The Null Ritual (2004) [pdf]
library.mpib-berlin.mpg.de
library.mpib-berlin.mpg.de
However, an explicit estimate of predictive power (sensitivity) would be very useful too - confidence intervals do not provide information on the model used to derive them which may or may not fit. Instead one has to use proxies like experimental design descriptions and N.
The important part is to choose the correct hypothesis to test. However, there is no requirement to use the effect size to test any hypothesis, although eventually someone should come up with a model that explains why it has the value it does.
> "as opposed to significance which is information about validity of data."
Huh?
Such as by chance or because the alternate hypothesis is bunkum or because your experiment lacks power. (It does not distinguish between these, and middle item is only minimally handled by most waste of obtaining it, esp. not by Fischer's or ANOVA which are facts about data only, using error in measurement model.)
Thus p-value is a statement about validity of data (type 1 error, false positive) and maybe about the null hypothesis as a secondary notion.
(E.g. about the control group of you use that as reference group for null.)