help > WARNING sign and 2nd-Level Covariate
Showing 1-3 of 3 posts
May 24, 2019 04:05 AM | Dilip Kumar
WARNING sign and 2nd-Level Covariate
Dear Alfonso and CONN community,
Using CONN latest version, I have performed ROI connectivity analysis for 164 x 164 ROIs resting state data and classified the ROI connectivity results into two groups (e.g Group_A [1 0 1 0 .... n] and Group_B [0 1 0 1 .... n] under Setup Covariates 2nd-level tab). With aim to see group-level connectivity difference in all 164x164 ROIs and removing any age related effect, when I try to control for age in following 3 ways I read on this forum, I'm getting different results for each method;
1- Defining Age as single covariate with actual sample age values and choosing Group_A, Group_B, and Age under ROI Results Explorer tab (3 eye); for Between-subject Contrast [1 -1 0], I get the WARNING sign under Between-sources Contrast and connectivity results on attachment page 1.
2- Defining Age as separate group-level covariates with actual sample age values and zeros for non-group members and choosing Group_A, Group_B, and Age_A, Age_B under ROI Results Explorer tab (4 eye); for Between-subject Contrast [1 -1 0 0], I get the WARNING sign again under Between-sources Contrast and connectivity results on attachment page 2 (drastically different from page 1).
3- Defining Age as separate group-level covariates with centered sample age values (i.e. mean age Group_A/B minus actual sample age and zeros for non-group members) and choosing Group_A, Group_B, and Age_Centered_A, Age_Centered_B under ROI Results Explorer tab (4 eye); for Between-subject Contrast [1 -1 0 0], I get the WARNING sign again under Between-sources Contrast and connectivity results on attachment page 3 (greatly similar to page 1 but yet somewhat different at the same time).
Page 4 in attachment shows how the results look without using age as covariate. In a simple 2-sample TTEST on age vector, the age difference between two groups isn't significant (i.e p>0.05). Hence I was assuming the results would not change drastically when controlling for age. I'm of the view that all 3 methods were right and using any one would be right and should suffice. I was expecting similar results between the 3 methods, however results are quite different for each. Now I'm wondering on whats basis should I choose any one of the 3 ways above? And what the WARNING sign was about?
Can someone please shed some light on possible reason behind difference in results and WARNING sign? Thank you.
Sincerely,
Dilip
Using CONN latest version, I have performed ROI connectivity analysis for 164 x 164 ROIs resting state data and classified the ROI connectivity results into two groups (e.g Group_A [1 0 1 0 .... n] and Group_B [0 1 0 1 .... n] under Setup Covariates 2nd-level tab). With aim to see group-level connectivity difference in all 164x164 ROIs and removing any age related effect, when I try to control for age in following 3 ways I read on this forum, I'm getting different results for each method;
1- Defining Age as single covariate with actual sample age values and choosing Group_A, Group_B, and Age under ROI Results Explorer tab (3 eye); for Between-subject Contrast [1 -1 0], I get the WARNING sign under Between-sources Contrast and connectivity results on attachment page 1.
2- Defining Age as separate group-level covariates with actual sample age values and zeros for non-group members and choosing Group_A, Group_B, and Age_A, Age_B under ROI Results Explorer tab (4 eye); for Between-subject Contrast [1 -1 0 0], I get the WARNING sign again under Between-sources Contrast and connectivity results on attachment page 2 (drastically different from page 1).
3- Defining Age as separate group-level covariates with centered sample age values (i.e. mean age Group_A/B minus actual sample age and zeros for non-group members) and choosing Group_A, Group_B, and Age_Centered_A, Age_Centered_B under ROI Results Explorer tab (4 eye); for Between-subject Contrast [1 -1 0 0], I get the WARNING sign again under Between-sources Contrast and connectivity results on attachment page 3 (greatly similar to page 1 but yet somewhat different at the same time).
Page 4 in attachment shows how the results look without using age as covariate. In a simple 2-sample TTEST on age vector, the age difference between two groups isn't significant (i.e p>0.05). Hence I was assuming the results would not change drastically when controlling for age. I'm of the view that all 3 methods were right and using any one would be right and should suffice. I was expecting similar results between the 3 methods, however results are quite different for each. Now I'm wondering on whats basis should I choose any one of the 3 ways above? And what the WARNING sign was about?
Can someone please shed some light on possible reason behind difference in results and WARNING sign? Thank you.
Sincerely,
Dilip
May 25, 2019 04:05 PM | Alfonso Nieto-Castanon - Boston University
RE: WARNING sign and 2nd-Level Covariate
Dear Dilip,
If you click on the 'Warning' sign, a window will pop up displaying the design matrix and providing a few more details about the nature of the warning. Could you please copy/paste that description or better yet take a snapshot of that window and attach that information as well? (that should help figure out what is it that CONN finds questionable about your design)
Thanks
Alfonso
Originally posted by Dilip Kumar:
If you click on the 'Warning' sign, a window will pop up displaying the design matrix and providing a few more details about the nature of the warning. Could you please copy/paste that description or better yet take a snapshot of that window and attach that information as well? (that should help figure out what is it that CONN finds questionable about your design)
Thanks
Alfonso
Originally posted by Dilip Kumar:
Dear Alfonso and CONN community,
Using CONN latest version, I have performed ROI connectivity analysis for 164 x 164 ROIs resting state data and classified the ROI connectivity results into two groups (e.g Group_A [1 0 1 0 .... n] and Group_B [0 1 0 1 .... n] under Setup Covariates 2nd-level tab). With aim to see group-level connectivity difference in all 164x164 ROIs and removing any age related effect, when I try to control for age in following 3 ways I read on this forum, I'm getting different results for each method;
1- Defining Age as single covariate with actual sample age values and choosing Group_A, Group_B, and Age under ROI Results Explorer tab (3 eye); for Between-subject Contrast [1 -1 0], I get the WARNING sign under Between-sources Contrast and connectivity results on attachment page 1.
