress Y on [E(X|W) minus linear projection of X on W].

110

David Slichter @davidslichter.bsky.social · 08/05/2026

(1/n) Regression estimates are a weighted avg of within- and between-covariate variation in X. Within: Compare observations with W_i=w for controls W, but one person has more X than the other. Formally, this is regressing Y on X-E(X|W).

110

David Slichter @davidslichter.bsky.social · 06/05/2026

In the summer, my upstate NY kids complain about how it isn't winter.

120

David Slichter @davidslichter.bsky.social · 06/05/2026

You get jumps in earnings at degree completion years, but also jumps in selection bias at degree completion years.

010

David Slichter @davidslichter.bsky.social · 06/05/2026

That link is reporting the correlation between schooling and earnings, not the effect of schooling on earnings. Literature says that the effect of the marginal year of schooling on earnings is maybe 5-10% per year. (I personally think it's on the lower end of that.)

110

David Slichter @davidslichter.bsky.social · 05/05/2026

In this sense, the HC acquired in college was pivotal to a successful career. But it doesn't necessarily feel that way because you graduated with only the bare minimum skill level needed to be useful. (3/3)

010

David Slichter @davidslichter.bsky.social · 05/05/2026

Engineering undergrads can't reliably design planes, but they can achieve the minimal competence to be useful to Boeing. Without learning some math, physics, and engineering, Boeing cannot hire you in an engineering role and you will never learn to design planes. (2/3)

110

David Slichter @davidslichter.bsky.social · 05/05/2026

I think part of it is also that college does not confer full expertise or anything close to it. It confers the minimum degree of semi-expertise to qualify for jobs where you will then learn things. (1/3)

140

David Slichter @davidslichter.bsky.social · 05/05/2026

(6/6) Education inputs which are unobserved by employers or do not signal ability nonetheless seem to cause large earnings changes: academic.oup.com/qje/article/... www.aeaweb.org/articles?id=... www.barbarabiasi.com/uploads/1/0/... papers.ssrn.com/sol3/papers....

barbarabiasi.com

1404

David Slichter @davidslichter.bsky.social · 05/05/2026

(5/n) College major differences caused by tiny score changes on admissions exams have huge effects on earnings, and with a pattern of comparative advantage that is far easier to rationalize with HC model: academic.oup.com/qje/article/...

academic.oup.com

Field of Study, Earnings, and Self-Selection*

Abstract. This article examines the labor market payoffs to different types of postsecondary education, including field and institution of study. Instrumen

1332

David Slichter @davidslichter.bsky.social · 05/05/2026

(4/n) Signaling theory predicts that going to a more selective colleges should increase earnings by a lot, yet college selectivity is essentially uncorrelated with value-added to wages: blueprintcdn.com/wp-content/u...

blueprintcdn.com

1394

David Slichter @davidslichter.bsky.social · 05/05/2026

(3/n) Signaling theory implies largest benefits at the start of career when employers know the least, but actually returns to schooling are negative until late 20s, and peak in mid or late career: www.journals.uchicago.edu/doi/full/10....

journals.uchicago.edu

University of Chicago Press Journals: Cookie absent

1423