Hi Freesurfers,
For obtaining my undergraduate degree I'm performing a study using Freesurfer. I'm stuck with the centering (de-meaning) issue. I have two discrete variables, gender, and patient-control, and a continuous variable which is age. The age population range goes from 20 to 55 years and is paired between patients and controls. Patient's disease affects cortical thickness over time and every patient has been presenting the disease for 6 years (no more, no less). I've runned QDEC and mri_glmfit using DODS obtaining the same results, the problem is that when I demean the age (I used the grandmean) the results are very differente. If I don't demean some clusters survive the monte-carlo multiple comparisons correction (p<0.05), but, when I demean no cluster survive. I have this questions:
1. Since most of my subjects are in the range of 20-35 years and only a few have more than 35 years, it is correct to use the mean? Wouldn't be better to use the median or some other value? 2. This is maybe a silly question, but why should I care about the intercept? I know that demeaning doesn't change the slope, just the intercept, but I don't understand why I should care, I read in some publications that demeaning is not always necessary.
Other questions:
1. Most of my subjects are women (around 80%, N=18), Does it have an advantage to remove the males to get rid of the gender variable?, or should I continue as I've been doing, controlling for the effect of gender. 2. Since I don't know if this disease is more aggressive when is developed at some stage of life, I mean, if young patients are less affected than older patients after six years of presenting it, it is correct to control the effect of age? (Entering the variable as a nuisance factor), or should I search for an interaction between age and the condition variable? I only care about differences in cortical thickness. Any suggestion is very welcome, thanks in advance for your answer.
Rodrigo
Just to clarify, where is said N=18 I wanted to meant 18 controls + 18 patients. Thanks in advance.
Rodrigo
________________________________ De: freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Rodrigo Gonzalez Huerta rghbecker2908@hotmail.com Enviado: martes, 29 de noviembre de 2016 02:19:26 p. m. Para: freesurfer@nmr.mgh.harvard.edu Asunto: [Freesurfer] Demeaning variables
Hi Freesurfers,
For obtaining my undergraduate degree I'm performing a study using Freesurfer. I'm stuck with the centering (de-meaning) issue. I have two discrete variables, gender, and patient-control, and a continuous variable which is age. The age population range goes from 20 to 55 years and is paired between patients and controls. Patient's disease affects cortical thickness over time and every patient has been presenting the disease for 6 years (no more, no less). I've runned QDEC and mri_glmfit using DODS obtaining the same results, the problem is that when I demean the age (I used the grandmean) the results are very differente. If I don't demean some clusters survive the monte-carlo multiple comparisons correction (p<0.05), but, when I demean no cluster survive. I have this questions:
1. Since most of my subjects are in the range of 20-35 years and only a few have more than 35 years, it is correct to use the mean? Wouldn't be better to use the median or some other value? 2. This is maybe a silly question, but why should I care about the intercept? I know that demeaning doesn't change the slope, just the intercept, but I don't understand why I should care, I read in some publications that demeaning is not always necessary.
Other questions:
1. Most of my subjects are women (around 80%, N=18), Does it have an advantage to remove the males to get rid of the gender variable?, or should I continue as I've been doing, controlling for the effect of gender. 2. Since I don't know if this disease is more aggressive when is developed at some stage of life, I mean, if young patients are less affected than older patients after six years of presenting it, it is correct to control the effect of age? (Entering the variable as a nuisance factor), or should I search for an interaction between age and the condition variable? I only care about differences in cortical thickness. Any suggestion is very welcome, thanks in advance for your answer.
Rodrigo
When you say that the results are different, what do you mean? What is your contrast? What is your command line? What is your fsgd file?
On 11/29/2016 03:34 PM, Rodrigo Gonzalez Huerta wrote:
Just to clarify, where is said N=18 I wanted to meant 18 controls + 18 patients. Thanks in advance.
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Rodrigo Gonzalez Huerta rghbecker2908@hotmail.com *Enviado:* martes, 29 de noviembre de 2016 02:19:26 p. m. *Para:* freesurfer@nmr.mgh.harvard.edu *Asunto:* [Freesurfer] Demeaning variables
Hi Freesurfers,
For obtaining my undergraduate degree I’m performing a study using Freesurfer. I’m stuck with the centering (de-meaning) issue. I have two discrete variables, gender, and patient-control, and a continuous variable which is age. The age population range goes from 20 to 55 years and is paired between patients and controls. Patient’s disease affects cortical thickness over time and every patient has been presenting the disease for 6 years (no more, no less). I’ve runned QDEC and mri_glmfit using DODS obtaining the same results, the problem is that when I demean the age (I used the grandmean) the results are very differente. If I don’t demean some clusters survive the monte-carlo multiple comparisons correction (p<0.05), but, when I demean no cluster survive. I have this questions:
- Since most of my subjects are in the range of 20-35 years and only a few have more than 35 years, it is correct to use the mean? Wouldn’t be better to use the median or some other value?
