topic Difference in Risk Ratio stats between JMP 15 & JMP 16? in Discussions

Difference in Risk Ratio stats between JMP 15 & JMP 16?

sfigard — Fri, 09 Jun 2023 00:57:05 GMT

OK, have I got a question for y'all! I'm teaching survival analysis and looking at Proportional Hazards Fit and Risk Ratios. In JMP 15 I'm getting a Prob>Chisq of <.0001 for my risk ratios (yes, it's an odd dataset where the differences in survival are rather large) with understandable 95% confidence limits. I've assumed this is testing the Ho that the risk ratio is not significantly different from a value of one. (Could be wrong there.) In JMP 16, however, evaluating the exact same data, I'm getting a Prob>Chisq of 0.9987 and zero for the lower 95% CI and nothing for the upper. Can anyone explain what is going on here?

Re: Difference in Risk Ratio stats between JMP 15 & JMP 16?

Mark_Bailey — Thu, 10 Nov 2022 19:58:03 GMT

Steve, I would report this anomaly to JMP Technical Support (support@jmp.com). It looks like a bug to me. Could be a code change that caused a regression. We usually have unit tests to catch such things.

I notice that the Wald chi-square is calculated correctly but it is way different from the likelihood ratio chi-square, which is also calculated correctly. I also notice that you have complete separation in the response, which can be a problem case.

What if you treat it as a nominal logistic regression?

Re: Difference in Risk Ratio stats between JMP 15 & JMP 16?

sfigard — Mon, 14 Nov 2022 19:24:29 GMT

Mark,

Thanks for taking the time to look at this. The data itself is artificial and used for teaching survival analysis, so it is designed to have a significant difference (although it is a rather large difference). In the past there hasn't been a problem interpreting the numbers. I'm not sure how one would treat this as a nominal logistic regression given the structure of the data. I will send the question over to tech support and see what they can come up with.