3dLMEr voxelwise effect size for each fixed effect

AFNI Message Board
1

Hello everyone,

I need to present effect sizes for my 3dLMEr results. What is the convention? the -R2 option provides overall model fit, but not individual fixed effects effect sizes. Would converting the Chi-squared to a Cohen's w be the best or running a follow up on the mean of the significant voxels in R?

Andrew

AFNI version info (afni -ver):

Precompiled binary macos_13_ARM_clang: Jan 24 2024 (Version AFNI_24.0.02 'Caracalla')

2

Andrew,

For effect size, are you focusing on the marginal effects estimated using the -gltCode option? I generally prefer using the point estimate along with its standard error (or the associated t-statistic) as a measure of effect size. However, if Cohen's d is preferred for some reason, it can be calculated as d = \frac{t}{\sqrt{df}} , where df is the degrees of freedom for the t-statistic.

Gang Chen

3

That's basically what I was thinking -- I have Chi-squared output tho. Perhaps its because I just used the general fixed effect output instead of -gltCode

When you say point estimate -- what is the Brik label?

4

For each effect estimated using -gltCode, the first sub-brick contains the point estimate for each voxel, while the second sub-brick provides the corresponding statistical value.

Gang Chen

5

I’m sorry, I’ve not heard the term “point estimate” before. Is this the beta weight for the linear regression?

6

You can find the definition of a "point estimate" on this Wikipedia page.

Gang Chen