Biotech Values

[iwfal]

Statistics -

ABT’s PR (#msg-43589307) says Niaspan was statsig better than Zetia on the MACE secondary endpoint.

As you could see from my post there is actually a choice of statistics tests you can run and it makes a substantial (i.e. almost a factor of 2) difference in p value. Fisher's is the test usually run for small numbers of events - but there is a whole saga behind that. There was a decades long feud between Fisher and Bernard - which Fisher won primarily because he was more powerful (the father of clinical statistics) and, by reputation, he was a complete a**h***. Everyone thinks of statistics as an exact science - but it is far far from it. The buried assumptions are where the meat is and just like in clinical medicine some of the assumptions are not what you might think (e.g. Fisher's assumptions appear to have been chosen in large part to make the calculations simpler - which might have been important in the 30's and 40's, but is a complete non-issue with modern computers).

PS Even if you use a given test, e.g. Fisher's, it is still possible to get somewhat different p value depending upon whether you do things like stratification or regression.