3 ms·
Are you saying 32,000 people isn't a large enough study to be meaningful?
by cestith 5y ago
Are you saying 32,000 people isn't a large enough study to be meaningful?
- epistasis 5y agoFor low-prevalence events the number of events that happened is just as important for statistical power as the big 32,000. Especially when doing relative risk, the confidence interval on that "27 times" could be extremely large.
- peteradio 5y agoTrue, but we aren't comparing a 1 to 0 or 1 to 2. Its 238 to 19. sqrt(238)=+/-15 sqrt(19)=+/-4 (if the numbers are to be believed) . 238/19=12.5 (238-15)/(19+4)=9.7 so it becomes 20% less on the lower end of the fluctuation which is only nominal.
- Hinrik 5y agoIt's more about the small effect sizes of 8 people vs 1. Still a lot of room for noise in the result.
- cestith 5y agoTrue, but where's a better study at the moment? I think it would be easier to gather more studies and do a cross-study meta analysis rather than recruiting a million people into a single study to eliminate the effects of noise. The "27 times" is factually their finding in one study, which someone might read as being given too much weight in the article but doesn't change the data.