I'd say using past papers is advanced stuff. For beginners I'd use range/plausibility (weakly informative), e.g. outcome is blood pressure, you have lm, so e.g. N(150, 30) for intercept and N(0, 15) for between group difference captures the plausible range well enough (dtto for log/logit link).

March 27, 2026 - 06:33 UTC

1
0
1
That's really good advice, I like that.
1
0
1
To expand a little bit - especially on the log scale this often ends up giving surprisingly narrow priors. E.g. N(0, 1) for a group indicator coef could be seen as "informative" but for GLM with log link, it implies 7-fold reduction/increase is plausible - so its _very_ wide for many use cases.
0
0
2