Hacker News Reader

Comment by cscheid | Hacker News Reader

This kind of spatial scan statistics is really easy to misuse without careful multiple-comparison corrections. Would they have reported a 52% effect? a 51% effect? What about 1.4 inch bands? or 1.2 inch bands? That "remarkably thin" 0.3 percent will disappear pretty quickly.

Sorry that this is dismissive, but this piece is just a slightly more complicated case of http://xkcd.com/882/

View on HN

It's not a statistical model for future predictions. The whole point here is to over-fit as much as possible, because you only care about a single dataset.

Sorry, but if that were the case, then just showing the list of pages and their locations would get 100% of them, with 0% probability of being "by chance".

Then, you say "that's a complicated rule: no one can memorize the list", and that's fine. But finding the simplest rule that has the best fit can be shown, via the Minimum Description Length principle, to be equivalent to models with deep connections to statistical learning.

The gist of the matter is: even if you have a single dataset, statistical considerations are very important to make sure that the statements you make are sensible. That does not appear to be the case in this Slate piece.

He made no claim to minimal or maximal anything. His technique does give you an edge, though.

Agreed. It would have been more interesting if he'd developed his model/heuristic based on the first five or so books and then tested it on the remaining two to see if it held up.

Reply on news.ycombinator.com