Sign in

Nikolaj Harmon

@nikharmon.bsky.social
438 followers 253 following 37 posts

Associate Professor of Economics, University of Copenhagen Labor Economics, Econometrics and Political Economics web.econ.ku.dk/nharmon #econsky

PostsRepliesMedia
Nikolaj Harmon @nikharmon.bsky.social · 23/04/2026
#EconSky : Any recommendations/ressources for setting up a good research/teaching workflow including AI? At the moment I’ve ended up with something pretty haphazard (using different LLMs, some via browser some via apps, lots of drag and drop of individual files, etc.) I’m looking for more structure.
121
Reposted by Nikolaj Harmon
Carl T. Bergstrom @carlbergstrom.com · 18/11/2025
Way back in 1947, Alan Turing had thoughts on how AI would influence the demand for skilled labor. (via Matteo Pasquinelli 2023 _The Eye of the Master_)
Roughly speaking those who work in connection with the [automatic computing engine] will be divided into its masters and its servants. Its masters will plan out instruction tables for it, thinking up deeper and deeper ways of using it...The masters are liable to get replaced because as soon as any technique becomes at all stereotyped it becomes possible to devise a system of instruction tables which will enable the electronic computer to do it for itself. It may happen however that the masters will refuse to do this. They may be unwilling to let their jobs be stolen from them in this way. In that case they would surround the whole of their work with mystery and make excuses, couched in well chosen gibberish, whenever any dangerous suggestions were made. 

