P-VALUES
(thread)
1/ There’s this paper trending today claiming p < 0.001 on drug effectiveness even with:
- controls selected not randomly
- controls treated in a different 🏥
- patients ending up in ICU removed from sample (!)
That any of those factors has p < 0.001 is BS.
(thread)
1/ There’s this paper trending today claiming p < 0.001 on drug effectiveness even with:
- controls selected not randomly
- controls treated in a different 🏥
- patients ending up in ICU removed from sample (!)
That any of those factors has p < 0.001 is BS.
2/ I’m not naming the paper because it’s not about that specific paper, but about methodology in general.
The point is: any factor not controlled for should have its probability of being significant deducted from the p-value,
or some similar adjustment.
Thoughts?
The point is: any factor not controlled for should have its probability of being significant deducted from the p-value,
or some similar adjustment.
Thoughts?
3/ This would be in addition to the points presented here:
4/ Quiz: you take 200 sick people. 100 you put in hospital A & give a drug. 100 you put in hospital B and give no drug (they’re the control)
The first group shows much better improvements than the control, p < 0.001
What’s the probability that’s fully explained by the hospital?
The first group shows much better improvements than the control, p < 0.001
What’s the probability that’s fully explained by the hospital?
5/ How does the answer to the above change the p-value?
(Rhetorical quiz; for full context see the first tweet of the thread.)
(Rhetorical quiz; for full context see the first tweet of the thread.)
