Hacker News Reader

p values cutoff can't salvage that

as a matter of fact artificially raising the cutoff to make the results sound more convincing is a form a p-hacking

it is exactly the opposite of how p-values work. a smaller p-value does not make a result more reliable, it makes no sense whatsoever to reject a 10^-7 but trust in 10^-8

the misuse of p-values in the world is rampant - the only people that don't misuse them are those that do not use them ;-)

View on HN

See here: https://en.wikipedia.org/wiki/Genome-wide_significance

"The most commonly accepted threshold is p < 5 × 10−8, which is based on performing a Bonferroni correction for all the independent common SNPs across the human genome"

The reason for the low cutoff is to correct for multiple hypothesis testing.

Any what do you mean by "opposite of how p-values work"?

Obviously if P(data or more extreme|null hypothesis) = 10^-8 it's significantly less likely to see the data if the null hypothesis was true, than if P(data or more extreme| null hypothesis) = 10^-7.

Reply on news.ycombinator.com