Hi Fabio

If you want to compare a lme model including only a single random effect for the intercept term against the same model but including only a single random effect for the slope term all you need to do is to choose the one with higher maximum likelihood after model fitting.

However if you are using time-since-baseline as your time measure then a model with a single random effect for the slope doesn't make sense. In that particular case the fact that t0(baseline time)=0 for every subject makes the model unable to model correlations between the first and other measurements on the same subject. Therefore, in that particular case, you have to include intercept as a random effect and compare that model with the same model including intercept and slope. If you only have at most 2 repeated measurements for each subject then the only possible model is one with a single random effect otherwise you are over-fitting the data.

lme_mass_LR can only be used with q>=1 which means you can not use it to compare a model without random effects with a model including random effects.

Best -Jorge

---

De: "Bernardoni, Fabio" Fabio.Bernardoni@uniklinikum-dresden.de Para: "freesurfer@nmr.mgh.harvard.edu" freesurfer@nmr.mgh.harvard.edu Enviado: Martes 16 de septiembre de 2014 8:27 Asunto: [Freesurfer] how to choose the best random effect in LME models

Dear Freesurfer Users,

I am trying different LME models to describe the time evolution of cortical thickness in anorexia patients. The LME tools provide a function (lme_mass_LR) to test whether a model with q+1 random effects is better (fits the data better) than one with q random

effects, eg whether a model with random intercept and slope is better than one with only intercept random.

I have two questions: 1- is there a similar function to test whether the model with intercept as a random effect is better than the model with slope as a random effect (ie two models with the same number of random effects)? In another thread I read one should judge which random

effect is to be chosen from the percentage of voxels where one has convergence. It sounds reasonable, but an explanation of why this should be so would be welcome.

2- does it make sense to use lme_mass_LR to test whether a model with 1 random effect is better than one with 0 random effects? I have it with something like:

LR_pval = lme_mass_LR(lhstats_0RE,lhstats_1RE,0);

And I get as a result LR_pval= 0.3 (a constant vector) which I do not know how to interpret: something wrong, this test does not make any sense, choose the 0 random effect model ...

Thanks in advance for your help.

Cheers, Fabio

Dr. Fabio Bernardoni wiss. Mitarbeiter Klinik und Poliklinik für Kinder- und Jugendpsychiatrie und -psychotherapie

Tel. +49 (0)351 458-5245 Fax +49 (0)351 458-7206

Universitätsklinikum Carl Gustav Carus an der Technischen Universität Dresden Anstalt des öffentlichen Rechts des Freistaates Sachsen Fetscherstraße 74, 01307 Dresden http://www.uniklinikum-dresden.de Vorstand: Prof. Dr. med. D. M. Albrecht (Sprecher), Wilfried E. B. Winzer Vorsitzender des Aufsichtsrates: Prof. Dr. med. Peter C. Scriba USt.-IDNr.: DE 140 135 217, St.-Nr.: 203 145 03113

---

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.