Dear FreeSurfer experts,
Thank you for your fast reply. We are extracting cortical thickness values for statistical sub-analyses in SPSS (due to corrections required with regard to family relatedness within the data); therefore we want to be a 100% certain that the statistical model we use in SPSS is identical to the model used by mri_glmfit. Can you confirm that, out of the multiple linear regression types listed below, mri_glmfit makes use of the Direct (or Standard) regression model?
Type
Characteristics
Direct (or Standard)
. All IVs are entered simultaneously
Hierarchical
. IVs are entered in steps, i.e., some before others
. Interpret: R2 change, F change
Forward
. The software enters IVs one by one until there are no more significant IVs to be entered
Backward
. The software removes IVs one to one until there are no more non-significant IVs to removed
Stepwise
. A combination of Forward and Backward MLR
Thank you again for your help,
Best wishes,
Lizanne
Date: Wed, 12 Mar 2014 11:04:00 -0400
From: Douglas Greve greve@nmr.mgh.harvard.edu
Subject: Re: [Freesurfer] glmfit with correlated covariates
To: freesurfer@nmr.mgh.harvard.edu
Message-ID: 53207760.8010809@nmr.mgh.harvard.edu
Content-Type: text/plain; charset="iso-8859-1"
It is just a standard multiple regression analysis where all regressors
are fit simultaneously. One weight does not include that of another weight.
doug
On 3/12/14 5:16 AM, L. Schweren wrote:
Dear FreeSurfer experts,
I am attempting to disentangle the effects of different features of
pharmacological treatment on cortical thickness.
I am running glmfit from the commandline, with multiple covariates
(a.o. Z_Age, Z_TreatmentDuration and Z_StartAge) in the fsgd-file.
These covariates are correlated up to approximately 0.6 , which to my
understanding is not ideal yet not inducing collinearity. Running
glmfit, I do not get any errors such as ill-designed matrix or so.
My question regards the way the different regression weights are
calculated in each voxel. If I test the variance in CT of voxel A
explained by for example TreatmentDuration, and part of the variance
in voxel A is explained by both TreatmentDuration and StartAge, will
the regression weigth of TreatmentDuration than include the part that
is also explained by StartAge? Or are all other covariates first
"regressed out" of the variance, such that the variable I test can
only explain the variance that was not explained by any of the other
covariates?
Thank you very much, your help is very much appreciated!
Best wishes,
Lizanne
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
I don't want to speak for the Freesurfer team, but I am 99% certain that it is Direct/Standard as you list them. There isn't any stepwise variable selection going on.
________________________________________ From: freesurfer-bounces@nmr.mgh.harvard.edu [freesurfer-bounces@nmr.mgh.harvard.edu] on behalf of L. Schweren [l.j.s.schweren@umcg.nl] Sent: Tuesday, March 25, 2014 6:08 AM To: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] glmfit with correlated covariates
Dear FreeSurfer experts,
Thank you for your fast reply. We are extracting cortical thickness values for statistical sub-analyses in SPSS (due to corrections required with regard to family relatedness within the data); therefore we want to be a 100% certain that the statistical model we use in SPSS is identical to the model used by mri_glmfit. Can you confirm that, out of the multiple linear regression types listed below, mri_glmfit makes use of the Direct (or Standard) regression model?
Type
Characteristics
Direct (or Standard)
· All IVs are entered simultaneously
Hierarchical
· IVs are entered in steps, i.e., some before others · Interpret: R2 change, F change
Forward
· The software enters IVs one by one until there are no more significant IVs to be entered
Backward
· The software removes IVs one to one until there are no more non-significant IVs to removed
Stepwise
· A combination of Forward and Backward MLR
Thank you again for your help,
Best wishes, Lizanne
Date: Wed, 12 Mar 2014 11:04:00 -0400
From: Douglas Greve greve@nmr.mgh.harvard.edu
Subject: Re: [Freesurfer] glmfit with correlated covariates
To: freesurfer@nmr.mgh.harvard.edu
Message-ID: 53207760.8010809@nmr.mgh.harvard.edu
Content-Type: text/plain; charset="iso-8859-1"
It is just a standard multiple regression analysis where all regressors
are fit simultaneously. One weight does not include that of another weight.
doug
On 3/12/14 5:16 AM, L. Schweren wrote:
Dear FreeSurfer experts,
I am attempting to disentangle the effects of different features of
pharmacological treatment on cortical thickness.
