Hi Francois

I think that you missunderstood a point of my previous answer.  You should always include ALL subjects (those with 1,2,3,4... and so on repeated measures) in your analysis whether or not the model for the covariance includes one, two, three or more random effects.

What I wanted to say in my previous answer is that you should have several subjects with more than four longitudinal measurements in your data set to start thinking of using such a complicated random effects covariance matrix as the one determined by an lme model including three random effects.

Yes, subjects with a single measure contribute to more efficient and unbiased estimation of the between-subject variability.

Best -Jorge

---

De: "Lalonde, Francois (NIH/NIMH) [E]" flalonde@mail.nih.gov Para: "freesurfer@nmr.mgh.harvard.edu" freesurfer@nmr.mgh.harvard.edu Enviado: Jueves 28 de marzo de 2013 16:45 Asunto: Re: [Freesurfer] specifying random effects in LME (Linear Mixed Effects models)

Jorge,

Thanks for the clarification.  I will try an analysis using [1 2 3] with all of the subjects with a minimum of 4 repeats and compare the results using the same analysis on all subjects with a minimum of 3 repeats.  This is worthwhile for us since we lose quite a few when excluding those subjects with only 3 repeats.  Your response also brings up the interesting point of what we can expect when including subjects with a single measure (I think a new feature in your longitudinal analysis).  I guess they would contribute to specifying group differences at the level of the intercept?

--Francois

From: jorge luis <jbernal0019@yahoo.esmailto:jbernal0019@yahoo.es> Reply-To: jorge luis <jbernal0019@yahoo.esmailto:jbernal0019@yahoo.es> Date: Wednesday, March 27, 2013 4:58 PM To: Francois Lalonde <flalonde@mail.nih.govmailto:flalonde@mail.nih.gov>, "freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu" <freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu> Subject: Re: [Freesurfer] specifying random effects in LME (Linear Mixed Effects models)

Hi Francois

If you want to test the model with three random effects including intercept, time, and time*time as the random effects then you should use [1 2 3] (these are the columns corresponding to those covariates in X). Actually, for the example in the wiki page we first tested [1 2 3] but the model [1 2] was the best at most vertices. In general, you need more than 4 repeated measures to think of including three random effects in the model for the covariance. Otherwise two random effects are usually enough (you can still include time*time in the model for the mean as in the wiki ). Also, computation time increases quickly with the number of random effects.

There is an oncoming paper that will expand more on our longitudinal mass-univariate analyses with lme (hopefully soon).

Best -Jorge

---

De: "Lalonde, Francois (NIH/NIMH) [E]" <flalonde@mail.nih.govmailto:flalonde@mail.nih.gov> Para: "freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu" <freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu> Enviado: Miércoles 27 de marzo de 2013 15:20 Asunto: [Freesurfer] specifying random effects in LME (Linear Mixed Effects models)

I am following the wiki page for LME analysis and I have a quick question.  The Mass-univariate example near the bottom of the page proposes an initial model that contains intercept, linear and quadratic terms as random effects.  However, the examples just below for lme_mass_fit_EM_init(),  lme_mass_fit_EM_Rgw() only have [1 2] as selected random effects.  Should the vector Zcols contain [1 2 3] as selected random effects in order to test the proposed model?

Thanks, Francois

François Lalonde, Ph.D. Child Psychiatry Branch NIMH / NIH 10 Center Drive, Room 3N202 Bethesda, MD  20892

---

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edumailto: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.

---

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

Yes, it is. You should use the same design matrix X and vary the Zcols selection .

Just a note: If you find the spatiotemporal mixed effects model fitting procedure  described in the wiki too complicated (the paper explaining it is still under revision) you have the option to use the simpler vertex-wise mixed effects model. Something like this:

lhstats = lme_mass_fit_vw(X,[1 2],Y,ni,lhcortex);

This will simply fit a linear mixed effects model independently at each vertex.

Best -Jorge

---

De: "Lalonde, Francois (NIH/NIMH) [E]" flalonde@mail.nih.gov Para: "freesurfer@nmr.mgh.harvard.edu" freesurfer@nmr.mgh.harvard.edu Enviado: Viernes 29 de marzo de 2013 16:03 Asunto: Re: [Freesurfer] specifying random effects in LME (Linear Mixed Effects models)

Hi Jorge,

Thanks for correcting my misunderstanding.  I will include all of the subjects to generate the covariance estimates.  Sorry to be so concrete but in comparing models, for instance, 1 random effect versus 2 random effects, is the same design matrix, X, used for all covariance estimates, the only difference being that the Zcols selection is different?

