LME and multiple comparisons

Hello, I have a model with several categorical fixed factors, each having several levels. Fitlme reports confidence intervals and p values for n-1 levels of each fixed factor compared to a selected reference. How can I determine these values for other combinations of factor levels? (e.g., level 1 vs. level 2, level 1 vs. level 3, level 2 vs. level 3).

Thanks, Chen

Answer 1

[Peng Li]

If I understand your question correctly, you were asking a post hoc comparison issue? coefTest method of LinearMixedModel object could help with this. basically you need a comtrast matrix to indicate which two levels you'd like to compare.

https://www.mathworks.com/help/stats/linearmixedmodel.coeftest.html

Here is reference page for coefTest.

Answer 2

[Paul]

Hi Chen,

You talk about confidence intervals and p-values, and how you can get those for other combinations/comparisons of levels. The answer about p-values I have given recently, here.

With regards to confidence intervals I'm not sure I can give you a satisfying answer, but I first want to ask you to clarify your question a bit more. Do you mean the confidence interval of the difference between level 1 and level 2, for example? There are ways to calculate this, but I'm not sure if this is at all useful in practice. I found it hard to determine a best way of doing this, and I found I had very little need for it in the end, so I never figured out exactly how to do it.

If you could explain what kind of confidence bounds you are talking about, exactly, and what you want to use them for, I can see if I can help you.

Best,

Paul