The five filters between a study and a decision

A five-stage diagram of how publication bias happens inside a company: the fielding, write-up, escalation, narrative and memory filters. At each stage a null result is more likely to be lost than a positive one, without anyone suppressing anything. Use it to audit where your own research programme leaks findings. Koji closes the write-up filter by producing a structured report for every study as responses arrive.

Where this appears

• Publication Bias and the File-Drawer Problem in Product Research: Why Your Evidence Base Only Remembers the Studies That Worked (2026)

What's in this process

1. The fielding filter: Studies that seem unlikely to give a clear answer get dropped before they run, so no record exists.
2. The write-up filter: The study ran and the data sits in a tool, but no one built the deck. Usually the largest leak.
3. The escalation filter: The readout exists but never leaves the research channel, because a no-difference slide is hard to lead with.
4. The narrative filter: A flat primary outcome is pushed aside for an interesting secondary finding.
5. The memory filter: Six months on, the finding cited in planning is the one that made a good story.