Friday, April 11, 2008

Economist survey - An important caveat

We'll revisit the the Economist survey over the weekend: this time it's the turn of the "religion" questions. However in the mean time...

PJ, guru in all matters statistical, adds an important caveat:

"The errors on that survey must be pretty large - and splitting by political affiliation will end up nigh on meaningless. Not to mention the myriad other limitations of survey data. But at least they do the very minimum necessary to take a survey seriously - i.e. provide full questions and answers, and details of the sample. Didn't see a response rate though - but that is widely seen as the Achilles heel of survey data (making, for instance, the published confidence intervals meaningless)."

[Incidentally, if you haven't read PJ's brilliant series of blogs about the controversial Kirsch et al. meta-analysis on antidepressants, why on earth are you wasting time reading my blog? - go on, it's much more interesting, and much more important, than anything you'll find here.]

No comments: