Showing posts from February, 2011

Wagenmakers-Bem: Round 3

Continuing what may turn into a 12-round prize fight, Eric-Jan Wagenmakers was kind enough to send me a link to his team's response to the Bem, Utts & Johnson reply:

Bem et al reply to Wagenmakers et al

Daryl Bem, Jessica Utts, and Wes Johnson have written a reply to the Wagenmakers et al Bayesian analysis of Bem's precognition experiments. You can download it from here . It explains why Wagenmakers et al's conclusion of a null effect is, well, just plain wrong. Utts and Johnson are statistics professors at U C Irvine. Both are authors of popular statistics textbooks, including one on Bayesian methods. In general, if Bayesian and frequentist methods lead to wildly different conclusions about the same underlying dataset, then something is out of whack. In most cases one likely reason is that the Bayesian enthusiast has selected impossibly low Bayesian priors. Bem, Utts and Johnson clearly show that this was the case for Wagenmaker's article, and that a more reasonable choice of priors leads to strong evidence in favor of precognition.