I am reviewing a paper for publication, and they have reported an adjusted odds ratio for their intervention variable of 123x10^5 with a 95% CI of (0.00-), p=.994. Is this an error? Should they rerun their model and look for outliers or multicollinearity? They report on the other variables which were significant in the paper, but this number is very odd to me.
Check this paper. If the authors are on RG, you can ask them: Article Explaining Odds Ratio
From what you wrote the data is entirely uninformative (I suppose the upper limit of the CI is +Infinity). Thus, the message is that the data they analyzed is (not too in-)compatible with any OR (no matter how close to 0, no matter how big). The given value is ony a point estimate. There is not much meaing in this value; it is somewhere whithin the limits of the CI, which is infinitively large. Given that, an OR of 10 or 100 or 10^6 or 10^8 or 10^10 would not be much less plausible given the data.
Likely one of the cell values is very small, so the table
33 45
0 32
Will produce an infinity OR (and log odds ratio) and the standard methods produce infinity SE (includes sqrt(1/0 + ...)), Suggest to them adding a small amount like .5 to each cell. This starting value has different names in different disciplines.
They do not know what they are doing, it is your duty to point this out firmly and politely
See answers to this question
https://www.researchgate.net/post/Is_it_weird_to_get_a_very_big_odds_ratio_in_logistic_regression
- Badges
- Science topic
- Similar topics
- Publishing