I don't know how to explain the failure to replicate, but it is duly noted

could be a sample size issue. if the first study had middling power, you wouldn't necessarily expect identical studies to replicate. plus the effect size was almost certainly overestimated (consistent with the results of the first replication) due to the beavis effect or "winner's curse".

http://www.genetics.org/cgi/cont...full/165/4/ 2259
http://www.ajhg.org/AJHG/abstrac...9297(07)61097- 0

I posted this in the open thread earlier, but it's somewhat on topic about a possible cure for Alzheimers.

From the Need et al. non-replication:

From Papassotiropoulos et al. we can infer that the genotype explains 6.2% of total variation for the 5-min delayed verbal recall in Swiss cohort 1, and 4.1% for the 30-min delayed VLMT in the US cohort. We note that our Genetics of Memory cohort was 90% powered to detect an effect size of 3.2% at a significance of 0.05, and the German cohort was 90% powered to detect an effect size of 3.0%. Any effect size greater than 1.7% should have produced a P-value less than the observed one with a probability of at least 95%.