Thanks for your help and patience.

--Francois

From: jorge luis <jbernal0019@yahoo.esmailto:jbernal0019@yahoo.es> Reply-To: jorge luis <jbernal0019@yahoo.esmailto:jbernal0019@yahoo.es> Date: Thursday, March 28, 2013 5:36 PM To: Francois Lalonde <flalonde@mail.nih.govmailto:flalonde@mail.nih.gov>, "freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu" <freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu> Subject: Re: [Freesurfer] specifying random effects in LME (Linear Mixed Effects models)

Hi Francois

I think that you missunderstood a point of my previous answer.  You should always include ALL subjects (those with 1,2,3,4... and so on repeated measures) in your analysis whether or not the model for the covariance includes one, two, three or more random effects.

What I wanted to say in my previous answer is that you should have several subjects with more than four longitudinal measurements in your data set to start thinking of using such a complicated random effects covariance matrix as the one determined by an lme model including three random effects.

Yes, subjects with a single measure contribute to more efficient and unbiased estimation of the between-subject variability.

Best -Jorge

---

De: "Lalonde, Francois (NIH/NIMH) [E]" <flalonde@mail.nih.govmailto:flalonde@mail.nih.gov> Para: "freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu" <freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu> Enviado: Jueves 28 de marzo de 2013 16:45 Asunto: Re: [Freesurfer] specifying random effects in LME (Linear Mixed Effects models)

Jorge,

Thanks for the clarification.  I will try an analysis using [1 2 3] with all of the subjects with a minimum of 4 repeats and compare the results using the same analysis on all subjects with a minimum of 3 repeats.  This is worthwhile for us since we lose quite a few when excluding those subjects with only 3 repeats.  Your response also brings up the interesting point of what we can expect when including subjects with a single measure (I think a new feature in your longitudinal analysis).  I guess they would contribute to specifying group differences at the level of the intercept?

--Francois

From: jorge luis <jbernal0019@yahoo.esmailto:jbernal0019@yahoo.es<mailto:jbernal0019@yahoo.esmailto:jbernal0019@yahoo.es>> Reply-To: jorge luis <jbernal0019@yahoo.esmailto:jbernal0019@yahoo.es<mailto:jbernal0019@yahoo.esmailto:jbernal0019@yahoo.es>> Date: Wednesday, March 27, 2013 4:58 PM To: Francois Lalonde <flalonde@mail.nih.govmailto:flalonde@mail.nih.gov<mailto:flalonde@mail.nih.govmailto:flalonde@mail.nih.gov>>, "freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu<mailto:freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu>" <freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu<mailto:freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu>> Subject: Re: [Freesurfer] specifying random effects in LME (Linear Mixed Effects models)

Hi Francois

If you want to test the model with three random effects including intercept, time, and time*time as the random effects then you should use [1 2 3] (these are the columns corresponding to those covariates in X). Actually, for the example in the wiki page we first tested [1 2 3] but the model [1 2] was the best at most vertices. In general, you need more than 4 repeated measures to think of including three random effects in the model for the covariance. Otherwise two random effects are usually enough (you can still include time*time in the model for the mean as in the wiki ). Also, computation time increases quickly with the number of random effects.

There is an oncoming paper that will expand more on our longitudinal mass-univariate analyses with lme (hopefully soon).

Best -Jorge

---

De: "Lalonde, Francois (NIH/NIMH) [E]" <flalonde@mail.nih.govmailto:flalonde@mail.nih.gov<mailto:flalonde@mail.nih.govmailto:flalonde@mail.nih.gov>> Para: "freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu<mailto:freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu>" <freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu<mailto:freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu>> Enviado: Miércoles 27 de marzo de 2013 15:20 Asunto: [Freesurfer] specifying random effects in LME (Linear Mixed Effects models)

I am following the wiki page for LME analysis and I have a quick question.  The Mass-univariate example near the bottom of the page proposes an initial model that contains intercept, linear and quadratic terms as random effects.  However, the examples just below for lme_mass_fit_EM_init(),  lme_mass_fit_EM_Rgw() only have [1 2] as selected random effects.  Should the vector Zcols contain [1 2 3] as selected random effects in order to test the proposed model?

Thanks, Francois

François Lalonde, Ph.D. Child Psychiatry Branch NIMH / NIH 10 Center Drive, Room 3N202 Bethesda, MD  20892

---

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edumailto:Freesurfer@nmr.mgh.harvard.edu<mailto:Freesurfer@nmr.mgh.harvard.edumailto: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.

---

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

---

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