- From: Maciej Stachowiak <mjs@apple.com>
- Date: Thu, 26 Feb 2009 17:37:44 -0800
- To: Philip TAYLOR <P.Taylor@Rhul.Ac.Uk>
- Cc: Ian Hickson <ian@hixie.ch>, Steven Faulkner <faulkner.steve@gmail.com>, HTMLWG <public-html@w3.org>, W3C WAI Protocols & Formats <w3c-wai-pf@w3.org>
On Feb 26, 2009, at 3:20 PM, Philip TAYLOR wrote: > > > Ian Hickson wrote: > >>> Doesn't that require that both samples be taken from the same >>> homogeneous population ? >> The fact that first two points are so closely correlated means that >> the two samples probably are both proportional samples of the same >> overall population, which is why we can draw the conclusion from >> the third data point from only one sample. > > We are getting dangerously off-topic, but I disagree. > There may well be a correlation between the prevalence > of lemons and the prevalence of cones, yet no such > correlation between (the prevalence of) either lemons > and apples or between cones and apples. If two independent population samples show similar distribution of many characteristics, Bayesian inference would lead us to conclude that they are likely (though not certain) to show similar distribution of other characteristics. This is particularly the case with traits that are not fully independent but likely to correlate with each other. If we had evidence that one sample or the other had a systematic selection bias relating to the characteristic in question, we might conclude otherwise. Regards, Maciej
Received on Friday, 27 February 2009 01:38:46 UTC