- This is maybe a silly question, but why should I care about the intercept? I know that demeaning doesn’t change the slope, just the intercept, but I don’t understand why I should care, I read in some publications that demeaning is not always necessary.
Other questions:
- Most of my subjects are women (around 80%, N=18), Does it have an advantage to remove the males to get rid of the gender variable?, or should I continue as I’ve been doing, controlling for the effect of gender.
- Since I don’t know if this disease is more aggressive when is developed at some stage of life, I mean, if young patients are less affected than older patients after six years of presenting it, it is correct to control the effect of age? (Entering the variable as a nuisance factor), or should I search for an interaction between age and the condition variable? I only care about differences in cortical thickness.
Any suggestion is very welcome, thanks in advance for your answer.
Rodrigo
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Hi Doug,
I mean that when I run QDEC selecting gender and diagnosis as fixed factors, age as nuisance factor, and display the results, I obtain totally different significant statistical regions (different statistical maps) when ages are demeaned ((demeaned age = age of each subject – the mean of all subjects (grandmean)) than the case when ages aren't demeaned. Therefore, when I run the monte carlo simulation, when I demean the age variable I don’t get any clusters, but when age is not demeaned I obtain some clusters of considerable size. The contrast I used in mri_glmfit is:
0.5 -0.5 0.5 -0.5 0 0 0 0 ("is there a difference between patient and control regressing out the effects of gender and age?")
In QDEC it seems to correspond to the output 'Does the average thickness accounting for gender differ between patient and control? (age as nuisance variable)’ because the statistical maps look the same.
From what I understand, this difference in the results when age variable is demeaned is something I should expect when doing a DODS analysis. My main doubt would be if it was necessary to demean the age variable and enter it as a nuisance factor based on the characteristics of my population and the analysis I want to perform. I tried to focus my questions in that direction. I’m very sorry if I am not giving myself to understand. Thanks!
Rodrigo
________________________________ De: freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas N Greve greve@nmr.mgh.harvard.edu Enviado: martes, 29 de noviembre de 2016 03:22:15 p. m. Para: freesurfer@nmr.mgh.harvard.edu Asunto: Re: [Freesurfer] Demeaning variables
When you say that the results are different, what do you mean? What is your contrast? What is your command line? What is your fsgd file?
On 11/29/2016 03:34 PM, Rodrigo Gonzalez Huerta wrote:
Just to clarify, where is said N=18 I wanted to meant 18 controls + 18 patients. Thanks in advance.
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Rodrigo Gonzalez Huerta rghbecker2908@hotmail.com *Enviado:* martes, 29 de noviembre de 2016 02:19:26 p. m. *Para:* freesurfer@nmr.mgh.harvard.edu *Asunto:* [Freesurfer] Demeaning variables
Hi Freesurfers,
For obtaining my undergraduate degree I’m performing a study using Freesurfer. I’m stuck with the centering (de-meaning) issue. I have two discrete variables, gender, and patient-control, and a continuous variable which is age. The age population range goes from 20 to 55 years and is paired between patients and controls. Patient’s disease affects cortical thickness over time and every patient has been presenting the disease for 6 years (no more, no less). I’ve runned QDEC and mri_glmfit using DODS obtaining the same results, the problem is that when I demean the age (I used the grandmean) the results are very differente. If I don’t demean some clusters survive the monte-carlo multiple comparisons correction (p<0.05), but, when I demean no cluster survive. I have this questions:
- Since most of my subjects are in the range of 20-35 years and only a few have more than 35 years, it is correct to use the mean? Wouldn’t be better to use the median or some other value?
- This is maybe a silly question, but why should I care about the intercept? I know that demeaning doesn’t change the slope, just the intercept, but I don’t understand why I should care, I read in some publications that demeaning is not always necessary.
Other questions:
- Most of my subjects are women (around 80%, N=18), Does it have an advantage to remove the males to get rid of the gender variable?, or should I continue as I’ve been doing, controlling for the effect of gender.
- Since I don’t know if this disease is more aggressive when is developed at some stage of life, I mean, if young patients are less affected than older patients after six years of presenting it, it is correct to control the effect of age? (Entering the variable as a nuisance factor), or should I search for an interaction between age and the condition variable? I only care about differences in cortical thickness.
Any suggestion is very welcome, thanks in advance for your answer.
Rodrigo
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.htmlhttp://www.nmr.mgh.harvard.edu/facility/filedrop/index.html Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/
_______________________________________________ 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.
It comes down to the contrast you care about. If you want to look at the difference in mean thickness between the groups regressing out the effect of age, then demeaning can make a big difference. If this is what you want, then you need to do the analysis in two stages. First run with DODS and see if there are any places where there is an interaction between age and group; demeaning is unimportant for this contrast. If there is not a difference, then reanalyze using DOSS (in which case demeaning is also unimportant). You cannot do DOSS in QDEC.
