Original title: Are geographical and socioeconomic contexts exchangeable? On simple means versus random effects in the study of health inequalities
Juan Merlo, Kani Khalaf · September 2026
Health researchers often use a statistical tool called random-effects models to study health gaps between different groups. This tool assumes exchangeabilityThe assumption that groups are interchangeable, so their averages come from one shared distribution.. That means it treats all groups as interchangeable samples pulled from one shared bucket. This math was built for repeated body tests on one person, where a shared biology makes that assumption work. Later, researchers quietly started using it on neighborhoods, schools, and social categories. The authors show this tool fails for society. People are not just repeated measurements of a group. Also, social groups are shaped by real forces like racism, not random chance. When the model pulls odd or small groups toward the average, it actually erases the very inequalities researchers want to see. Instead, the authors recommend using Simple-Means MAIHDA. This method calculates each group’s results using only its own people and shows the true uncertainty openly instead of smoothing it away.
Imagine you are a teacher grading essays from different classrooms. You assume all classrooms are basically the same, so you grade each essay by pulling it toward the school average. But one classroom has a student who writes brilliantly, and another has a student who struggles. By pulling everyone toward the average, you erase the standout and the struggler. In this paper, the researchers argue that health researchers do the same thing with social groups: they assume groups are interchangeable, so they shrink away the very differences that show inequality. They recommend grading each classroom on its own merits instead.
Health researchers use models that assume neighborhoods and social groups are interchangeable, like lottery draws. This assumption can shrink or hide the small, deviant groups where inequality is most visible. The paper argues that treating injustice as random noise makes it seem natural rather than something to fix.
The abstract does not provide empirical examples or quantify the impact of shrinkage. It focuses on conceptual critique and recommendations, but does not test S-MAIHDA against RE models in real data. It also does not address practical implementation challenges or when RE models might still be preferred beyond the listed uses.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Are geographical and socioeconomic contexts exchangeable? On simple means versus random effects in the study of health inequalities — Juan Merlo, Kani Khalaf — SocArXiv — doi:10.31235/osf.io/dhwrx_v2
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.