2- Defining Age as separate group-level covariates with actual sample age values and zeros for non-group members and choosing Group_A, Group_B, and Age_A, Age_B under ROI Results Explorer tab (4 eye); for Between-subject Contrast [1 -1 0 0], I get the WARNING sign again under Between-sources Contrast and connectivity results on attachment page 2 (drastically different from page 1).
3- Defining Age as separate group-level covariates with centered sample age values (i.e. mean age Group_A/B minus actual sample age and zeros for non-group members) and choosing Group_A, Group_B, and Age_Centered_A, Age_Centered_B under ROI Results Explorer tab (4 eye); for Between-subject Contrast [1 -1 0 0], I get the WARNING sign again under Between-sources Contrast and connectivity results on attachment page 3 (greatly similar to page 1 but yet somewhat different at the same time).
Page 4 in attachment shows how the results look without using age as covariate. In a simple 2-sample TTEST on age vector, the age difference between two groups isn't significant (i.e p>0.05). Hence I was assuming the results would not change drastically when controlling for age. I'm of the view that all 3 methods were right and using any one would be right and should suffice. I was expecting similar results between the 3 methods, however results are quite different for each. Now I'm wondering on whats basis should I choose any one of the 3 ways above? And what the WARNING sign was about?
Can someone please shed some light on possible reason behind difference in results and WARNING sign? Thank you.
Sincerely,
Dilip
Using CONN latest version, I have performed ROI connectivity analysis for 164 x 164 ROIs resting state data and classified the ROI connectivity results into two groups (e.g Group_A [1 0 1 0 .... n] and Group_B [0 1 0 1 .... n] under Setup Covariates 2nd-level tab). With aim to see group-level connectivity difference in all 164x164 ROIs and removing any age related effect, when I try to control for age in following 3 ways I read on this forum, I'm getting different results for each method;
1- Defining Age as single covariate with actual sample age values and choosing Group_A, Group_B, and Age under ROI Results Explorer tab (3 eye); for Between-subject Contrast [1 -1 0], I get the WARNING sign under Between-sources Contrast and connectivity results on attachment page 1.
2- Defining Age as separate group-level covariates with actual sample age values and zeros for non-group members and choosing Group_A, Group_B, and Age_A, Age_B under ROI Results Explorer tab (4 eye); for Between-subject Contrast [1 -1 0 0], I get the WARNING sign again under Between-sources Contrast and connectivity results on attachment page 2 (drastically different from page 1).
3- Defining Age as separate group-level covariates with centered sample age values (i.e. mean age Group_A/B minus actual sample age and zeros for non-group members) and choosing Group_A, Group_B, and Age_Centered_A, Age_Centered_B under ROI Results Explorer tab (4 eye); for Between-subject Contrast [1 -1 0 0], I get the WARNING sign again under Between-sources Contrast and connectivity results on attachment page 3 (greatly similar to page 1 but yet somewhat different at the same time).
Page 4 in attachment shows how the results look without using age as covariate. In a simple 2-sample TTEST on age vector, the age difference between two groups isn't significant (i.e p>0.05). Hence I was assuming the results would not change drastically when controlling for age. I'm of the view that all 3 methods were right and using any one would be right and should suffice. I was expecting similar results between the 3 methods, however results are quite different for each. Now I'm wondering on whats basis should I choose any one of the 3 ways above? And what the WARNING sign was about?
Can someone please shed some light on possible reason behind difference in results and WARNING sign? Thank you.
Sincerely,
Dilip
May 27, 2019 02:05 AM | Dilip Kumar
RE: WARNING sign and 2nd-Level Covariate
Dear Alfonso,
Thanks for looking into it. I have noticed that the WARNING sign appears only when I choose covariates like age or cognitive scores (continuous values). It does not appear when I choose these covariates in combination with subject groups hence I only took screen shot for age covariate. Please find the screen shot attached, it doesn't offer any WARNING description though despite showing WARNING instead of showing n=X. Could it be because I have n=63 in the pipeline and the covariate has 63 entries but with a zero entry for one subject which I want to discard due to quality control failure? Then again when I do add non-zero value for that one subject, the WARNING is still there.
Any thoughts on second part of my query (difference in results)?
Kind regards,
Dilip
Thanks for looking into it. I have noticed that the WARNING sign appears only when I choose covariates like age or cognitive scores (continuous values). It does not appear when I choose these covariates in combination with subject groups hence I only took screen shot for age covariate. Please find the screen shot attached, it doesn't offer any WARNING description though despite showing WARNING instead of showing n=X. Could it be because I have n=63 in the pipeline and the covariate has 63 entries but with a zero entry for one subject which I want to discard due to quality control failure? Then again when I do add non-zero value for that one subject, the WARNING is still there.
Any thoughts on second part of my query (difference in results)?
Kind regards,
Dilip