I have a clarifying question about fsgd file format. We have 40 participants, and a large battery of individual difference measures (egg., vocabulary, phonological skill, etc.) We applied a PCA to the battery and found 4 components, corresponding to WMC, processing speed, comprehension ability, and phonological ability. We tested for main effects of each of these on thickness, area, and volume via QDEC. However, we also want to check for interactions - and since this is essentially 4 continuous factors, this is not QDEC-able (at least, it does not appear to be). A nuisance variable (e.g., mean thickness) is also present. Nregressors = Nclasses*(Nvariables+1) = 1*((4 components + 1 nuisance) +1) = 1*(5+1) = 6
Null contrast is 1 0 0 0 0 0
Main effect, e.g., of the working memory component (#1) is
0 1 0 0 0 0
Main effect, e.g., of the WM component modulated by the nuisance variable is
0 1 0 0 0 1
So the format to examine possible interactions, e.g., component 1 and component 2 interact to affect thickness, is
0 1 1 0 0 0
And with the nuisance variable
0 1 1 0 0 1
Do we have this right?
THANKS!!
Clint
If you want to test for interactions among nuisance variables, then you have to create a new regressor where you multiply them together. doug
On 3/30/14 4:20 PM, Clint Johns wrote:
I have a clarifying question about fsgd file format. We have 40 participants, and a large battery of individual difference measures (egg., vocabulary, phonological skill, etc.) We applied a PCA to the battery and found 4 components, corresponding to WMC, processing speed, comprehension ability, and phonological ability. We tested for main effects of each of these on thickness, area, and volume via QDEC. However, we also want to check for interactions - and since this is essentially 4 continuous factors, this is not QDEC-able (at least, it does not appear to be). A nuisance variable (e.g., mean thickness) is also present. Nregressors = Nclasses*(Nvariables+1) = 1*((4 components + 1 nuisance) +1) = 1*(5+1) = 6
Null contrast is 1 0 0 0 0 0
Main effect, e.g., of the working memory component (#1) is
0 1 0 0 0 0
Main effect, e.g., of the WM component modulated by the nuisance variable is
0 1 0 0 0 1
So the format to examine possible interactions, e.g., component 1 and component 2 interact to affect thickness, is
0 1 1 0 0 0
And with the nuisance variable
0 1 1 0 0 1
Do we have this right?
THANKS!!
Clint
-- I remain...
Clinton L. Johns, Ph.D. Research Scientist, Haskins Laboratories 300 George Street New Haven CT 06511 speech: 203-865-6163 x240 fax: 203-865-8963 net: johns@haskins.yale.edu mailto:johns@haskins.yale.edu
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
I'm not sure I understand (sorry for being obtuse!) We don't want to check for interactions among nuisance variables (inasmuch as we only have the one, the average thickness). Rather, we want to check for interactions among the four factors revealed by our PCA, controlling for the nuisance variable. Without the nuisance variable, we have 5 regressors, and the nuisance variable makes six.
So, if I *do* understand you, we did right by including the mean thickness as a new regressor (the final '1' in the various .fsgd files we created, where the various other combinations of 1s reflect the other potential interactions of our 4 variables). So the nuisance variable effectively doubles the number of contrasts, so we have .fsgd files for possible interactions absent the nuisance variable, and a second set that includes the nuisance variable.
Again, I'm sorry if I'm missing the obvious. I'm really just not sure that we have formatted them.fsgd files correctly!
Thanks for your time and attention.
clint
On Sunday, March 30, 2014, Douglas Greve greve@nmr.mgh.harvard.edu wrote:
If you want to test for interactions among nuisance variables, then you have to create a new regressor where you multiply them together. doug
On 3/30/14 4:20 PM, Clint Johns wrote:
I have a clarifying question about fsgd file format. We have 40 participants, and a large battery of individual difference measures (egg., vocabulary, phonological skill, etc.) We applied a PCA to the battery and found 4 components, corresponding to WMC, processing speed, comprehension ability, and phonological ability. We tested for main effects of each of these on thickness, area, and volume via QDEC. However, we also want to check for interactions - and since this is essentially 4 continuous factors, this is not QDEC-able (at least, it does not appear to be). A nuisance variable (e.g., mean thickness) is also present. Nregressors = Nclasses*(Nvariables+1) = 1*((4 components + 1 nuisance) +1) = 1*(5+1) = 6
Null contrast is 1 0 0 0 0 0
Main effect, e.g., of the working memory component (#1) is
0 1 0 0 0 0
Main effect, e.g., of the WM component modulated by the nuisance variable is
0 1 0 0 0 1
So the format to examine possible interactions, e.g., component 1 and component 2 interact to affect thickness, is
0 1 1 0 0 0
And with the nuisance variable
0 1 1 0 0 1
Do we have this right?
