nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Transforming skewed count data for analysis

Hi Everyone,

I have some heavily skewed count data that I'd like to use to build a model. The data looks much better with a log transform shape wise but being count data, there's an abundance of zeros that won't work with a log model. From reading online, I should be using something like a Poisson distribution but I am not sure how to do so with JMP?

Any guidance would be appreciated, thank you in advanced!

Accepted answers

Peter_Bartell

A couple thoughts for you:

1. You might want to try the Generalized Linear Model personality in the Fit Model platform. You can specify the response distribution (Poisson is an option) and link function (log is one of the options) as part of the model specification window. Here's the link to the JMP online documentation for the Generalized Linear Model personality:

https://www.jmp.com/support/help/14-1/generalized-linear-models.shtml#

2. If you have JMP Pro, within the Fit Model, Generalized Regression personality, there is a zero inflated Poisson distribution option for situations such as you describe...a predominance of zeros within the response population. Here's the link to the JMP Pro online documentation for the Generalized Regression personality distribution notes:

https://www.jmp.com/support/help/14-1/distribution.shtml

All comments

Paul_J

I often use the Gamma-Poisson to fit the a count distribution with peak at zero. I like the JMP help on this as well.

Peter_Bartell

A couple thoughts for you:

https://www.jmp.com/support/help/14-1/generalized-linear-models.shtml#

https://www.jmp.com/support/help/14-1/distribution.shtml

Thanks Peter! I will check out the GLM link.