While the latest case if more about very LARGE effects, it may reflect potential publication bias against null results.
Since 2000s, the share of claims that are null have declined in almost all fields.
Since 2000s, the share of claims that are null have declined in almost all fields.
Comments
(@i4replication.bsky.social ) and coauthors make a compelling point by studying p-hacking
RCTs and RDDs have relatively high null results across all fields. Typically, with RCTs it is harder to do specification mining to get your desired large result.
Sadly, the scandal shows there may be other research malpractice that can get you there.