External Email - Use Caution
Hi Douglas,
Actually I have one more question.
I went to look at the contrast examples on the wiki and in the example that fits my analysis (Two Groups (One Factor/Two Levels), One Covariate) it says "DODS only", and you mentioned in the previous email that my analysis turned out to be DOSS, because the interaction is non-significant.
So my question is why on the wiki (https://secure-web.cisco.com/1lSrkY1qx9oAVl14F4HmZZXqrWillSwx2jFAY6OURvMGOy6...) it's like "DODS only"? My fear is that I have not explained my analysis correctly to you.
Best, Laura Willers de Souza
Master Student in Biochemistry
Zimmer Lab - Federal University of Rio Grande do Sul
- Porto Alegre/RS - Brazil
---------------------------------------------------------------------------------------------------
De: Douglas N. Greve dgreve@mgh.harvard.edu Enviado: domingo, 02 de janeiro de 2022 Para: Laura Willers de Souza laura.w.desouza@hotmail.com Cc: freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu Assunto: Re: [Freesurfer] Questions about contrasts
Dear Douglas,
Again, thanks for the reply.
Please check if I understand correctly:
* First I do the contrast [0 0 -1 1] to test if there is an interaction between diagnosis and cortisol; if there is, I do DOSS [1 -1 0], if there isnt’t, I continue with DODS. * Having no interaction, I do the contrast [1 -1 0 0] to check if only the diagnosis changes cortical thickness regardless of cortisol.
Did I interpret correctly?
You have it backwards. If there is no interaction, use DOSS. If there is an interaction, consult a statistician. Fortunately, there is no interaction (whew!)
And how do I check if cortisol changes cortical thickness regardless of diagnosis (assuming there is no interaction)?
[0 0 1] with DOSS
Thanks again!
Best,
*Laura Willers de Souza*
---------------------------------------------------------------------------------------------------
De: Laura Willers de Souza laura.w.desouza@hotmail.com Enviado: quarta-feira, 15 de dezembro de 2021 Para: Douglas N. Greve dgreve@mgh.harvard.edu Cc: freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu Assunto: Re: Questions about contrasts
Dear Douglas,
Again, thanks for the reply.
Please check if I understand correctly:
* First I do the contrast [0 0 -1 1] to test if there is an interaction between diagnosis and cortisol; if there is, I do DOSS [1 -1 0], if there isnt’t, I continue with DODS. * Having no interaction, I do the contrast [1 -1 0 0] to check if only the diagnosis changes cortical thickness regardless of cortisol.
Did I interpret correctly?
And how do I check if cortisol changes cortical thickness regardless of diagnosis (assuming there is no interaction)?
Thanks again!
Best, Laura Willers de Souza
---------------------------------------------------------------------------------------------------
De: Douglas N. Greve dgreve@mgh.harvard.edu Enviado: terça-feira, 14 de dezembro de 2021 Para: Laura Willers de Souza laura.w.desouza@hotmail.com Cc: freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu Assunto: Re: [Freesurfer] Questions about contrasts
This contrast will do it: [1 -1 0 0]
But I would first test [0 0 -1 1] to see if the slopes are different. If not, then use [1 -1 0] with DOSS instead of DODS
---------------------------------------------------------------------------------------------------
De: Laura Willers de Souza laura.w.desouza@hotmail.com Enviado: terça-feira, 14 de dezembro de 2021 17:30 Para: Douglas N. Greve dgreve@mgh.harvard.edu Cc: freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu Assunto: Re: Questions about contrasts
Dear Douglas,
Thank you so much for your answer!
Sorry, but I don't think I explained myself correctly in the previous email.
I want to analyze the difference in cortical thickness between two groups (CU and CI), adjusting for a covariate (cortisol CSF) and the interaction between them.
