Mixed effect logistic regression- Disagreement between p-value and confidence interval

[Nicole Carbert]

Hi all,

I just ran a mixed effect logistic regression model and found a significant p value but when I look at my confidence intervals they cross 1 (.37-2.47). Both variables are dichotomous (0 and 1). The sample size is 377 with 263 groups.

I ran the following:

xtmelogit mg_kid120mindichotomous mgareain_dichot || facilityidnumber: , intpoints(10)

mg_kid120mindichotomous= DV
mgareain_dicho=IV
facilityidnumber=group variable

Do you have any suggestions as to how I should interpret these results (e.g., should I go with the p-value or the CI)? Or is something completely wrong with this model?

Any help would be greatly appreciated!

Thanks,

Nicole

[Clyde Schechter]

There is nothing obviously wrong with your model. I think you are misinterpreting the output. As you did not specify the -or- option, Stata by default reports the regression coefficients (and their confidence intervals), not odds ratios. That confidence interval does not include 0, and so it is consistent with the coefficient having a significant p-value.

If you prefer to see your results in the odds ratio metric, use the -or- option.

By the way, if you are using Stata version 13 or later, the command -xtmelogit- has been renamed and is now -meqrlogit-. The old name still works, but at some future time it may not, so best to start using the current terminology.

[Nicole Carbert]

Great, thanks so much for your help Clyde!

[Hassen Ali]

Thank you very much Clyde Schechter, I have learned a lot from your reply!!
Good News,I will start using the command -meqrlogit-.
With Best Wishes,Hassen