On 11/29/16 6:23 PM, Rodrigo Gonzalez Huerta wrote:
Hi Doug,
I mean that when I run QDEC selecting gender and diagnosis as fixed factors, age as nuisance factor, and display the results, I obtain totally different significant statistical regions (different statistical maps) when ages are demeaned ((demeaned age = age of each subject – the mean of all subjects (grandmean)) than the case when ages aren't demeaned. Therefore, when I run the monte carlo simulation, when I demean the age variable I don’t get any clusters, but when age is not demeaned I obtain some clusters of considerable size. The contrast I used in mri_glmfit is:
0.5 -0.5 0.5 -0.5 0 0 0 0 ("is there a difference between patient and control regressing out the effects of gender and age?")
In QDEC it seems to correspond to the output 'Does the average thickness accounting for gender differ between patient and control? (age as nuisance variable)’ because the statistical maps look the same.
From what I understand, this difference in the results when age variable is demeaned is something I should expect when doing a DODS analysis. My main doubt would be if it was necessary to demean the age variable and enter it as a nuisance factor based on the characteristics of my population and the analysis I want to perform. I tried to focus my questions in that direction. I’m very sorry if I am not giving myself to understand. Thanks!
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas N Greve greve@nmr.mgh.harvard.edu *Enviado:* martes, 29 de noviembre de 2016 03:22:15 p. m. *Para:* freesurfer@nmr.mgh.harvard.edu *Asunto:* Re: [Freesurfer] Demeaning variables When you say that the results are different, what do you mean? What is your contrast? What is your command line? What is your fsgd file?
On 11/29/2016 03:34 PM, Rodrigo Gonzalez Huerta wrote:
Just to clarify, where is said N=18 I wanted to meant 18 controls + 18 patients. Thanks in advance.
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Rodrigo Gonzalez Huerta rghbecker2908@hotmail.com *Enviado:* martes, 29 de noviembre de 2016 02:19:26 p. m. *Para:* freesurfer@nmr.mgh.harvard.edu *Asunto:* [Freesurfer] Demeaning variables
Hi Freesurfers,
For obtaining my undergraduate degree I’m performing a study using Freesurfer. I’m stuck with the centering (de-meaning) issue. I have two discrete variables, gender, and patient-control, and a continuous variable which is age. The age population range goes from 20 to 55 years and is paired between patients and controls. Patient’s disease affects cortical thickness over time and every patient has been presenting the disease for 6 years (no more, no less). I’ve runned QDEC and mri_glmfit using DODS obtaining the same results, the problem is that when I demean the age (I used the grandmean) the results are very differente. If I don’t demean some clusters survive the monte-carlo multiple comparisons correction (p<0.05), but, when I demean no cluster survive. I have this questions:
- Since most of my subjects are in the range of 20-35 years and only a few have more than 35 years, it is correct to use the mean? Wouldn’t be better to use the median or some other value?
- This is maybe a silly question, but why should I care about the intercept? I know that demeaning doesn’t change the slope, just the intercept, but I don’t understand why I should care, I read in some publications that demeaning is not always necessary.
Other questions:
- Most of my subjects are women (around 80%, N=18), Does it have an advantage to remove the males to get rid of the gender variable?, or should I continue as I’ve been doing, controlling for the effect of gender.
- Since I don’t know if this disease is more aggressive when is developed at some stage of life, I mean, if young patients are less affected than older patients after six years of presenting it, it is correct to control the effect of age? (Entering the variable as a nuisance factor), or should I search for an interaction between age and the condition variable? I only care about differences in cortical thickness.
Any suggestion is very welcome, thanks in advance for your answer.
Rodrigo
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 http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/
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.
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
It makes sense to me, I only have two more questions,
1. When you say “see if there are any places where there is an interaction between age and group”, do you mean finding significant statistical regions, right? And this would be before or after using monte carlo? 2. What about using the median or other value instead of the mean since most of my subjects are in the range from 20 to 35 years old and only a few in the range from 35 to 55 years old. Would this make sense to you? Thanks in advance Rodrigo
________________________________ De: freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas Greve greve@nmr.mgh.harvard.edu Enviado: martes, 29 de noviembre de 2016 07:33:47 p. m. Para: freesurfer@nmr.mgh.harvard.edu Asunto: Re: [Freesurfer] Demeaning variables
It comes down to the contrast you care about. If you want to look at the difference in mean thickness between the groups regressing out the effect of age, then demeaning can make a big difference. If this is what you want, then you need to do the analysis in two stages. First run with DODS and see if there are any places where there is an interaction between age and group; demeaning is unimportant for this contrast. If there is not a difference, then reanalyze using DOSS (in which case demeaning is also unimportant). You cannot do DOSS in QDEC.
On 11/29/16 6:23 PM, Rodrigo Gonzalez Huerta wrote:
Hi Doug,
I mean that when I run QDEC selecting gender and diagnosis as fixed factors, age as nuisance factor, and display the results, I obtain totally different significant statistical regions (different statistical maps) when ages are demeaned ((demeaned age = age of each subject – the mean of all subjects (grandmean)) than the case when ages aren't demeaned. Therefore, when I run the monte carlo simulation, when I demean the age variable I don’t get any clusters, but when age is not demeaned I obtain some clusters of considerable size. The contrast I used in mri_glmfit is:
0.5 -0.5 0.5 -0.5 0 0 0 0 ("is there a difference between patient and control regressing out the effects of gender and age?")
In QDEC it seems to correspond to the output 'Does the average thickness accounting for gender differ between patient and control? (age as nuisance variable)’ because the statistical maps look the same.
From what I understand, this difference in the results when age variable is demeaned is something I should expect when doing a DODS analysis. My main doubt would be if it was necessary to demean the age variable and enter it as a nuisance factor based on the characteristics of my population and the analysis I want to perform. I tried to focus my questions in that direction. I’m very sorry if I am not giving myself to understand. Thanks!
Rodrigo
________________________________ De: freesurfer-bounces@nmr.mgh.harvard.edumailto:freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edumailto:freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas N Greve greve@nmr.mgh.harvard.edumailto:greve@nmr.mgh.harvard.edu Enviado: martes, 29 de noviembre de 2016 03:22:15 p. m. Para: freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu Asunto: Re: [Freesurfer] Demeaning variables
When you say that the results are different, what do you mean? What is your contrast? What is your command line? What is your fsgd file?
On 11/29/2016 03:34 PM, Rodrigo Gonzalez Huerta wrote:
Just to clarify, where is said N=18 I wanted to meant 18 controls + 18 patients. Thanks in advance.
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edumailto:freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edumailto:freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Rodrigo Gonzalez Huerta rghbecker2908@hotmail.commailto:rghbecker2908@hotmail.com *Enviado:* martes, 29 de noviembre de 2016 02:19:26 p. m. *Para:* freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu *Asunto:* [Freesurfer] Demeaning variables
Hi Freesurfers,
For obtaining my undergraduate degree I’m performing a study using Freesurfer. I’m stuck with the centering (de-meaning) issue. I have two discrete variables, gender, and patient-control, and a continuous variable which is age. The age population range goes from 20 to 55 years and is paired between patients and controls. Patient’s disease affects cortical thickness over time and every patient has been presenting the disease for 6 years (no more, no less). I’ve runned QDEC and mri_glmfit using DODS obtaining the same results, the problem is that when I demean the age (I used the grandmean) the results are very differente. If I don’t demean some clusters survive the monte-carlo multiple comparisons correction (p<0.05), but, when I demean no cluster survive. I have this questions:
- Since most of my subjects are in the range of 20-35 years and only a few have more than 35 years, it is correct to use the mean? Wouldn’t be better to use the median or some other value?
- This is maybe a silly question, but why should I care about the intercept? I know that demeaning doesn’t change the slope, just the intercept, but I don’t understand why I should care, I read in some publications that demeaning is not always necessary.
Other questions:
- Most of my subjects are women (around 80%, N=18), Does it have an advantage to remove the males to get rid of the gender variable?, or should I continue as I’ve been doing, controlling for the effect of gender.
- Since I don’t know if this disease is more aggressive when is developed at some stage of life, I mean, if young patients are less affected than older patients after six years of presenting it, it is correct to control the effect of age? (Entering the variable as a nuisance factor), or should I search for an interaction between age and the condition variable? I only care about differences in cortical thickness.
Any suggestion is very welcome, thanks in advance for your answer.
Rodrigo
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edumailto: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.edumailto: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.htmlhttp://www.nmr.mgh.harvard.edu/facility/filedrop/index.html Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/
_______________________________________________ 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.edumailto:Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
On 11/29/16 8:52 PM, Rodrigo Gonzalez Huerta wrote:
It makes sense to me, I only have two more questions,
- When you say “see if there are any places where there is an interaction between age and group”, do you mean finding significant statistical regions, right? And this would be before or after using monte carlo?
After. Even if you have some sig areas, they won't matter if they do not overlap with sig areas in DOSS
What about using the median or other value instead of the mean since most of my subjects are in the range from 20 to 35 years old and only a few in the range from 35 to 55 years old. Would this make sense to you?
It has nothing to do with using the mean or median. The problem is that the differences in the means of the groups is not a valid concept when the age slopes are different
Thanks in advance
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas Greve greve@nmr.mgh.harvard.edu *Enviado:* martes, 29 de noviembre de 2016 07:33:47 p. m. *Para:* freesurfer@nmr.mgh.harvard.edu *Asunto:* Re: [Freesurfer] Demeaning variables
It comes down to the contrast you care about. If you want to look at the difference in mean thickness between the groups regressing out the effect of age, then demeaning can make a big difference. If this is what you want, then you need to do the analysis in two stages. First run with DODS and see if there are any places where there is an interaction between age and group; demeaning is unimportant for this contrast. If there is not a difference, then reanalyze using DOSS (in which case demeaning is also unimportant). You cannot do DOSS in QDEC.
On 11/29/16 6:23 PM, Rodrigo Gonzalez Huerta wrote:
Hi Doug,
I mean that when I run QDEC selecting gender and diagnosis as fixed factors, age as nuisance factor, and display the results, I obtain totally different significant statistical regions (different statistical maps) when ages are demeaned ((demeaned age = age of each subject – the mean of all subjects (grandmean)) than the case when ages aren't demeaned. Therefore, when I run the monte carlo simulation, when I demean the age variable I don’t get any clusters, but when age is not demeaned I obtain some clusters of considerable size. The contrast I used in mri_glmfit is:
0.5 -0.5 0.5 -0.5 0 0 0 0 ("is there a difference between patient and control regressing out the effects of gender and age?")
In QDEC it seems to correspond to the output 'Does the average thickness accounting for gender differ between patient and control? (age as nuisance variable)’ because the statistical maps look the same.
From what I understand, this difference in the results when age variable is demeaned is something I should expect when doing a DODS analysis. My main doubt would be if it was necessary to demean the age variable and enter it as a nuisance factor based on the characteristics of my population and the analysis I want to perform. I tried to focus my questions in that direction. I’m very sorry if I am not giving myself to understand. Thanks!
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas N Greve greve@nmr.mgh.harvard.edu *Enviado:* martes, 29 de noviembre de 2016 03:22:15 p. m. *Para:* freesurfer@nmr.mgh.harvard.edu *Asunto:* Re: [Freesurfer] Demeaning variables When you say that the results are different, what do you mean? What is your contrast? What is your command line? What is your fsgd file?
On 11/29/2016 03:34 PM, Rodrigo Gonzalez Huerta wrote:
Just to clarify, where is said N=18 I wanted to meant 18 controls + 18 patients. Thanks in advance.
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Rodrigo Gonzalez Huerta rghbecker2908@hotmail.com *Enviado:* martes, 29 de noviembre de 2016 02:19:26 p. m. *Para:* freesurfer@nmr.mgh.harvard.edu *Asunto:* [Freesurfer] Demeaning variables
Hi Freesurfers,
For obtaining my undergraduate degree I’m performing a study using Freesurfer. I’m stuck with the centering (de-meaning) issue. I have two discrete variables, gender, and patient-control, and a continuous variable which is age. The age population range goes from 20 to 55 years and is paired between patients and controls. Patient’s disease affects cortical thickness over time and every patient has been presenting the disease for 6 years (no more, no less). I’ve runned QDEC and mri_glmfit using DODS obtaining the same results, the problem is that when I demean the age (I used the grandmean) the results are very differente. If I don’t demean some clusters survive the monte-carlo multiple comparisons correction (p<0.05), but, when I demean no cluster survive. I have this questions:
- Since most of my subjects are in the range of 20-35 years and only a few have more than 35 years, it is correct to use the mean? Wouldn’t be better to use the median or some other value?
- This is maybe a silly question, but why should I care about the intercept? I know that demeaning doesn’t change the slope, just the intercept, but I don’t understand why I should care, I read in some publications that demeaning is not always necessary.
Other questions:
- Most of my subjects are women (around 80%, N=18), Does it have an advantage to remove the males to get rid of the gender variable?, or should I continue as I’ve been doing, controlling for the effect of gender.
- Since I don’t know if this disease is more aggressive when is developed at some stage of life, I mean, if young patients are less affected than older patients after six years of presenting it, it is correct to control the effect of age? (Entering the variable as a nuisance factor), or should I search for an interaction between age and the condition variable? I only care about differences in cortical thickness.
Any suggestion is very welcome, thanks in advance for your answer.
Rodrigo
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 https://gate.nmr.mgh.harvard.edu/filedrop2 www.nmr.mgh.harvard.edu/facility/filedrop/index.html http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/
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.
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
Hey Doug,
Sorry for keep bothering, today I was trying to do what you told me yesterday, but some doubts arose. I ran QDEC using DODS to see if there were any interactions (significant regions) between age (age variable was not demeaned) and group. I didn’t found anything, no significant regions survived monte carlo. If that came to happen, you suggested me to use DOSS instead, but… because of the characteristics of the disease I’m studying, the slopes shouldn’t be similar between patients and controls. My questions are:
1) Can I proceed using DOSS anyway?
