Questions on LMM
Dear community,
I have some practical questions regarding LMM in JASP:
1) If my model fails to converge ("Model fit is singular") with random slopes (selected as the main effects), I understand that I can try to simplify the model. For instance, if I have two factors with two levels each, is it correct to use the "variance estimates" or the "random effect estimates" option and first remove the factor with the lowest variance to determine which one to exclude first?
2) Regarding the contrast option in the marginal means section, I find it puzzling that, regardless of the p-value correction I choose, the 95% CI for the contrast remains exactly the same. Shouldn't the CI change based on the correction applied, or am I misunderstanding this?
3) I can't find a way to display for instance the Q-Q plots of residuals and test the normality among the variables. Would this come in a next version?
4) For my contrast, the option of displaying the Cohen's d is not available. But I guess I can calculate it as any Cohen's d based on the data from the contrast? Or if it does not appear in the options this is for a specific reasons?
Thank you!
Comments
I've asked our expert!
Hi Emily,
1) I think that's quite an acceptable approach. (There are different suggestions for how to go about this, but there is no clear guidance. Other are also not accessible from JASP, e.g., covariance matrix eigen decomposition etc.)
2) There are different opinions on that. Some would argue that CIs should be corrected too, some not. I don't personally believe in the CI adjustment.
3) Yes, that is a great suggestion and it is on our roadmap.
4) You can use the output to compute Cohen's d manually. We didn't implement effect size calculations for (G)LMMs yet because there are many different ways of standardizing the mean differences by total/within/between subject variances etc. It is also on our roadmap though!
Cheers,
Frantisek