But could there be aggravating circumstances that our methodology did not capture? We conducted sensitivity checks for such “omitted variable bias”. And we found that there would need to be an omitted variable of equal strength to “repeated offense” to increase our p-values beyond 5%

We used regression analysis to control for such differences: Our preferred models include all judge and defendant personal characteristics, and dummy variables for the crimes for which defendants were convicted as well as aggravating and mitigating circumstances

Showing disparate patterns is hard: longer average sentences for men could mean that judges are biased or they could just be an indication that men commit crimes that carry longer sentences

Norway is an exception: in an amazing effort, @cbjoerke.bsky.social & Henrik Bøe collected more than 9000 verdicts and linked defendants and judges to the population register and national tax list