So my model would be: cortical thickness ~ diagnosis + CSF cortisol + diagnosis*CSF cortisol
Using the GLM DODS, the contrasts, according to the wiki example (https://secure-web.cisco.com/1lSrkY1qx9oAVl14F4HmZZXqrWillSwx2jFAY6OURvMGOy6...https://secure-web.cisco.com/1qbucUNWhNS-2uW6Lox1w2bMK8EUHU0lm2XC7ywMcgAz1opgFdJ9TpOloRtF1uTmyVc70aldT5QkaELBu9kC0X6YiGpVUzZ3hV6ImEDrYAPF9VgAmDOudc6SysPVlhaUvzzGFirVYNLChsA_roK3faZNzYnWbdGPEJSgasor3qtQpKeAN7XvnQ4X24hWQfi35Rh8kF9yIsndHzOZGNxOdCH7U2CUMc9KxEL3N9u2-8weS-Fz01NHj2B_xMyBA8BlbS0n_wekioRXCPViGU3a_0zVQxE5Wu9sLPwFsQ3peECtbgAqHVmzWbyks9AhB3fWY/https%3A%2F%2Fsurfer.nmr.mgh.harvard.edu%2Ffswiki%2FFsgdf2G1V), would be:
* To verify if only the diagnosis regression coefficient is significant (diagnosis changes cortical thickness regardless of cortisol):
1 -1 0 0
* To verify if the regression coefficient of the interaction between diagnosis and cortisol is significant (the two change cortical thickness dependently):
0 0 1 -1
* To verify if only the cortisol regression coefficient is significant (cortisol changes cortical thickness regardless of diagnosis):
0 0 0.5 0.5
Are these contrasts corret to represent the model we designed?
Thank you so much again!
Best regards,
Laura Willers de Souza
Master Student in Biochemistry
Zimmer Lab - Federal University of Rio Grande do Sul
- Porto Alegre/RS - Brazil
---------------------------------------------------------------------------------------------------
De: Douglas N. Greve dgreve@mgh.harvard.edu Enviado: terça-feira, 12 de dezembro de 2021 18:23 Para: Laura Willers de Souza laura.w.desouza@hotmail.com Cc: freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu Assunto: Re: [Freesurfer] Questions about contrasts
The model is actually, two independent linear models:
thicknessCU = CUOffset + CUcortisol*CUslope
thicknessCI = CIOffset + CIcortisol*CIslope
There will be four regression parameters:
[CUOffset CUslope CIOffset CIslope]
---------------------------------------------------------------------------------------------------
De: Douglas N. Greve dgreve@mgh.harvard.edu Enviado: terça-feira, 12 de dezembro de 2021 10:48 Para: Laura Willers de Souza laura.w.desouza@hotmail.com Cc: freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu Assunto: Re: [Freesurfer] Questions about contrasts
Hello FreeSurfer Developers,
I want to perform an analysis of the difference between two groups in cortical thickness, having cortisol levels as covariate.
This is my model: cortical thickness ~ pathology (CU/CI) + CSF cortisol + pathology*CSF cortisol
From the GLM DODS analysis I want to know if:
* the regression coefficient is significant only for the difference
between groups,
this could possibly mean two things: the difference in offset [1 -1 0 0] or the difference in slopes ([0 0 1 -1]
* or only for cortisol,
[0 0 0.5 0.5] will test whether the cortisol slope is different than 0
* or for both,
both what?
* or if the coefficient of the intercept between group and cortisol
is significant.
this does not make sense to me
Is my FSGD file (attached) correct? What would the contrasts look like?
---------------------------------------------------------------------------------------------------
De: Laura Willers de Souza laura.w.desouza@hotmail.com Enviado: terça-feira, 09 de dezembro de 2021 19:50 Para: Douglas N. Greve dgreve@mgh.harvard.edu Cc: freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu Assunto: Questions about contrasts
Hello FreeSurfer Developers,
I want to perform an analysis of the difference between two groups in cortical thickness, having cortisol levels as covariate.
This is my model: cortical thickness ~ pathology (CU/CI) + CSF cortisol + pathology*CSF cortisol
From the GLM DODS analysis I want to know if:
* the regression coefficient is significant only for the difference between groups, * or only for cortisol, * or for both, * or if the coefficient of the intercept between group and cortisol is significant.
Is my FSGD file (attached) correct? What would the contrasts look like?
Thanks in advance!
Laura Willers de Souza
Master Student in Biochemistry
Zimmer Lab - Federal University of Rio Grande do Sul
- Porto Alegre/RS - Brazil
The contrasts are for DODS, you will need different contrasts for DOSS. I just added something onto that web page for DOSS
On 1/3/2022 10:10 PM, Laura Willers de Souza wrote:
External Email - Use Caution
Hi Douglas,
Actually I have one more question.
I went to look at the contrast examples on the wiki and in the example that fits my analysis (Two Groups (One Factor/Two Levels), One Covariate) it says "DODS only", and you mentioned in the previous email that my analysis turned out to be DOSS, because the interaction is non-significant.