 — Alan Turing (1947) Lecture on the Automatic Computing Engine
11513159
Nikolaj Harmon @nikharmon.bsky.social · 23/10/2025
Ah! I like that example! It’s effectively pointing out that we might be particularly prone to do implicit meta studies across our own works (inferring V as a particular important control from several papers). Fixed seeds can bias conclusions from those.
120
Nikolaj Harmon @nikharmon.bsky.social · 23/10/2025
I can very much see the practical implication *within* a project. Across projects it feels less clear. Picking a different seed introduces random variation in results (conclusions) across projects which does not seem particularly useful Maybe an argument is to make projects “meta-analysis-friendly”
110
Nikolaj Harmon @nikharmon.bsky.social · 23/10/2025
I’m curious: How would you articulate the problem with using the same seed across different projects?
100
Nikolaj Harmon @nikharmon.bsky.social · 19/08/2025
Interesting study but worth noting that “correlation is not causality” applies here. This shows a correlation between political beliefs and conclusions, it does NOT show that political beliefs affect conclusions.
000
Reposted by Nikolaj Harmon
John Holbein @johnholbein1.bsky.social · 05/08/2025
In Denmark, new mothers are quasi-randomly assigned to mother groups by nurses. This new paper shows that being assigned to a group with a depressed peer decreases mothers self-reported mental health + increases mental health care uptake. Fascinating!
24310
Nikolaj Harmon @nikharmon.bsky.social · 11/02/2025
Jeg ville jo nok insistere på at det første er lidt et definitionsspørgsmål :) Tak for oplysning. Nu er jeg med igen (på både substans og sprogbrug)
010
Nikolaj Harmon @nikharmon.bsky.social · 11/02/2025
Ja, mindre samlet trafik ved road pricing, men også en modsatrettet effekt fra af at nye veje nu giver indtægter fra roadpricebetalinger. Men det sidste du nævner lyder som om den gængse nomenklatur er at se dette som en sænkning af nettoomkostningerne (selvfinansiering) frem for højere afkast?
100
Nikolaj Harmon @nikharmon.bsky.social · 11/02/2025
Hvis man regnede helt rigtig, burde vejprojekter så ikke se bedre ud med optimal Road pricing? Road pricing løser et eksternalitetsoroblem som sikrer optimal udnyttelse af vejnettet = større samf. afkast
100
Reposted by Nikolaj Harmon
Paul Hünermund @p-hunermund.com · 11/02/2025
4603
Nikolaj Harmon @nikharmon.bsky.social · 11/02/2025
Yes, nu er jeg med. Så den overordnet gode policy er congestion pricing plus vejudvidelse indtil omkostning>=gevinst i sparet rejsetid.
110
Nikolaj Harmon @nikharmon.bsky.social · 11/02/2025
Jeg er nysgerrig om jeg er bagud på min viden her, men jeg troede: 1) udvidelse af transport-infrastruktur løser ikke trængselsproblemer (eksternalitetsomkostninger), 2) congestion road pricing gør. Er det ikke det der står i tweetet? (Plus noget tvivlsomt om kollektiv transport)
100
Reposted by Nikolaj Harmon
Torben Heien Nielsen @torbenheien.bsky.social · 27/01/2025
Hello BlueSky!
0135
Reposted by Nikolaj Harmon
Ben Golub @bengolub.bsky.social · 24/12/2024
In economics, editors, referees, and authors often behave as if a published paper should reflect some kind of authoritative consensus. As a result, valuable debate happens in secret, and the resulting paper is an opaque compromise with anonymous co-authors called referees. 1/
28547158
Reposted by Nikolaj Harmon
John Holbein @johnholbein1.bsky.social · 19/12/2024
435678
Reposted by Nikolaj Harmon
Jonas Jessen @jonasjessen.bsky.social · 16/12/2024
🚨 PUBLICATION ALERT 🚨 How are firms are affected by (longer) parental leave absences? Paper joint with @mathiashuebener.bsky.social, @danielkuehnle.bsky.social & Michael Oberfichtner is forthcoming at The Economic Journal doi.org/10.1093/ej/u... Short thread below 👇
14713
Reposted by Nikolaj Harmon
Jeppe Druedahl @jeppedruedahl.bsky.social · 27/11/2024
Imorgen torsdag har vi Økonomisk Eksploratorium om årets nobelpris.
0123
Nikolaj Harmon @nikharmon.bsky.social · 25/11/2024
TL;DR: The novel data is the universe of human written text. The LLM is just a way to process this data.
000
Nikolaj Harmon @nikharmon.bsky.social · 25/11/2024
I share your skepticism but there is a coherent argument for why this could work: An LLM is (approx) trained on the universe of text to behave like humans do. To someone who has read and understood the universe of text; the LLM will never surprise. But none of us have!
100
Nikolaj Harmon @nikharmon.bsky.social · 20/11/2024
Let’s just say my enthusiasm for accepting referee tasks dropped for a while after. I’m sure I have lots of bad (subjective) opinions as reviewer. But this was not that.
010
Nikolaj Harmon @nikharmon.bsky.social · 20/11/2024
Later I got the revised version to review. The authors had BEAUTIFULLY addressed 1… … but also doubled down on the bias claim, STILL without properly defining an estimand. 🤯 Situation was weird after that. Eventually ended with editor half-agreeing with both of us and accepting then paper. 🤷‍♂️ N/N
110
Nikolaj Harmon @nikharmon.bsky.social · 20/11/2024
… the paper never defined the estimand! And in fact for a sensible choice of estimand, there is NO bias in the existing method. I wrote a positive referee report detailing how I would fix 1. Plus suggesting that claim 2. be taken out (the paper’s contribution was easily high enough without it) 2/N
110
Nikolaj Harmon @nikharmon.bsky.social · 20/11/2024
I’ll keep this vague since I hold no ill will here. But basically: I refereed a really good paper with only two drawbacks: 1. The exposition/framing was doing the results a disservice (IMO) 2. The paper claimed to fix a bias in an existing method that I have worked on extensively, except… 1/N
110
Nikolaj Harmon @nikharmon.bsky.social · 20/11/2024
Man, I have a war story about this one.
130
Nikolaj Harmon @nikharmon.bsky.social · 20/11/2024
Cue that Lucas quote in 3…2…1
000
Reposted by Nikolaj Harmon
Robert (Bob) Kubinec @rmkubinec.bsky.social · 19/11/2024
#bluesky has over 20 million users! 🥳👏 here's to a quality media environment that promotes information, not misinformation! 🫡 (except for @kjhealy.bsky.social posts, those are just bad) (@timgill924.bsky.social is pretty bad too) (what is it with sociologists... 🤔)
085
Reposted by Nikolaj Harmon
Mark Wittek @markwittek.bsky.social · 20/11/2024