THANKS!!
Clint
-- I remain...
Clinton L. Johns, Ph.D. Research Scientist, Haskins Laboratories 300 George Street New Haven CT 06511 speech: 203-865-6163 x240 fax: 203-865-8963 net: johns@haskins.yale.edujavascript:_e(%7B%7D,'cvml','johns@haskins.yale.edu');
Freesurfer mailing listFreesurfer@nmr.mgh.harvard.edu javascript:_e(%7B%7D,'cvml','Freesurfer@nmr.mgh.harvard.edu');https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Sorry, that should have been interaction among continuous (not necessarily nuisance) variables. You will need to create new variables that are products of your variables of interest in order to test for interactions.
doug
On 3/30/14 9:20 PM, Clint Johns wrote:
I'm not sure I understand (sorry for being obtuse!) We don't want to check for interactions among nuisance variables (inasmuch as we only have the one, the average thickness). Rather, we want to check for interactions among the four factors revealed by our PCA, controlling for the nuisance variable. Without the nuisance variable, we have 5 regressors, and the nuisance variable makes six.
So, if I *do* understand you, we did right by including the mean thickness as a new regressor (the final '1' in the various .fsgd files we created, where the various other combinations of 1s reflect the other potential interactions of our 4 variables). So the nuisance variable effectively doubles the number of contrasts, so we have .fsgd files for possible interactions absent the nuisance variable, and a second set that includes the nuisance variable.
Again, I'm sorry if I'm missing the obvious. I'm really just not sure that we have formatted them.fsgd files correctly!
Thanks for your time and attention.
clint
On Sunday, March 30, 2014, Douglas Greve <greve@nmr.mgh.harvard.edu mailto:greve@nmr.mgh.harvard.edu> wrote:
If you want to test for interactions among nuisance variables, then you have to create a new regressor where you multiply them together. doug On 3/30/14 4:20 PM, Clint Johns wrote:I have a clarifying question about fsgd file format. We have 40 participants, and a large battery of individual difference measures (egg., vocabulary, phonological skill, etc.) We applied a PCA to the battery and found 4 components, corresponding to WMC, processing speed, comprehension ability, and phonological ability. We tested for main effects of each of these on thickness, area, and volume via QDEC. However, we also want to check for interactions - and since this is essentially 4 continuous factors, this is not QDEC-able (at least, it does not appear to be). A nuisance variable (e.g., mean thickness) is also present. Nregressors = Nclasses*(Nvariables+1) = 1*((4 components + 1 nuisance) +1) = 1*(5+1) = 6 Null contrast is 1 0 0 0 0 0 Main effect, e.g., of the working memory component (#1) is 0 1 0 0 0 0 Main effect, e.g., of the WM component modulated by the nuisance variable is 0 1 0 0 0 1 So the format to examine possible interactions, e.g., component 1 and component 2 interact to affect thickness, is 0 1 1 0 0 0 And with the nuisance variable 0 1 1 0 0 1 Do we have this right? THANKS!! Clint -- I remain... Clinton L. Johns, Ph.D. Research Scientist, Haskins Laboratories 300 George Street New Haven CT 06511 speech: 203-865-6163 x240 fax: 203-865-8963 net: johns@haskins.yale.edu <javascript:_e(%7B%7D,'cvml','johns@haskins.yale.edu');> _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu <javascript:_e(%7B%7D,'cvml','Freesurfer@nmr.mgh.harvard.edu');> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer-- I remain...