2) What if DOSS give me some clusters after running monte carlo? Which is the case since I’ve already ran mri_glmfit and obtained three clusters. What should I do now? Should I re-run DODS with age-variable demeaned?
3) Since most of my subjects are women (around 80%) and my sample is small (18 patients, 18 controls), does it have an advantage to remove the males to be able to get rid of the gender variable that I’m also regressing out? In order to have just two classes (patient and control) instead of four (malecontrol, malepatient,….) , or it is better to continue as I’ve been doing, controlling for the effect of gender.
Thanks for your help!
Rodrigo
________________________________ De: freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas Greve greve@nmr.mgh.harvard.edu Enviado: martes, 29 de noviembre de 2016 08:31:20 p. m. Para: freesurfer@nmr.mgh.harvard.edu Asunto: Re: [Freesurfer] Demeaning variables
On 11/29/16 8:52 PM, Rodrigo Gonzalez Huerta wrote: It makes sense to me, I only have two more questions,
1. When you say “see if there are any places where there is an interaction between age and group”, do you mean finding significant statistical regions, right? And this would be before or after using monte carlo?
After. Even if you have some sig areas, they won't matter if they do not overlap with sig areas in DOSS
1. 2. What about using the median or other value instead of the mean since most of my subjects are in the range from 20 to 35 years old and only a few in the range from 35 to 55 years old. Would this make sense to you?
It has nothing to do with using the mean or median. The problem is that the differences in the means of the groups is not a valid concept when the age slopes are different
1. Thanks in advance Rodrigo
________________________________ De: freesurfer-bounces@nmr.mgh.harvard.edumailto:freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edumailto:freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas Greve greve@nmr.mgh.harvard.edumailto:greve@nmr.mgh.harvard.edu Enviado: martes, 29 de noviembre de 2016 07:33:47 p. m. Para: freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu Asunto: Re: [Freesurfer] Demeaning variables
It comes down to the contrast you care about. If you want to look at the difference in mean thickness between the groups regressing out the effect of age, then demeaning can make a big difference. If this is what you want, then you need to do the analysis in two stages. First run with DODS and see if there are any places where there is an interaction between age and group; demeaning is unimportant for this contrast. If there is not a difference, then reanalyze using DOSS (in which case demeaning is also unimportant). You cannot do DOSS in QDEC.
On 11/29/16 6:23 PM, Rodrigo Gonzalez Huerta wrote:
Hi Doug,
I mean that when I run QDEC selecting gender and diagnosis as fixed factors, age as nuisance factor, and display the results, I obtain totally different significant statistical regions (different statistical maps) when ages are demeaned ((demeaned age = age of each subject – the mean of all subjects (grandmean)) than the case when ages aren't demeaned. Therefore, when I run the monte carlo simulation, when I demean the age variable I don’t get any clusters, but when age is not demeaned I obtain some clusters of considerable size. The contrast I used in mri_glmfit is:
0.5 -0.5 0.5 -0.5 0 0 0 0 ("is there a difference between patient and control regressing out the effects of gender and age?")
In QDEC it seems to correspond to the output 'Does the average thickness accounting for gender differ between patient and control? (age as nuisance variable)’ because the statistical maps look the same.
From what I understand, this difference in the results when age variable is demeaned is something I should expect when doing a DODS analysis. My main doubt would be if it was necessary to demean the age variable and enter it as a nuisance factor based on the characteristics of my population and the analysis I want to perform. I tried to focus my questions in that direction. I’m very sorry if I am not giving myself to understand. Thanks!
Rodrigo
________________________________ De: freesurfer-bounces@nmr.mgh.harvard.edumailto:freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edumailto:freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas N Greve greve@nmr.mgh.harvard.edumailto:greve@nmr.mgh.harvard.edu Enviado: martes, 29 de noviembre de 2016 03:22:15 p. m. Para: freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu Asunto: Re: [Freesurfer] Demeaning variables
When you say that the results are different, what do you mean? What is your contrast? What is your command line? What is your fsgd file?
On 11/29/2016 03:34 PM, Rodrigo Gonzalez Huerta wrote:
Just to clarify, where is said N=18 I wanted to meant 18 controls + 18 patients. Thanks in advance.
