Wide CI for odds ratio

Hello all,

I ran a model that yields odds ratio, like the logistic regression does.

I have received a very wide CI. To illustrate the reason, imagine I have 2 groups, and an event. Let's say that in the treatment group, most subjects succeeded, and some didn't, while in the control group, only 1 succeeded and the rest didn't. By having only 1 observation in the category of success+control, the CI collapse and become wide, especially when the rest of the data is not THAT large (15 for the control in general).

So I ran this model, and got OR=60 (CI: 6,600)

How can I report something like this ? Is there anything I can do, is there a manipulation I can do to get the standard errors down ?


Omega Contributor
As Dason is alluding to, the n-value is in the denominator of the standard error calculation. If you have a larger sample size you will automatically get smaller standard errors.

Not sure if this comes into play but collinearity in terms can increase standard errors. Have you examined for collinearity in the model?
sample size is not large, and worse than that, I had 3 treatments and a binary outcome, and I found that one treatment (one control) is so bad, that it had only 1 success, which is although very good for the research, bad for the model, it makes it "collapse". Like in chi square test, we don't want empty cells...I don't know how to overcome this.


Ambassador to the humans
I don't really know what you mean by overcome it. I mean the answer to that is obvious - you need more data if you want a smaller confidence interval. But it looks like the confidence interval itself doesn't cross 1. What is it about the width of the confidence intervals that bugs you this much?
The fact that it doesn't cross 1 is good, I guess that what bothers me here is the precision. But like you said, apart from getting more data, which is not possible in this case, there is nothing much I can do.