[ExI] Statistical tests

[rex]

Anders Sandberg <anders at aleph.se> [2013-11-13 06:52]:
>But even sharp p-values can be useless if effect sizes are not 
>stated. If a medicine reduces the length of colds at p=0.000001 but 
>has an effect size of only 1%, it is useless in practice. A 
>surprising number of papers downplay or hide effect sizes.

Surprising until one recalls, "Publish or perish." What fraction of
authors will (correctly) note, "X is related to Y at the p=0.000001
level, but the effect is utterly insignificant."

Who wants to be known as the fellow who finds lots of useless, but
statistically significant, relationships?

-rex
-- 
Tales are like trees: both grow taller over time.