0

I am working with a binomial dependent variable (fail=1, not fail=0), and using ratios as independent variables to predict the outcome.

My dataset is n=34, so it isn't. I'm using R. When I use the binomial family with anything more than one predictor, I get a fail to converge warning and the output has humungous standard errors and coefficients, as well as z probabilities that are essentially 1.

When I compute the odds ratios, I literally get either 0 or 1. However, if I use Gaussian regression, the coefficients and p-values look, well, what I would hope to see and the odds ratios look believable.

My question is: can I actually use Gaussian in this case?

  • 2
    search for information on complete separation, e.g. https://stats.oarc.ucla.edu/other/mult-pkg/faq/general/faqwhat-is-complete-or-quasi-complete-separation-in-logisticprobit-regression-and-how-do-we-deal-with-them/ – Ben Bolker Mar 12 '24 at 16:37

0 Answers0