Nikolaj Harmon

Associate Professor of Economics, University of Copenhagen, interested in Labor Economics, Political Economy and Econometrics

[Nikolaj Harmon @NikHarmon]

🚨 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 #econtwitter #differenceindifferences

[Nikolaj Harmon @NikHarmon]

@SolomonKurz @jfiksel1 Yes that would be biased.

[Nikolaj Harmon @NikHarmon]

@ybytata @kearney_melissa But note that the decisions you mention are all ex ante wrt. the specific child in questions: postponing a birth is done before having
the child and so is deciding not to have the child altogether.

[Nikolaj Harmon @NikHarmon]

🇺🇸💔😢

Worst coin flip ever

[SODAS, Copenhagen (Bluesky: @cphsodas.bsky.social) @CPH_SODAS]

Sebastian Barfort from consulting company Round chatted with us about using social data science in the business world. 🕴️We talked about grown-up toy lovers, whether models can replace anthropologists, and why Sebastian just loves the internet. Read more: https://t.co/9qk7i0oZWl

[Nikolaj Harmon @NikHarmon]

@ChelseaParlett Yeah, you just need to add “but mostly close enough” at the end.

[Irena Buzarewicz @IrenaBuzarewicz]

Good morning

[Alicia Curth @AliciaCurth]

When Double Descent & Benign
Overfitting became a thing, I was a masters student in statistics — and so confused. I couldn't reconcile what l had literally just learned about bias-variance&co with modern ML

Here's what I wish someone had told me then: https://t.co/Crt5zTIoqC 1/n

[Nikolaj Harmon @NikHarmon]

@_MattSavino @vsbuffalo Maybe a more illustrative example is if the two numbers are -1 and 2. If revealing 2 my chance of switching is now 1/6 so I win 5/6 of the time.

[Nikolaj Harmon @NikHarmon]

@_MattSavino @vsbuffalo Not sure what you are asking but say the numbers are -1 and 1. If revealing -1 there is a 3/4 chance I switch and win. If revealing 1 there is a 1/4 chance I switch so there is 3/4 chance I do not switch and win. In both cases my chance of winning is 3/4. Makes sense?

[Nikolaj Harmon @NikHarmon]

@_MattSavino @vsbuffalo The version that I find most surprising and interesting involves not making any assumptions on the two hidden numbers.

[Nikolaj Harmon @NikHarmon]

@JeppeDruedahl @RuneStahl Jeps, min reservation går også både på relevansen af spansk evidens og på nogle potentielle grundlæggende udfordringer med studiet. (Tror bare ikke du skal undervurdere potentiel impact af et tweet som dit ovenfor)

[Nikolaj Harmon @NikHarmon]

@JeppeDruedahl @RuneStahl Jeg synes sgu du er nødt til at kvalificere det udsagn lidt Jeppe. Jeg er oplagt forudindtaget her, men det er min klare vurdering at resultaterne i det working paper er helt irrelevante i en dansk kontekst.

[Guilherme Jardim Duarte @guilhermejd1]

No defiers is a required assumption for identification of LATE in IV analysis. If that is false, the best thing you can do is to calculate bounds for the LATE.

But what happens if limit the maximum proportion of defiers. For instance, what if I don't believe is > 10%?

[The Review of Economics and Statistics (REStat) @restatjournal]

Expanding language training increased earnings of refugees. In the July issue, by Mette Foged @MetteFoged_econ Linea Hasager @HasagerLinea Giovanni Peri @UCD_GlobalMC Jacob Arendt @Jacob_N_Arendt Iben Bolvig @IBolvig https://t.co/7QXStbfKeW

[Nikolaj Harmon @NikHarmon]

@Alkalimeter @vsbuffalo Ah right, some version of that should work for sure. Nice extension.

[Nikolaj Harmon @NikHarmon]

@Alkalimeter @vsbuffalo No, for tails and n>=0 you always stick with the first number. This implies that for n>=0 the probability of switching is 1/2 * 1/n

[Nikolaj Harmon @NikHarmon]

@economoser Sure. But adding assumptions on the distribution of the numbers and/or guaranteeing only weakly better than 50% makes it much less impressive and surprising to me. The “strictly better” and “ for ANY two numbers” is what I found really counterintuitive.

[Nikolaj Harmon @NikHarmon]

@vsbuffalo Cc (and thanks): @JohnnyBlack_HS , @DeadFishCapital , @ArthurB , @b_subercaseaux , @Alkalimeter

(4/3)

[Nikolaj Harmon @NikHarmon]

@vsbuffalo For any n this strategy completes in finite time and gives >50% win probability.

It still pushes on the limits of feasibility since (ex ante) you risk of running out of balls unless infinite supply (or: prob. of spending arbitrarily long placing balls in the urn is >0)

(3/3)