Yes, that’s suspicious. Good instinct. I’m sure there’s some bias against publishing a marginally-significant result that’s got a low (outside the framework of the paper’s statistical model) prior. I’d bet some of the unlucky ones got file-drawered, and others (dishonestly or not) kept on collecting more data until the noise (I presume) was averaged down.
However, you might be missing that on an iso-P contour, false positives have diminishing effect size as sample size increases.
Yes, that’s suspicious. Good instinct. I’m sure there’s some bias against publishing a marginally-significant result that’s got a low (outside the framework of the paper’s statistical model) prior. I’d bet some of the unlucky ones got file-drawered, and others (dishonestly or not) kept on collecting more data until the noise (I presume) was averaged down.
However, you might be missing that on an iso-P contour, false positives have diminishing effect size as sample size increases.