Firth-adjusted GLM logit regression and confidence intervals

[#1]

Goodmorning,

I ran a logistic regression model, through a GLM with binomial distribution and logit link function, with Firth adjustment as I got a warning on quasi-separation of data.

I need to assess if my variable of interest (CHIP) is associated to the outcome (I have a binary response variable), in a model adjusted for other covariates.

Now, I run the model, and I get a parameter estimate with a significant p-value, but a confidence interval for the same estimate that is large enough to include the value 0.

My urgent question is:

How should I interpret that? Can I make inferences or should I be more cautious?

Thank you!

Ps. I know the model could be poorly fit, but that is the best I can do at the moment with the data I have

[#2]

I forgot to attach a screenshot. Here it is (it is in Italian)

[#3 · Accepted solution]

JMP help documentation tells us that these are profile-likelihood confidence intervals. I understand that you can expect some discrepancy between these and the ChiSquare p-values.

In this case the discrepancy is not very big. The p-value for this effect is very close to 0.05 and your confidence intervals only just include 0.

So while you might use the p-value < 0.05 to say that CHIP is "significant", you can't say that the confidence intervals are telling you something completely different!

[#4]

Thank you Phil, you really helped!