Thread by @jholbo1

I'm thinking of including this in my segment on p-values next sem. Lab-perfect example of p-values not telling us the likelihood that something happened 'by chance'. They tell, at most, relative to a model - in this case a suspiciously vague 'normal'. votepatternanalysis.substack.com/p/anomalies-in…

It is hard to think straight about such problems, partly due to absolute cosmic unlikelihood of anything, in particular, happening - rather than some other thing. When every event is unlikely - but aren't they all? - you get a sort of one-size fits all cosmic paranoid style .

What do people think of it? I mean: set aside that anyone beavering away at proving fraud is probably severely confabulating at this point, it seems like a nice example of garden variety crude analysis, showing how genuinely hard good analysis of this sort is. Baseline agonistes.

I'm trying to set aside my contempt for the 'stop the steal' crowd to be objective about how much 'weight' to give. My specific reason is that, in teaching, I talk about the relationship between the 'huh, that's kind of weird' feeling and actual assessment of likelihoods.

P-hacking is methodologically bad but life is p-hacking, insofar as oddities grab your eye.