🎺 Open Assistant Professor position in Social Data Science at our Department (DNDS). Great research, supportive social atmosphere, willingness to communicate across disciplinary boundaries, and a beautiful city with affordable rents. For details see: careers.ceu.edu/job/Vienna-A...
careers.ceu.edu
Assistant Professor (f/m/d)
Assistant Professor (f/m/d)
01312
Nikolaj Harmon @nikharmon.bsky.social · 19/11/2024
Either I don’t understand your question or I’ve already answered it above.
000
Nikolaj Harmon @nikharmon.bsky.social · 19/11/2024
But adding a spline or polynomial makes it much harder to summarize the relationship. If the log model is a good fit, you can summarize it in one number (elasticity blah blah). (Obviously many cases where you can do better than logging but I’m happy to defend it as a good norm/first-pass)
100
Nikolaj Harmon @nikharmon.bsky.social · 19/11/2024
Hot take: 80% of the time logging covariates leads to a model that better fits the data. (and honestly who cares about the units of the covariates?)
120
Reposted by Nikolaj Harmon
Dave Itzkoff @ditzkoff.bsky.social · 19/11/2024
this is like something a character in Animal Farm would say, and every 6th grader in the classroom would understand why it's wrong www.nytimes.com/2024/11/15/o...
489160863280
Reposted by Nikolaj Harmon
Jeppe Druedahl @jeppedruedahl.bsky.social · 19/11/2024
To all job market candidates in macro: University of Copenhagen has a 6 year tenure-track assistant professor position. See more here: econjobmarket.org/positions?show… If you have any questions, please contact me.
econjobmarket.org
EJM - Econ Job Market
073
Reposted by Nikolaj Harmon
Jonathan Roth @jondr44.bsky.social · 19/11/2024
One of us turned 1 and the other got tenure 😃😃
4876311
Reposted by Nikolaj Harmon
Hannes Ullrich @hannesullrich.com · 19/11/2024
An all new Econ of AI starter pack (by @arpitrage.bsky.social) : go.bsky.app/DfnDyqb
3209
Reposted by Nikolaj Harmon
Thomas Wilken @thomaswilken.bsky.social · 19/11/2024
Glædelig international mandedag! I dag går mine tanker ud til følgende figur fra @rockwoolfonden.bsky.social. Mænd: 👉 Er et helt karakterpoint bagefter i gennemsnit ved 9. klasses afgangseksamen 👉 Får i MARKANT lavere grad børn 👉 Dør i langt højere grad af fortvivlelsesdødsfald #dkpol
4398
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
Bonus 4: For the more technically inclined: My paper establishes a certain equivalence between the csdid/did_multiplegt estimators and did_imputation estimator. The former estimators can be viewed as applying the latter estimator to a particular (first-differenced) model. N+4/N
000
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
Bonus 3: csdid and did_multiplegt requires a choice of a baseline period. This is sometimes referred to as being “arbitrary”. Results in my paper show that it is not. For the efficiency result, the baseline MUST be the period just before treatment (the standard choice). N+3/N
100
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
Bonus 2: There are important special cases where csdid/did_multiplegt is efficient also with staggered adoption (and random walk errors). The key example is when estimating treatment effects ONLY in the very first period where treatment kicks in (see paper for more). N+2/N
100
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
Bonus 1:   Struggling to think about persistence? Simple test: Say your outcome variable Y happens to increase for some individual in some time period, what do you expect next period? Probably Y drops back down again -> impersistent shocks Probably Y stays high -> persistent shocks   N+1/N
100
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
Conclusion: Below is a simple flowchart that shows you (and your referees) how to pick the best DID estimator for your setting. Full paper and Stata package for the Stepwise DID adjustment is on my website: web.econ.ku.dk/nharmon/Rese... Comments very welcome! N/N
100
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
Ok but does this matter? Simulation + application: Yes! Both outcome variables with “persistent” and “impersistent” shocks are common. Picking the right estimator improves precision by as much as 50 pct more data. (Does not require “(im)persistent shocks” to hold exactly.) 8/N
100
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
Ok but non-staggered adoption is restrictive; what if adoption is staggered? My paper (pt 2): A small new adjustment to csdid/did_multiplegt is enough for them to be efficient under staggered adoption. I call this “Stepwise DID” (see paper for the details and intuition) 7/N
100
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
But what about the other estimators then? Is there ever a reason to use them??? My paper (pt 1): YES! If treatment adoption is non-staggered and the outcome has “persistent shocks” (Random Walk errors), then in fact csdid/did_multiplegt are THE best unbiased estimators. 6/N
100
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
But WHICH estimator is more efficient? Borusyak et al, ReStud: did_imputation is THE best unbiased estimator under “impersistent shocks” (Gauss-Markov errors) So if noise in your outcome variable is driven by impersistent shocks, you almost surely should use did_imputation 5/N
100
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
In a given sample, these estimators require the EXACT same assumptions for unbiasedness. How to pick one? Stats 101 gives a simple answer: Efficiency! The best estimator is the one with smallest variance (Reminder: more efficient estimator -> smaller SEs with the same data) 4/N
100
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
The choice set: For a given sample and with standard assumptions there are effectively two DID estimators to choose from. For the purpose of this thread, I’ll refer to them by some of their popular Stata package names: Option 1: csdid/did_multiplegt Option 2: did_imputation 3/N
100
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
Recent work has yielded great new DID estimators for heterogeneous treatment effects. Awesome! 👍 But this creates a burden-of-choice for applied folks 🤯: Need to pick a (preferred) estimator (and this MUST be done ex ante) 2/N
100
Nikolaj Harmon @nikharmon.bsky.social · 25/01/2024
🚨 DID Paper Thread 🚨 Are you confused about which new DID estimator to use? A bit more than a year ago, I wrote a short paper that gives a simple answer. I just completed a revision of that paper so here’s a 🧵 to celebrate! 1/N
293