Kvasikokeelliset menetelmät terveydenhuollon ja terveystaloustieteen vaikutusarvioinneissa: journal.fi/sla/article/vi… #sote #terveydenhuolto #vaikuttavuus #kustannusvaikuttavuus #kausaalipäättely #terveystaloustiede

There's a commonly used method for causal inference that just doesn't work It's called Nonlinear Mendelian Randomization—NLMR—, and the hundreds of papers that have used it need to be investigated🧵 To grok the issue, answer: does something look off about this vitamin D result?

New paper: "What Does Consulting Do?" w/ Gert Bijnens, Simon Jäger Consulting is a storied but opaque industry. Our VAT-based B2B data permit the 1st comprehensive study: Which firms buy consulting? What happens when they do? nber.org/papers/w34072 eml.berkeley.edu/~schoefer/scho…

Having a child with Down syndrome increases the likelihood of parental co-habitation and subsequent child-bearing, from Liran Einav, Amy Finkelstein, and Petra Persson nber.org/papers/w34064

🚨 New study using data from 50,000+ Finnish conscripts: High-income youths earn 5.7% more long-term if dormed with peers from similarly wealthy families. Low-income conscripts see no such gain. Peer effects may reinforce, not reduce, inequality. 🔗 aeaweb.org/articles?from=…

💊 Uusi suomalaistutkimus tarkastelee, miten lääkkeiden omavastuurajan ylittäminen vaikuttaa lääkkeiden kulutukseen. Kun omavastuu ylittyy ja hinnat laskevat, lääkekulut kasvavat keskimäärin 17 % – pääosin siksi, että ihmiset ostavat enemmän lääkkeitä. 🔗onlinelibrary.wiley.com/doi/10.1002/he…

Incredibly happy to see my paper forthcoming in QJE! I quantify how much of worldwide growth and poverty reduction can be accounted for by education over 1980-2019. A quick summary 1/9 👇

New study from Denmark 🇩🇰 of over 1.5 MILLION people who received an mRNA Covid vaccine NO increased risk of 29 different adverse events Filter out the noise. Look at real data. mRNA vaccines are incredibly safe. jamanetwork.com/journals/jaman…

Many studies attempt to estimate the causal effect of exposures by matching groups and then comparing them cross-sectionally, longitudinally, and so on. The idea is that, after matching, changes will be causally interpretable as effects of that exposure. But, this rarely works.

What happens before and after tenure? Check out our latest paper in PNAS PNASNews: Tenure and Research Trajectories Led by Giorgio Tripodi, postdoc in my lab. Fun collaboration with Xiangzheng, Yifan Qian, Dakota Murray, Ben Jones & Chaoqun Ni 1/n

🧪 Satunnaistetut kokeet eivät takaa kliinisesti relevantteja tuloksia, jos kokeen osallistujat eivät noudata protokollaa (non-adherence). Angrist & Hull (PNAS): IV-menetelmä voi korjata tämän, silti yhä alikäytetty, erityisesti seulontatutkimuksissa. 🔗 doi.org/10.1073/pnas.2…

On cue . . .

Economics job market candidates p-hack. But, why? Well, doing so helps them get academic jobs. "We find evidence that marginally significant results in [job market papers] are associated with higher academic placement likelihoods."

New working paper! We ran a field experiment with >1,300 university students randomly assigned to online vs in-person lectures in first-semester courses. What happens? Online instruction hurts performance, but only for women♀️📉 wt Xiaoyue Shan Uschi Backes-Gellner

🧠 NEJM Evidence: Luonnolliset koeasetelmat tarjoavat kausaalinäyttöä vaikuttavuudesta, kun satunnaistettu koe (RCT) ei ole mahdollinen – etenkin kliinisessä päätöksenteossa. 👉 Uskottavat tutkimusasetelmat ratkaisevat soten vaikuttavuusarvioinneissa 🔗 doi.org/10.1056/EVIDra…

"We find no evidence that preregistration in itself reduces p-hacking and publication bias." But... "When preregistration is accompanied by a preanalysis plan we find evidence consistent with both reduced p-hacking and reduced publication bias."

🩺 Can AI reduce medical overtreatment? Wang et al. (2025, JHE) use a lab-in-the-field RCT with 196 physicians to study how AI affects clinical decisions. ➤ 𝑅𝑒𝑠𝑢𝑙𝑡𝑠: AI cuts overtreatment by up to 80% and boosts accuracy by up to 20%. 🔗 doi.org/10.1016/j.jhea… #sote