I am running glmfit from the commandline, with multiple covariates
(a.o. Z_Age, Z_TreatmentDuration and Z_StartAge) in the fsgd-file.
These covariates are correlated up to approximately 0.6 , which to my
understanding is not ideal yet not inducing collinearity. Running
glmfit, I do not get any errors such as ill-designed matrix or so.
My question regards the way the different regression weights are
calculated in each voxel. If I test the variance in CT of voxel A
explained by for example TreatmentDuration, and part of the variance
in voxel A is explained by both TreatmentDuration and StartAge, will
the regression weigth of TreatmentDuration than include the part that
is also explained by StartAge? Or are all other covariates first
"regressed out" of the variance, such that the variable I test can
only explain the variance that was not explained by any of the other
covariates?
Thank you very much, your help is very much appreciated!
Best wishes,
Lizanne
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Yes, they are all entered simultaneously doug
On 03/25/2014 11:09 AM, Watson, Christopher wrote:
I don't want to speak for the Freesurfer team, but I am 99% certain that it is Direct/Standard as you list them. There isn't any stepwise variable selection going on.
From: freesurfer-bounces@nmr.mgh.harvard.edu [freesurfer-bounces@nmr.mgh.harvard.edu] on behalf of L. Schweren [l.j.s.schweren@umcg.nl] Sent: Tuesday, March 25, 2014 6:08 AM To: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] glmfit with correlated covariates
Dear FreeSurfer experts,
Thank you for your fast reply. We are extracting cortical thickness values for statistical sub-analyses in SPSS (due to corrections required with regard to family relatedness within the data); therefore we want to be a 100% certain that the statistical model we use in SPSS is identical to the model used by mri_glmfit. Can you confirm that, out of the multiple linear regression types listed below, mri_glmfit makes use of the Direct (or Standard) regression model?
Type
Characteristics
Direct (or Standard)
· All IVs are entered simultaneously
Hierarchical
· IVs are entered in steps, i.e., some before others · Interpret: R2 change, F change
Forward
· The software enters IVs one by one until there are no more significant IVs to be entered
Backward
· The software removes IVs one to one until there are no more non-significant IVs to removed
Stepwise
· A combination of Forward and Backward MLR
Thank you again for your help,
Best wishes, Lizanne
Date: Wed, 12 Mar 2014 11:04:00 -0400
From: Douglas Greve greve@nmr.mgh.harvard.edu
Subject: Re: [Freesurfer] glmfit with correlated covariates
To: freesurfer@nmr.mgh.harvard.edu
Message-ID: 53207760.8010809@nmr.mgh.harvard.edu
Content-Type: text/plain; charset="iso-8859-1"
It is just a standard multiple regression analysis where all regressors
are fit simultaneously. One weight does not include that of another weight.