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edumailto:freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edumailto:freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Rodrigo Gonzalez Huerta rghbecker2908@hotmail.commailto:rghbecker2908@hotmail.com *Enviado:* martes, 29 de noviembre de 2016 02:19:26 p. m. *Para:* freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu *Asunto:* [Freesurfer] Demeaning variables
Hi Freesurfers,
For obtaining my undergraduate degree I’m performing a study using Freesurfer. I’m stuck with the centering (de-meaning) issue. I have two discrete variables, gender, and patient-control, and a continuous variable which is age. The age population range goes from 20 to 55 years and is paired between patients and controls. Patient’s disease affects cortical thickness over time and every patient has been presenting the disease for 6 years (no more, no less). I’ve runned QDEC and mri_glmfit using DODS obtaining the same results, the problem is that when I demean the age (I used the grandmean) the results are very differente. If I don’t demean some clusters survive the monte-carlo multiple comparisons correction (p<0.05), but, when I demean no cluster survive. I have this questions:
- Since most of my subjects are in the range of 20-35 years and only a few have more than 35 years, it is correct to use the mean? Wouldn’t be better to use the median or some other value?
- This is maybe a silly question, but why should I care about the intercept? I know that demeaning doesn’t change the slope, just the intercept, but I don’t understand why I should care, I read in some publications that demeaning is not always necessary.
Other questions:
- Most of my subjects are women (around 80%, N=18), Does it have an advantage to remove the males to get rid of the gender variable?, or should I continue as I’ve been doing, controlling for the effect of gender.
- Since I don’t know if this disease is more aggressive when is developed at some stage of life, I mean, if young patients are less affected than older patients after six years of presenting it, it is correct to control the effect of age? (Entering the variable as a nuisance factor), or should I search for an interaction between age and the condition variable? I only care about differences in cortical thickness.
Any suggestion is very welcome, thanks in advance for your answer.
Rodrigo
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edumailto: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.edumailto: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.htmlhttp://www.nmr.mgh.harvard.edu/facility/filedrop/index.html Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/
_______________________________________________ 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.edumailto:Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edumailto:Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
If you want to find the effect of group regressing out the effect of age and there is truly a difference in the age slopes, then you have a problem. Imagine a scatter plot of thickness vs age. Each group gets it's own best fit line (offset and slope). If the slopes are the same, then the two lines will always have the same distance between them. However, if the slopes are different, then the difference between the groups depends on age. When you "regress out age", what you are doing is selecting a single age to test the groups. If you don't demean, then you test at age=0. If you demean, you test at age=mean age. The problem is that you get different results depending upon what age you test at. There will be an age where the two lines intersect at which point there will be no difference. Before that age, one group will be greater than the other. After that, it will reverse. Some statisticians say that you should not test for a difference between groups if the slopes are different (ie, your analysis has come to an end). Most people just demean and plow ahead without understanding what it means. If you really think that there is a difference in the slopes despite the negative test result, then you can abandon the analysis or proceed with the understanding of the problems with interpretation. I would probably trust the negative result and proceed with the DOSS test. So, ...
1) Yes
2) Publish
3) Probably not. Adding the males gives you a net increase of 17 DOF, which is probably not trivial
On 11/30/2016 09:45 PM, Rodrigo Gonzalez Huerta wrote:
Hey Doug,
Sorry for keep bothering, today I was trying to do what you told me yesterday, but some doubts arose. I ran QDEC using DODS to see if there were any interactions (significant regions) between age (age variable was not demeaned) and group. I didn’t found anything, no significant regions survived monte carlo. If that came to happen, you suggested me to use DOSS instead, but… because of the characteristics of the disease I’m studying, the slopes shouldn’t be similar between patients and controls. My questions are:
Can I proceed using DOSS anyway?
What if DOSS give me some clusters after running monte carlo? Which
is the case since I’ve already ran mri_glmfit and obtained three clusters. What should I do now? Should I re-run DODS with age-variable demeaned?
- Since most of my subjects are women (around 80%) and my sample is
small (18 patients, 18 controls), does it have an advantage to remove the males to be able to get rid of the gender variable that I’m also regressing out? In order to have just two classes (patient and control) instead of four (malecontrol, malepatient,….) , or it is better to continue as I’ve been doing, controlling for the effect of gender.
Thanks for your help!
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas Greve greve@nmr.mgh.harvard.edu *Enviado:* martes, 29 de noviembre de 2016 08:31:20 p. m. *Para:* freesurfer@nmr.mgh.harvard.edu *Asunto:* Re: [Freesurfer] Demeaning variables
On 11/29/16 8:52 PM, Rodrigo Gonzalez Huerta wrote:
It makes sense to me, I only have two more questions,
- When you say “see if there are any places where there is an interaction between age and group”, do you mean finding significant statistical regions, right? And this would be before or after using monte carlo?
After. Even if you have some sig areas, they won't matter if they do not overlap with sig areas in DOSS
What about using the median or other value instead of the mean since most of my subjects are in the range from 20 to 35 years old and only a few in the range from 35 to 55 years old. Would this make sense to you?