So my question is why on the wiki (https://surfer.nmr.mgh.harvard.edu/fswiki/Fsgdf2G1V https://secure-web.cisco.com/1URi9Q-jl56ZpEMHrCqZrpvsv5PT3SKRjQnDVbwQuE-opRgESmFu0WTNlYDWtajuIHMT_w6hWtyhfqdiez9QCELg4ffzmYOtrgQ_wxNcUHZqFwWj4cwqvoAJdCnVN46emE4HMp9FMzZS3OreIqXSlYnBTS3uTaTX-zODfDN2SNvIESLamQSlLhVCJXV3vGAFqpPmaTRVwZoDpcannldUrH6pXKZ8t6xQM_BekqhMg__0dlG_rI7XHFgFFKj_jHml1fBDIxcUbP6L60CZx_bK1SKcsMw9DrJzTzEd9-u8jfQnTLOAcWWK_EXkQxHAFFZWb/https%3A%2F%2Fsurfer.nmr.mgh.harvard.edu%2Ffswiki%2FFsgdf2G1V) it's like "DODS only"? My fear is that I have not explained my analysis correctly to you.
Best,
*Laura Willers de Souza *
Master Student in Biochemistry
Zimmer Lab - Federal University of Rio Grande do Sul
- Porto Alegre/RS - Brazil
*De:* Douglas N. Greve dgreve@mgh.harvard.edu *Enviado:* domingo, 02 de janeiro de 2022 *Para:* Laura Willers de Souza laura.w.desouza@hotmail.com *Cc:*freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu *Assunto:* Re: [Freesurfer] Questions about contrasts
Dear Douglas, Again, thanks for the reply. Please check if I understand correctly: * First I do the contrast [0 0 -1 1] to test if there is an interaction between diagnosis and cortisol; if there is, I do DOSS [1 -1 0], if there isnt’t, I continue with DODS. * Having no interaction, I do the contrast [1 -1 0 0] to check if only the diagnosis changes cortical thickness regardless of cortisol. Did I interpret correctly?You have it backwards. If there is no interaction, use DOSS. If thereis an interaction, consult a statistician. Fortunately, there is nointeraction (whew!)
And how do I check if cortisol changes cortical thickness regardlessof diagnosis (assuming there is no interaction)?[0 0 1] with DOSS
Thanks again! Best, *Laura Willers de Souza*
*De:* Laura Willers de Souza laura.w.desouza@hotmail.com *Enviado:* quarta-feira, 15 de dezembro de 2021 *Para:* Douglas N. Greve dgreve@mgh.harvard.edu *Cc:*freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu *Assunto:* Re: Questions about contrasts
Dear Douglas, Again, thanks for the reply.
Please check if I understand correctly:
* First I do the contrast [0 0 -1 1] to test if there is an interaction between diagnosis and cortisol; if there is, I do DOSS [1 -1 0], if there isnt’t, I continue with DODS. * Having no interaction, I do the contrast [1 -1 0 0] to check if only the diagnosis changes cortical thickness regardless of cortisol.
Did I interpret correctly?
And how do I check if cortisol changes cortical thickness regardless of diagnosis (assuming there is no interaction)?
Thanks again!
Best, Laura Willers de Souza
*De:* Douglas N. Greve dgreve@mgh.harvard.edu *Enviado:* terça-feira, 14 de dezembro de 2021 *Para:* Laura Willers de Souza laura.w.desouza@hotmail.com *Cc:*freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu *Assunto:* Re: [Freesurfer] Questions about contrasts
This contrast will do it: [1 -1 0 0]
But I would first test [0 0 -1 1] to see if the slopes are different. If not, then use [1 -1 0] with DOSS instead of DODS
*De:* Laura Willers de Souza laura.w.desouza@hotmail.com *Enviado:* terça-feira, 14 de dezembro de 2021 17:30 *Para:* Douglas N. Greve dgreve@mgh.harvard.edu *Cc:*freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu *Assunto:* Re: Questions about contrasts
Dear Douglas,
Thank you so much for your answer!
Sorry, but I don't think I explained myself correctly in the previous email.
I want to analyze the difference in cortical thickness between two groups (CU and CI), adjusting for a covariate (cortisol CSF) and the interaction between them.
