4 ms·
I don’t know, the findings about African Americans seem to be a case of “outcome switching” [0] since it does not appear that this trial was specifically design
by tarr11 8y ago
I don’t know, the findings about African Americans seem to be a case of “outcome switching” [0] since it does not appear that this trial was specifically designed to test against that group.
[0] https://www.vox.com/2015/12/29/10654056/ben-goldacre-compare-trials https://www.vox.com/2015/12/29/10654056/ben-goldacre-compare...
- LeifCarrotson 8y agoA more approachable description of why this is bad: "Green Jellybeans cause acne, p=0.05" https://xkcd.com/882/ https://xkcd.com/882/ No, outcome switching is not exactly the same as P-hacking, but either could be blamed here.
- Dylan16807 8y agoBut on the other hand if you do the equivalent of 20 parallel trials and one of them has a naive p value of .001 then you still have a good result. I don't know how what heart attack values normally look like, but those percentages are pretty huge. They sound like numbers that might stand even after you compensate for the subsetting.