doug
On 3/12/14 5:16 AM, L. Schweren wrote:
Dear FreeSurfer experts, I am attempting to disentangle the effects of different features of pharmacological treatment on cortical thickness. I am running glmfit from the commandline, with multiple covariates (a.o. Z_Age, Z_TreatmentDuration and Z_StartAge) in the fsgd-file. These covariates are correlated up to approximately 0.6 , which to my understanding is not ideal yet not inducing collinearity. Running glmfit, I do not get any errors such as ill-designed matrix or so. My question regards the way the different regression weights are calculated in each voxel. If I test the variance in CT of voxel A explained by for example TreatmentDuration, and part of the variance in voxel A is explained by both TreatmentDuration and StartAge, will the regression weigth of TreatmentDuration than include the part that is also explained by StartAge? Or are all other covariates first "regressed out" of the variance, such that the variable I test can only explain the variance that was not explained by any of the other covariates? Thank you very much, your help is very much appreciated! Best wishes, Lizanne _______________________________________________ Freesurfer mailing list 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
Hi Lizanne, I do not use SPSS so I cannot comment on which model is equivalent. BTW, if you can save the design an contrast matrices in text form from SPSS, then you can feed them into mri_glmfit. I'm cc'ing Steve Hodge who has been working on mapping R design matrices to FS on the chance that he might have some insight. doug
On 03/25/2014 06:04 AM, L. Schweren wrote:
//
/Dear FreeSurfer experts, /
//
/Thank you for your fast reply. We are extracting cortical thickness values for statistical sub-analyses in SPSS (due to corrections required with regard to family relatedness within the data); therefore we want to be a 100% certain that the statistical model we use in SPSS is identical to the model used by mri_glmfit. Can you confirm that, out of the multiple linear regression types listed below, mri_glmfit makes use of the Direct (or Standard) regression model?/
//
*Type*
*Characteristics*
Direct (or Standard)
·All IVs are entered simultaneously
Hierarchical
·IVs are entered in steps, i.e., some before others
·Interpret: /R/^2 change, /F/ change
Forward
·The software enters IVs one by one until there are no more significant IVs to be entered
Backward
·The software removes IVs one to one until there are no more non-significant IVs to removed
Stepwise
·A combination of Forward and Backward MLR
//
/Thank you again for your help, /
//
/Best wishes, /
/Lizanne/
Date: Wed, 12 Mar 2014 11:04:00 -0400
From: Douglas Greve greve@nmr.mgh.harvard.edu
Subject: Re: [Freesurfer] glmfit with correlated covariates
To: freesurfer@nmr.mgh.harvard.edu
Message-ID: 53207760.8010809@nmr.mgh.harvard.edu
Content-Type: text/plain; charset="iso-8859-1"
It is just a standard multiple regression analysis where all regressors
are fit simultaneously. One weight does not include that of another weight.
doug
On 3/12/14 5:16 AM, L. Schweren wrote:
Dear FreeSurfer experts,
I am attempting to disentangle the effects of different features of
pharmacological treatment on cortical thickness.
I am running glmfit from the commandline, with multiple covariates
(a.o. Z_Age, Z_TreatmentDuration and Z_StartAge) in the fsgd-file.
These covariates are correlated up to approximately 0.6 , which to my
understanding is not ideal yet not inducing collinearity. Running
glmfit, I do not get any errors such as ill-designed matrix or so.
My question regards the way the different regression weights are
calculated in each voxel. If I test the variance in CT of voxel A
explained by for example TreatmentDuration, and part of the variance
in voxel A is explained by both TreatmentDuration and StartAge, will
the regression weigth of TreatmentDuration than include the part that
is also explained by StartAge? Or are all other covariates first
"regressed out" of the variance, such that the variable I test can
only explain the variance that was not explained by any of the other
covariates?
Thank you very much, your help is very much appreciated!
Best wishes,
Lizanne
Freesurfer mailing list
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
Hi Lizanne,
My method for determining the equivalent model in R is to extract single-vertex data from y.mgh (used in the mri_glmfit model), import it into R, and apply what I think is the same model. The expectation is that the results from R should match that particular vertex's values in beta.mgh, sig.mgh, F.mgh, etc.
So my suggestion is to use mri_segstats to get data from one vertex and play with it in SPSS to find the equivalent model. I'm guessing 'Direct (or Standard)', but each stats package will have default methods for scaling variables and coding factors that are only apparent when you compare results between packages (mri_glmfit, SPSS, R, etc.)
My virtual machine is down, otherwise I could test some this in SPSS ...
