normative

Abandon observational designs for tiny-signal nutrition questions

For questions where the noise is many orders of magnitude larger than the signal—such as single nutrients linked to disease outcomes—observational designs should be abandoned because they cannot recover such tiny effects; randomized or experimental trials minimizing confounding are needed, and continuing observational studies is largely a waste that may bias our thinking rather than provide preliminary insight.

normativepending

Speaker

John Ioannidis

Evidence Quote

“I would argue for this type of questions, where the error is 50 times bigger than the signal, we need to find designs that minimize the error.”

Source

John Ioannidis on Statistical Significance, Economics, and Replication— EconTalk
Created: 6/13/2026, 12:20:10 AM

Your notes

Loading notes...