Biotech Values

[iwfal]

Interesting analysis from a biostatistician.

1) I am familiar with the site - a statistics curmudgeon site (in the sense that there is always something wrong). That said, it is so extreme I suspect that there is some humor intended - or perhaps that he is doing it as a mental challenge.


2) It is also a site that has, like many statisticians, a large focus on increasing trial power. E.g. through the use of tools like ANCOVA or ... . The problem is that such tools come with penalties from buried assumptions (that aren't always true) or, worse, have models with (internal or external) dials that create p-hacking types of issues. ANCOVA is a good example since it has a ton of buried assumption (normality etc). There is a good reason the FDA likes categorical variables and non-parametric tests (both things that would generally be frowned upon by this site) - because they are generally much less subject to either post hocing or other forms of exaggerated p values.

I would suggest that we already have a problem with way, way, way too many published papers with false or very exaggerated positives and moving more in the direction that blog generally suggests would increase that problem substantially.