So my model would be: cortical thickness ~ diagnosis + CSF cortisol + diagnosis*CSF cortisol
Using the GLM DODS, the contrasts, according to the wiki example (https://surfer.nmr.mgh.harvard.edu/fswiki/Fsgdf2G1V https://secure-web.cisco.com/1qbucUNWhNS-2uW6Lox1w2bMK8EUHU0lm2XC7ywMcgAz1opgFdJ9TpOloRtF1uTmyVc70aldT5QkaELBu9kC0X6YiGpVUzZ3hV6ImEDrYAPF9VgAmDOudc6SysPVlhaUvzzGFirVYNLChsA_roK3faZNzYnWbdGPEJSgasor3qtQpKeAN7XvnQ4X24hWQfi35Rh8kF9yIsndHzOZGNxOdCH7U2CUMc9KxEL3N9u2-8weS-Fz01NHj2B_xMyBA8BlbS0n_wekioRXCPViGU3a_0zVQxE5Wu9sLPwFsQ3peECtbgAqHVmzWbyks9AhB3fWY/https%3A%2F%2Fsurfer.nmr.mgh.harvard.edu%2Ffswiki%2FFsgdf2G1V), would be:
- To verify if only the diagnosis regression coefficient is significant (diagnosis changes cortical thickness regardless of cortisol):
1 -1 0 0
- To verify if the regression coefficient of the interaction between diagnosis and cortisol is significant (the two change cortical thickness dependently):
0 0 1 -1
- To verify if only the cortisol regression coefficient is significant (cortisol changes cortical thickness regardless of diagnosis):
0 0 0.5 0.5
Are these contrasts corret to represent the model we designed?
Thank you so much again!
Best regards,
*Laura Willers de Souza*
Master Student in Biochemistry
Zimmer Lab - Federal University of Rio Grande do Sul
- Porto Alegre/RS - Brazil
*De:* Douglas N. Greve dgreve@mgh.harvard.edu *Enviado:* terça-feira, 12 de dezembro de 2021 18:23 *Para:* Laura Willers de Souza laura.w.desouza@hotmail.com *Cc:*freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu *Assunto:* Re: [Freesurfer] Questions about contrasts
The model is actually, two independent linear models:
thicknessCU = CUOffset + CUcortisol*CUslope
thicknessCI = CIOffset + CIcortisol*CIslope
There will be four regression parameters:
[CUOffset CUslope CIOffset CIslope]
*De:* Douglas N. Greve dgreve@mgh.harvard.edu *Enviado:* terça-feira, 12 de dezembro de 2021 10:48 *Para:* Laura Willers de Souza laura.w.desouza@hotmail.com *Cc:*freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu *Assunto:* Re: [Freesurfer] Questions about contrasts
Hello FreeSurfer Developers, I want to perform an analysis of the difference between two groups incortical thickness, having cortisol levels as covariate. This is my model: cortical thickness ~ pathology (CU/CI) + CSFcortisol + pathology*CSF cortisol From the GLM DODS analysis I want to know if: * the regression coefficient is significant only for the difference between groups,this could possibly mean two things: the difference in offset [1 -1 0 0]or the difference in slopes ([0 0 1 -1]
* or only for cortisol,[0 0 0.5 0.5] will test whether the cortisol slope is different than 0
* or for both,both what?
* or if the coefficient of the intercept between group and cortisol is significant.this does not make sense to me
Is my FSGD file (attached) correct? What would the contrasts look like?
*De:* Laura Willers de Souza laura.w.desouza@hotmail.com *Enviado:* terça-feira, 09 de dezembro de 2021 19:50 *Para:* Douglas N. Greve dgreve@mgh.harvard.edu *Cc:*freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu *Assunto:* Questions about contrasts
Hello FreeSurfer Developers,
I want to perform an analysis of the difference between two groups in cortical thickness, having cortisol levels as covariate.
This is my model: cortical thickness ~ pathology (CU/CI) + CSF cortisol + pathology*CSF cortisol
From the GLM DODS analysis I want to know if:
- the regression coefficient is significant only for the difference between groups,
- or only for cortisol,
- or for both,
- or if the coefficient of the intercept between group and cortisol is significant.
Is my FSGD file (attached) correct? What would the contrasts look like?
Thanks in advance!
*Laura Willers de Souza *
Master Student in Biochemistry
Zimmer Lab - Federal University of Rio Grande do Sul
- Porto Alegre/RS - Brazil
freesurfer@nmr.mgh.harvard.edu
-
Douglas N. Greve -
Laura Willers de Souza