Fit ZI Negative Binomial results change (from JMP 16 - JMP 17/18 )

Question

My client previously conducted analyses using JMP 16 before the release of JMP 17, and they have recently performed the same analyses using JMP 17. (JMP 18 and 17 has same results)

They used the "Analyze – Distribution – Discrete fit – Fit ZI Negative Binomial" function on the provided data. However, they noticed from the accompanying images that JMP 16 (left) and JMP 17 (right) produce different AIC and BIC values.

I am wondering if the formulae for calculating AIC and BIC for the ZI Negative Binomial distribution were modified with the version upgrade? how can i track this?

Additionally, while JMP 16 version estimates both the parameters of the ZI Negative Binomial distribution and the standard errors of these estimates, JMP 17 only provides the parameter estimates and omits the standard errors. Could it be that the method for calculating the standard errors of the parameter estimates has been modified with the upgrade, making it impossible to calculate standard errors for the attached data in the newer version?

or what am i missing..?

I posted the raw-data as attachement

Temp.jmp

HyeWonJung · Answer

Thanks for prompt reply. I cannot see Jira as i don't have access. Can I understand Version 16 was correctly calculated, but 17&18 has bug in terms of develop issues and from 18.1 and 19 will be fixed? I want to provide detailed response to my customer.

clay_barker · Answer

Thanks for the heads up!

As the Jira that @mia_stephens linked to mentions, it's a bug in the fitting algorithm's starting values. We've got it fixed in 19 and will get the fix into 18.1 as well.

mia_stephens · Answer

Thank you HyeWon, and thank you for submitting to support@jmp.com.

This Jira has been opened: https://jmpdiscovery.atlassian.net/browse/JMP-20852

Mia