Clinton L. Johns, Ph.D. Research Scientist, Haskins Laboratories 300 George Street New Haven CT 06511 speech: 203-865-6163 x240 fax: 203-865-8963 net: johns@haskins.yale.edu mailto:johns@haskins.yale.edu
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
I understand, I think! So our matrix should be something like this, which includes each of the 4 components we are using as variables, all possible interactions, and the covariate of mean cortical thickness?
group - Var1 - Var2 - Var3 - Var4 - 1x2 - 1x3 - 1x4 - 2x3 - 2x4 - 3x4 - 1x2x3 - 1x2x4 - 1x3x4 - 2x3x4 - 1x2x3x4 - avgthickness)
And testing for interactions takes the form of
E.g., the interaction of components 1 and 2, our matrix should be:
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
E.g., the interaction of components 1 and 2 and 3, our matrix should be:
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
And so on. Am I right? And if so, is it possible to include all the interaction tests in a single .fsgd file, like
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 etc.
And I have one final question (assuming that I got the first bit). In the .fsgd file that was set up this way:
group - Var1 - Var2 - Var3 - Var4 - avgthickness
What does a matrix like this:
0 1 1 0 0 0
actually do? Shouldn't this create some kind of error message?
Thanks again for all your advice on this!
clint
On Sun, Mar 30, 2014 at 10:32 PM, Douglas Greve greve@nmr.mgh.harvard.edu wrote:
Sorry, that should have been interaction among continuous (not necessarily nuisance) variables. You will need to create new variables that are products of your variables of interest in order to test for interactions.
doug
On 3/30/14 9:20 PM, Clint Johns wrote:
I'm not sure I understand (sorry for being obtuse!) We don't want to check for interactions among nuisance variables (inasmuch as we only have the one, the average thickness). Rather, we want to check for interactions among the four factors revealed by our PCA, controlling for the nuisance variable. Without the nuisance variable, we have 5 regressors, and the nuisance variable makes six.
So, if I *do* understand you, we did right by including the mean thickness as a new regressor (the final '1' in the various .fsgd files we created, where the various other combinations of 1s reflect the other potential interactions of our 4 variables). So the nuisance variable effectively doubles the number of contrasts, so we have .fsgd files for possible interactions absent the nuisance variable, and a second set that includes the nuisance variable.
Again, I'm sorry if I'm missing the obvious. I'm really just not sure that we have formatted them.fsgd files correctly!
Thanks for your time and attention.
clint
On Sunday, March 30, 2014, Douglas Greve greve@nmr.mgh.harvard.edu wrote:
If you want to test for interactions among nuisance variables, then you have to create a new regressor where you multiply them together. doug
On 3/30/14 4:20 PM, Clint Johns wrote:
I have a clarifying question about fsgd file format. We have 40 participants, and a large battery of individual difference measures (egg., vocabulary, phonological skill, etc.) We applied a PCA to the battery and found 4 components, corresponding to WMC, processing speed, comprehension ability, and phonological ability. We tested for main effects of each of these on thickness, area, and volume via QDEC. However, we also want to check for interactions - and since this is essentially 4 continuous factors, this is not QDEC-able (at least, it does not appear to be). A nuisance variable (e.g., mean thickness) is also present. Nregressors = Nclasses*(Nvariables+1) = 1*((4 components + 1 nuisance) +1) = 1*(5+1) = 6
Null contrast is 1 0 0 0 0 0
Main effect, e.g., of the working memory component (#1) is
0 1 0 0 0 0
Main effect, e.g., of the WM component modulated by the nuisance variable is
0 1 0 0 0 1
So the format to examine possible interactions, e.g., component 1 and component 2 interact to affect thickness, is
0 1 1 0 0 0
And with the nuisance variable
0 1 1 0 0 1
Do we have this right?
THANKS!!
Clint
-- I remain...
Clinton L. Johns, Ph.D. Research Scientist, Haskins Laboratories 300 George Street New Haven CT 06511 speech: 203-865-6163 x240 fax: 203-865-8963 net: johns@haskins.yale.edu
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
-- I remain...
Clinton L. Johns, Ph.D. Research Scientist, Haskins Laboratories 300 George Street New Haven CT 06511 speech: 203-865-6163 x240 fax: 203-865-8963 net: johns@haskins.yale.edu
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
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.
