By Keith Laws
Originally published February 27, 2013
Here is an excerpt:
Psychologists find significant statistical support for their hypotheses more frequently than any other science, and this is not a new phenomenon. More than 30 years ago, it was reported that psychology researchers are eight times as likely to submit manuscripts for publication when the results are positive rather than negative.
Unpublished, "failed" replications and negative findings stay in the file-drawer and therefore remain unknown to future investigators, who may independently replicate the null-finding (each also unpublished) - until by chance, a spuriously significant effect turns up.
It is this study that is published. Such findings typically emerge with large effect sizes (usually being tested with small samples), and then shrivel as time passes and replications fail to document the purported phenomenon. If the unreliability of the effect is eventually recognised, it occurs with little fanfare.
The entire story is here.