Steve . . . . . . . . . . . . . . . Steven M. Hodge Child and Adolescent Neurodevelopment Initiative S3-301 University of Massachusetts Medical School 55 Lake Avenue North Worcester, MA 01655
Steven.Hodge@umassmed.edu 774-455-4108 ________________________________________ From: Douglas N Greve [greve@nmr.mgh.harvard.edu] Sent: Tuesday, March 25, 2014 11:43 To: freesurfer@nmr.mgh.harvard.edu; Hodge, Steven Subject: Re: [Freesurfer] glmfit with correlated covariates
Hi Lizanne, I do not use SPSS so I cannot comment on which model is equivalent. BTW, if you can save the design an contrast matrices in text form from SPSS, then you can feed them into mri_glmfit. I'm cc'ing Steve Hodge who has been working on mapping R design matrices to FS on the chance that he might have some insight. doug
On 03/25/2014 06:04 AM, L. Schweren wrote:
//
/Dear FreeSurfer experts, /
//
/Thank you for your fast reply. We are extracting cortical thickness values for statistical sub-analyses in SPSS (due to corrections required with regard to family relatedness within the data); therefore we want to be a 100% certain that the statistical model we use in SPSS is identical to the model used by mri_glmfit. Can you confirm that, out of the multiple linear regression types listed below, mri_glmfit makes use of the Direct (or Standard) regression model?/
//
*Type*
*Characteristics*
Direct (or Standard)
·All IVs are entered simultaneously
Hierarchical
·IVs are entered in steps, i.e., some before others
·Interpret: /R/^2 change, /F/ change
Forward
·The software enters IVs one by one until there are no more significant IVs to be entered
Backward
·The software removes IVs one to one until there are no more non-significant IVs to removed
Stepwise
·A combination of Forward and Backward MLR
//
/Thank you again for your help, /
//
/Best wishes, /
/Lizanne/
Date: Wed, 12 Mar 2014 11:04:00 -0400
From: Douglas Greve greve@nmr.mgh.harvard.edu
Subject: Re: [Freesurfer] glmfit with correlated covariates
To: freesurfer@nmr.mgh.harvard.edu
Message-ID: 53207760.8010809@nmr.mgh.harvard.edu
Content-Type: text/plain; charset="iso-8859-1"
It is just a standard multiple regression analysis where all regressors
are fit simultaneously. One weight does not include that of another weight.
doug
On 3/12/14 5:16 AM, L. Schweren wrote:
Dear FreeSurfer experts,
I am attempting to disentangle the effects of different features of
pharmacological treatment on cortical thickness.
I am running glmfit from the commandline, with multiple covariates
(a.o. Z_Age, Z_TreatmentDuration and Z_StartAge) in the fsgd-file.
These covariates are correlated up to approximately 0.6 , which to my
understanding is not ideal yet not inducing collinearity. Running
glmfit, I do not get any errors such as ill-designed matrix or so.
My question regards the way the different regression weights are
calculated in each voxel. If I test the variance in CT of voxel A
explained by for example TreatmentDuration, and part of the variance
in voxel A is explained by both TreatmentDuration and StartAge, will
the regression weigth of TreatmentDuration than include the part that
is also explained by StartAge? Or are all other covariates first
"regressed out" of the variance, such that the variable I test can
only explain the variance that was not explained by any of the other
covariates?
Thank you very much, your help is very much appreciated!
Best wishes,
Lizanne
Freesurfer mailing list
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
-- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu Phone Number: 617-724-2358 Fax: 617-726-7422
Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting FileDrop: https://gate.nmr.mgh.harvard.edu/filedrop2 www.nmr.mgh.harvard.edu/facility/filedrop/index.html Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/
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.
Thanks Steve. BTW, you can use the --voxdump col row slice functionality with mri_glmfit to extract all that info from a single voxel (maybe a little easier than mri_segstats)
doug
On 03/25/2014 06:10 PM, Hodge, Steven wrote:
Hi Lizanne,
My method for determining the equivalent model in R is to extract single-vertex data from y.mgh (used in the mri_glmfit model), import it into R, and apply what I think is the same model. The expectation is that the results from R should match that particular vertex's values in beta.mgh, sig.mgh, F.mgh, etc.
So my suggestion is to use mri_segstats to get data from one vertex and play with it in SPSS to find the equivalent model. I'm guessing 'Direct (or Standard)', but each stats package will have default methods for scaling variables and coding factors that are only apparent when you compare results between packages (mri_glmfit, SPSS, R, etc.)
