[naasking]

Social sciences are by far the worst offenders, with sociology right at the bottom at ~30% replication rate.

> Seems like we need people who study societies to help figure it out.

No, we need people who understand robust quantitative analysis and empirical methodologies. Sociologists are clearly not them. Open science and preregistration initiatives help a lot, but I should note that social sciences were also the most vociferous objectors to such changes.

[wpietri]

Did you look at the article I linked? Even many chemists and physicists think significant portions of their fields have results that can't be reduced. Biology and medicine even more so.

Even if the problem were related to robust quantitative analysis, the question of why so many people are failing to apply particular methods isn't a question for physics or chemistry.

[naasking]

> Even many chemists and physicists think significant portions of their fields have results that can't be reduced. Biology and medicine even more so.

Yes, I'm aware of the reproducibility numbers, that's why I said social science was the absolute worst, by far. Despite only 50% of medical papers replicating, they've still provided significant improvements over the past few decades. Ditto with chemistry.

The same cannot be said for sociology and psychology. CBT is still the tool with the best track record and that was created in the 1960s. The robust, replicated results from sociology since the 1980s have modest effect sizes and are basically things we already knew, eg. that social isolation is not good, and that social norms influence people's behaviour.

> Even if the problem were related to robust quantitative analysis, the question of why so many people are failing to apply particular methods isn't a question of physics or chemistry.

Neither is it a question of sociology or psychology. Science is a systematic, self-correcting process that should be agnostic to any factors, including human fallibility, conflicts of interest, AI generated nonsense, etc.

[wpietri]

That's one way to look at science although it's a bit religious for my tastes. But it if were fully self-correcting, we wouldn't be having this discussion. Understanding why it isn't and how it fix it are social science problems.

[wpietri]

Oops! s/reduced/reproduced/