It has nothing to do with using the mean or median. The problem is that the differences in the means of the groups is not a valid concept when the age slopes are different
Thanks in advance
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas Greve greve@nmr.mgh.harvard.edu *Enviado:* martes, 29 de noviembre de 2016 07:33:47 p. m. *Para:* freesurfer@nmr.mgh.harvard.edu *Asunto:* Re: [Freesurfer] Demeaning variables
It comes down to the contrast you care about. If you want to look at the difference in mean thickness between the groups regressing out the effect of age, then demeaning can make a big difference. If this is what you want, then you need to do the analysis in two stages. First run with DODS and see if there are any places where there is an interaction between age and group; demeaning is unimportant for this contrast. If there is not a difference, then reanalyze using DOSS (in which case demeaning is also unimportant). You cannot do DOSS in QDEC.
On 11/29/16 6:23 PM, Rodrigo Gonzalez Huerta wrote:
Hi Doug,
I mean that when I run QDEC selecting gender and diagnosis as fixed factors, age as nuisance factor, and display the results, I obtain totally different significant statistical regions (different statistical maps) when ages are demeaned ((demeaned age = age of each subject – the mean of all subjects (grandmean)) than the case when ages aren't demeaned. Therefore, when I run the monte carlo simulation, when I demean the age variable I don’t get any clusters, but when age is not demeaned I obtain some clusters of considerable size. The contrast I used in mri_glmfit is:
0.5 -0.5 0.5 -0.5 0 0 0 0 ("is there a difference between patient and control regressing out the effects of gender and age?")
In QDEC it seems to correspond to the output 'Does the average thickness accounting for gender differ between patient and control? (age as nuisance variable)’ because the statistical maps look the same.
From what I understand, this difference in the results when age variable is demeaned is something I should expect when doing a DODS analysis. My main doubt would be if it was necessary to demean the age variable and enter it as a nuisance factor based on the characteristics of my population and the analysis I want to perform. I tried to focus my questions in that direction. I’m very sorry if I am not giving myself to understand. Thanks!
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Douglas N Greve greve@nmr.mgh.harvard.edu *Enviado:* martes, 29 de noviembre de 2016 03:22:15 p. m. *Para:* freesurfer@nmr.mgh.harvard.edu *Asunto:* Re: [Freesurfer] Demeaning variables When you say that the results are different, what do you mean? What is your contrast? What is your command line? What is your fsgd file?
On 11/29/2016 03:34 PM, Rodrigo Gonzalez Huerta wrote:
Just to clarify, where is said N=18 I wanted to meant 18 controls
- 18
patients. Thanks in advance.
Rodrigo
*De:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu en nombre de Rodrigo
Gonzalez
Huerta rghbecker2908@hotmail.com *Enviado:* martes, 29 de noviembre de 2016 02:19:26 p. m. *Para:* freesurfer@nmr.mgh.harvard.edu *Asunto:* [Freesurfer] Demeaning variables
Hi Freesurfers,
For obtaining my undergraduate degree I’m performing a study using Freesurfer. I’m stuck with the centering (de-meaning) issue. I have two discrete variables, gender, and patient-control, and a continuous variable which is age. The age population range goes from 20 to 55 years and is paired between patients and controls. Patient’s disease affects cortical thickness over time and every patient has been presenting the disease for 6 years (no more, no less). I’ve runned QDEC and mri_glmfit using DODS obtaining the same results, the
problem
is that when I demean the age (I used the grandmean) the results are very differente. If I don’t demean some clusters survive the monte-carlo multiple comparisons correction (p<0.05), but, when I demean no cluster survive. I have this questions:
- Since most of my subjects are in the range of 20-35 years and only a few have more than 35 years, it is correct to use the mean? Wouldn’t be better to use the median or some other value?
- This is maybe a silly question, but why should I care about the intercept? I know that demeaning doesn’t change the slope, just the intercept, but I don’t understand why I should care, I read in some publications that demeaning is not always necessary.
Other questions:
- Most of my subjects are women (around 80%, N=18), Does it have an advantage to remove the males to get rid of the gender variable?, or should I continue as I’ve been doing, controlling for the effect of gender.
- Since I don’t know if this disease is more aggressive when is developed at some stage of life, I mean, if young patients are less affected than older patients after six years of presenting it, it is correct to control the effect of age? (Entering the variable as a nuisance factor), or should I search for an interaction between age and the condition variable? I only care about differences in cortical thickness.
Any suggestion is very welcome, thanks in advance for your answer.
Rodrigo
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 https://gate.nmr.mgh.harvard.edu/filedrop2 www.nmr.mgh.harvard.edu/facility/filedrop/index.html http://www.nmr.mgh.harvard.edu/facility/filedrop/index.html Outgoing: ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/
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.
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
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Douglas Greve -
Douglas N Greve
-
Rodrigo Gonzalez Huerta