(2018-03-22, 08:02 PM)fls Wrote: Effect sizes which are visible to the naked eye (i.e. found on informal observation) are likely to be "large", but at a minimum are "medium" (ref. Cohen, Statistical Power Analysis for the Behavioral Sciences, 1988 ed. pg 25-27). So when a "small" effect is ruled out, it rules out the possibility that the informal observations were identifying a valid effect (note that the "% relinquish" measure had a greater than .9 power to detect a small effect).
Of course it doesn't. Anecdotes relate to remarkable behaviour by individuals. A statistical study of the whole population may not be able to detect such behaviour, if it's exhibited by only a minority of individuals.