My virtual machine is down, otherwise I could test some this in SPSS ...
Steve . . . . . . . . . . . . . . . Steven M. Hodge Child and Adolescent Neurodevelopment Initiative S3-301 University of Massachusetts Medical School 55 Lake Avenue North Worcester, MA 01655
Steven.Hodge@umassmed.edu 774-455-4108 ________________________________________ From: Douglas N Greve [greve@nmr.mgh.harvard.edu] Sent: Tuesday, March 25, 2014 11:43 To: freesurfer@nmr.mgh.harvard.edu; Hodge, Steven Subject: Re: [Freesurfer] glmfit with correlated covariates
Hi Lizanne, I do not use SPSS so I cannot comment on which model is equivalent. BTW, if you can save the design an contrast matrices in text form from SPSS, then you can feed them into mri_glmfit. I'm cc'ing Steve Hodge who has been working on mapping R design matrices to FS on the chance that he might have some insight. doug
On 03/25/2014 06:04 AM, L. Schweren wrote:
//
/Dear FreeSurfer experts, /
//
/Thank you for your fast reply. We are extracting cortical thickness values for statistical sub-analyses in SPSS (due to corrections required with regard to family relatedness within the data); therefore we want to be a 100% certain that the statistical model we use in SPSS is identical to the model used by mri_glmfit. Can you confirm that, out of the multiple linear regression types listed below, mri_glmfit makes use of the Direct (or Standard) regression model?/
//
*Type*
*Characteristics*
Direct (or Standard)
·All IVs are entered simultaneously
Hierarchical
·IVs are entered in steps, i.e., some before others
·Interpret: /R/^2 change, /F/ change
Forward
·The software enters IVs one by one until there are no more significant IVs to be entered
Backward
·The software removes IVs one to one until there are no more non-significant IVs to removed
Stepwise
·A combination of Forward and Backward MLR
//
/Thank you again for your help, /
//
/Best wishes, /
/Lizanne/
Date: Wed, 12 Mar 2014 11:04:00 -0400
From: Douglas Greve greve@nmr.mgh.harvard.edu
Subject: Re: [Freesurfer] glmfit with correlated covariates
To: freesurfer@nmr.mgh.harvard.edu
Message-ID: 53207760.8010809@nmr.mgh.harvard.edu
Content-Type: text/plain; charset="iso-8859-1"
It is just a standard multiple regression analysis where all regressors
are fit simultaneously. One weight does not include that of another weight.
doug
On 3/12/14 5:16 AM, L. Schweren wrote:
Dear FreeSurfer experts, I am attempting to disentangle the effects of different features of pharmacological treatment on cortical thickness. I am running glmfit from the commandline, with multiple covariates (a.o. Z_Age, Z_TreatmentDuration and Z_StartAge) in the fsgd-file. These covariates are correlated up to approximately 0.6 , which to my understanding is not ideal yet not inducing collinearity. Running glmfit, I do not get any errors such as ill-designed matrix or so. My question regards the way the different regression weights are calculated in each voxel. If I test the variance in CT of voxel A explained by for example TreatmentDuration, and part of the variance in voxel A is explained by both TreatmentDuration and StartAge, will the regression weigth of TreatmentDuration than include the part that is also explained by StartAge? Or are all other covariates first "regressed out" of the variance, such that the variable I test can only explain the variance that was not explained by any of the other covariates? Thank you very much, your help is very much appreciated! Best wishes, Lizanne _______________________________________________ Freesurfer mailing list 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
-- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu Phone Number: 617-724-2358 Fax: 617-726-7422
Bugs: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting FileDrop: https://gate.nmr.mgh.harvard.edu/filedrop2 www.nmr.mgh.harvard.edu/facility/filedrop/index.html Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/
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
freesurfer@nmr.mgh.harvard.edu
-
Douglas N Greve -
Hodge, Steven -
L. Schweren -
Watson, Christopher