Intraclass Correlation using a random effects model versus the MSA platform

[#1]

Good evening JMP Community,

Does someone have an explanation as to why the ICC calculated in the MSA platform (EMP) is different than the percent of variance explained in the variability or the fit model platform with random effects (REML)?

For example, I used the Animal sample data and concatenated the species and subjectID into a new column and used the new column as the between subjects factor and miles as the outcome.  In the fit model and variability platforms I calculate an ICC of ~0.387 and in the MSA platform the ICC is ~0.4.

Attached is a pic of the two outputs.

Sincerely,

MG

 

[#2]

its all about the sigmas

https://www.jmp.com/support/help/en/15.1/#page/jmp/statistical-details-for-measurement-systems-analy...

The REML variance components are calculated differently, and perhaps a little more precisely. 

Take a look at the degrees of freedom in EMP too.

[#3 · Accepted solution]

Hi MG,

The EMP Measurement Systems Analysis platform uses ranges by default for calculating variation, but if you switch to using standard deviations in the launch dialog you should get results that match the Variability platform in your example.

Here are the results:

It is possible to set a platform preference to always use standard deviations if you prefer that to be the default.

I hope this helps. Thanks.

Laura

[#4]

Many thanks Byron and Laura.  I appreciate the quick support!

Sincerely,

MG

[#5]

Thank you @Laura_Lancaster for the clarification.

would it be possible to produce a list of ICC measurements available in JMP in the context of this list:
https://en.wikipedia.org/wiki/Intraclass_correlation#Calculation_in_software_packages 

Many thanks,

Ron