Logging hypotheses and protocols before performing research seems to work as intended: to reduce publication bias for positive results.
