Issues when running logistic regression analysis

[Jacob Ghotbi]

Dear everyone,

I am currently doing a study on pancreatic cancer, and one of the things I'm looking at is the likelihood of reaching resection based on the type of chemotherapy administered. Based on the crosstab analysis below, you can see that there is a significant difference, but this is something I also have to show in a univariate regression model, as well a multivariate analysis with other variables.

Stata would not perform the regression analysis, as it stated that "1 perfectly predicts faliure/outcome" (see the second image).
I eventually realized what this means: As the crosstab anaylisis shows, none of the patients receiving chemotherapy regimens 2-4 ended up with a resection. Appearantly, this creates an issue when running a regression analysis. Does anyone know how to circumvent this problem?

Thanks in advance

[Nick Cox]

I think the problem is that you've coded your outcome 1 and 2, so all values are treated as positive (not zero). You should feed logit regression outcomes coded 0 and 1.

[Jacob Ghotbi]

Thank you for replying.
Even when I change the coding to 0 and 1, I run into this problem.

[Jacob Ghotbi]

image two:

[Nick Cox]

https://stats.oarc.ucla.edu/other/mu...ith-the-issue/

[Jacob Ghotbi]

Thank you! This explains it.

I tried using firth logistic regression, the Bayesian method and the Exact method, and the Exact method fits the dataset best.

When doing a the analysis with more than one independent variable, an error occurs because I exceeded the memory limit of 10.0M bytes. Is this a limitation of the computer that I am using. And, is there a way of increasing the memory limit?

Thanks

[Nick Cox]

I am surprised to hear that. As I understand it you have 92 observations and even with many variables that's still a tiny dataset.