(Total Views: 437)
Posted On: 09/16/2020 11:44:52 AM
Post# of 155849
"Odds ratio" vs. "difference in proportions" tests (for stat geeks)
I've noticed more papers using an "odds ratio" test for significance, and I have found a calculator for it: https://www.medcalc.org/calc/odds_ratio.php
Roughly, the odds ratio test computes its confidence intervals based on the logs of the probability, not the probabilities themselves. It's a weird number to me - not the same as "relative risk," which means what it sounds like it means.
Fortunately, the results are very little different: Assuming 195 pats split 130/65 and a 50% reduction in death rate, 2-sided p-values are below.
P(die|control) = 25% = 16, P(die|LL) = 12.5% = 16
=> dif in props: p=.043
=> odds ratio: p=.032
P(die|control) = 50% = 32, P(die|LL) = 25% = 32
=> dif in props: p=.0003
=> odds ratio: p=.0004
I've noticed more papers using an "odds ratio" test for significance, and I have found a calculator for it: https://www.medcalc.org/calc/odds_ratio.php
Roughly, the odds ratio test computes its confidence intervals based on the logs of the probability, not the probabilities themselves. It's a weird number to me - not the same as "relative risk," which means what it sounds like it means.
Fortunately, the results are very little different: Assuming 195 pats split 130/65 and a 50% reduction in death rate, 2-sided p-values are below.
P(die|control) = 25% = 16, P(die|LL) = 12.5% = 16
=> dif in props: p=.043
=> odds ratio: p=.032
P(die|control) = 50% = 32, P(die|LL) = 25% = 32
=> dif in props: p=.0003
=> odds ratio: p=.0004
(0)
(0)
