3 ms·
A major problem with this kind of armchair analysis is that in practice variables are only ever approximately normal, and the differences are most noticeable in
by klodolph 9y ago
A major problem with this kind of armchair analysis is that in practice variables are only ever approximately normal, and the differences are most noticeable in the tail.
Taleb's book The Black Swan talks about this phenomenon, but I can't recommend the book--it's basically one chapter's worth of material stretched over 400 pages.
- tptacek 9y agoDarren Dahly is a biostatistician and lecturer at University College Cork. I don't think you get to dismiss his analysis as "armchair".
- klodolph 9y ago"Armchair" in this case means "without looking at the data". We're making hypotheses about the tail of a distribution based on the assumption that the distribution is normal, rather than looking at the data. It's still a problem with the argument, no matter who makes it. It sounds like you're saying that it's more important to know the identity of the person making the argument than the content of the argument itself?
- tpeo 9y agoI don't think would be a problem of armchair analysis. Rather, it would be an issue of trying to fit the wrong distribution to the data. Granted, not having even a peek at the data would increase the likelihood that the distribution someone pulled out of their hat is wrong. But it doesn't prevent people from making that mistake. I don't think the tail of the techiness distribution would be an issue either, because this tail isn't that fat. Techiness is most likely constrained to a definite variance. I mean, Googlers are part of that tail, yet none of them are some sort of post-singularitarian cyborg, beyond the comprehension of their own colleagues. It might sound stupid, but this is the sort of thing we expect to see in the fat-tailed distributions.
- Mattasher 9y agoTo fill this out with an example, adult male heights are normal for most of the distribution, but we have a handful of people who've grown to 8 standard deviations beyond the mean, and a large cluster of people even farther below the mean.