# 3dISC and effect size calculation

Hello,

I was hoping to calculate effect sizes for the continuous (quantitative) variables in a model I created with 3dISC. What would be the best way to go about this? Is there a certain measure of effect size (maybe Cohen’s f^2?) that would be the most appropriate for the kind of regression it uses (linear mixed model w/ crossed random effects)?

Also – is there a way to get an r^2 (coefficient of determination) value for a model made through 3dISC?

Thank you AFNI folks, I appreciate all the advice you provide on this forum!

Ryann

Ryann,

What would be the best way to go about this? Is there a certain measure of effect size (maybe Cohen’s f^2?)
that would be the most appropriate for the kind of regression it uses (linear mixed model w/ crossed random effects)?

I assume that your input files are correlation values. It would be difficult to directly obtain the effect size in the original physical unit. You may consider using Cohen’s d by converting the t-statistic in the output.

is there a way to get an r^2 (coefficient of determination) value for a model made through 3dISC?

Unfortunately R^2 is not clearly defined under the linear mixed-effects modeling framework even though there have been proposals (e.g., https://arxiv.org/abs/2007.08675)

Hi Gang,

Thanks for your reply. The inputs are the correlation maps, yes, that I generated with 3dTCorrelate. I was under the impression that Cohen’s d was for measuring the effect size of a difference between two groups? I am wondering if there is a way to measure an effect size for a quantitative variable in my model (i.e., head motion). Or is Cohen’s d still the most appropriate for this?

To get Cohen’s d from the t-stat map - you would divide the t-stat map by the number of subjects, correct? Would that be the number of individual subjects (i.e., 81) or the number of pairs that went into the 3dISC model (i.e., 3240)?

Thanks for all your help, I really appreciate it!

Ryann

Ryann,

Cohen’s d can be interpreted as the standardized effect that is independent of sample size. So, you can obtain Cohen’s d for any simple effect in a model including a covariate such as head motion.

When the t-statistic value is available, you can directly calculate the corresponding Cohen’s d: divide the t-stat by the square root of the sample size. Use the number of subjects as your sample size in this case.

Thanks for all your help Gang, much appreciated!