That all looks right. Your final contrast matrix that you have a question about tests whether Var1+Var2=0 doug
On 03/31/2014 08:33 AM, Clint Johns wrote:
I understand, I think! So our matrix should be something like this, which includes each of the 4 components we are using as variables, all possible interactions, and the covariate of mean cortical thickness?
group - Var1 - Var2 - Var3 - Var4 - 1x2 - 1x3 - 1x4 - 2x3 - 2x4 - 3x4
- 1x2x3 - 1x2x4 - 1x3x4 - 2x3x4 - 1x2x3x4 - avgthickness)
And testing for interactions takes the form of
E.g., the interaction of components 1 and 2, our matrix should be:
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
E.g., the interaction of components 1 and 2 and 3, our matrix should be:
0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
And so on. Am I right? And if so, is it possible to include all the interaction tests in a single .fsgd file, like
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 etc.
And I have one final question (assuming that I got the first bit). In the .fsgd file that was set up this way:
group - Var1 - Var2 - Var3 - Var4 - avgthickness
What does a matrix like this:
0 1 1 0 0 0
actually do? Shouldn't this create some kind of error message?
Thanks again for all your advice on this!
clint
On Sun, Mar 30, 2014 at 10:32 PM, Douglas Greve greve@nmr.mgh.harvard.edu wrote:
Sorry, that should have been interaction among continuous (not necessarily nuisance) variables. You will need to create new variables that are products of your variables of interest in order to test for interactions.
doug
On 3/30/14 9:20 PM, Clint Johns wrote:
I'm not sure I understand (sorry for being obtuse!) We don't want to check for interactions among nuisance variables (inasmuch as we only have the one, the average thickness). Rather, we want to check for interactions among the four factors revealed by our PCA, controlling for the nuisance variable. Without the nuisance variable, we have 5 regressors, and the nuisance variable makes six.
So, if I *do* understand you, we did right by including the mean thickness as a new regressor (the final '1' in the various .fsgd files we created, where the various other combinations of 1s reflect the other potential interactions of our 4 variables). So the nuisance variable effectively doubles the number of contrasts, so we have .fsgd files for possible interactions absent the nuisance variable, and a second set that includes the nuisance variable.
Again, I'm sorry if I'm missing the obvious. I'm really just not sure that we have formatted them.fsgd files correctly!
Thanks for your time and attention.
clint
On Sunday, March 30, 2014, Douglas Greve greve@nmr.mgh.harvard.edu wrote:
If you want to test for interactions among nuisance variables, then you have to create a new regressor where you multiply them together. doug
On 3/30/14 4:20 PM, Clint Johns wrote:
I have a clarifying question about fsgd file format. We have 40 participants, and a large battery of individual difference measures (egg., vocabulary, phonological skill, etc.) We applied a PCA to the battery and found 4 components, corresponding to WMC, processing speed, comprehension ability, and phonological ability. We tested for main effects of each of these on thickness, area, and volume via QDEC. However, we also want to check for interactions - and since this is essentially 4 continuous factors, this is not QDEC-able (at least, it does not appear to be). A nuisance variable (e.g., mean thickness) is also present. Nregressors = Nclasses*(Nvariables+1) = 1*((4 components + 1 nuisance) +1) = 1*(5+1) = 6
Null contrast is 1 0 0 0 0 0
Main effect, e.g., of the working memory component (#1) is
0 1 0 0 0 0
Main effect, e.g., of the WM component modulated by the nuisance variable is
0 1 0 0 0 1
So the format to examine possible interactions, e.g., component 1 and component 2 interact to affect thickness, is
0 1 1 0 0 0
And with the nuisance variable
0 1 1 0 0 1
Do we have this right?
THANKS!!
Clint
-- I remain...
Clinton L. Johns, Ph.D. Research Scientist, Haskins Laboratories 300 George Street New Haven CT 06511 speech: 203-865-6163 x240 fax: 203-865-8963 net: johns@haskins.yale.edu
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
-- I remain...
Clinton L. Johns, Ph.D. Research Scientist, Haskins Laboratories 300 George Street New Haven CT 06511 speech: 203-865-6163 x240 fax: 203-865-8963 net: johns@haskins.yale.edu
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
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@nmr.mgh.harvard.edu
-
Clint Johns -
Douglas Greve -
Douglas N Greve