Message-ID: <a9573a85cac442a19cee5eacb0411eee@qimrberghofer.edu.au>
Date: 2020-08-23T07:14:46Z
From: David Duffy
Subject: CLMM: Calculate ICC & Assessing Model Fit
In-Reply-To: <DM5PR0101MB308296310ECEA2920A7A5B29AB5A0@DM5PR0101MB3082.prod.exchangelabs.com>
> The distribution across the 9 levels appears rather haphazard, which dissuaded me from trying a GLMM:
Maybe. A certain amount of that can be soaked up by different intervals between thresholds in the probit-normal - floor and ceiling effects on your scale for the most extreme categories is one example. The equivalent of this for an ordinary LMM is an inverse-normal transformation of the scores (normit/rankit). This sometimes seems a bit lazy to me, but as I commented earlier, you may end up with similar results to the more elaborate models. Worth doing in parallel, at least.
Cheers, David Duffy.