help
help > RE: second-level analysis in Conn
Nov 3, 2015 05:11 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Hi Alfonso,
Thank you very much for all your help so far. I just wanted to confirm whether we should incorporate additional steps to control for multiple comparisons in the calculator viewing or not. Just to refresh your memory, we are examining the correlation between FC and symptom severity in children with ASD. At the second level analysis (ROI to ROI), we selected six regressors (3 age-groups and their scores: group1, group2, group3, score1, score2, score3) from the Subject Effects. Then, we entered the [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the Between-Subjects contrast. We selected all ROI's (36) from the Seeds/Sources and used the default eye(N) as the between-sources contrast. Several ROI's were revealed to be significant. However, we could not determine which factor(s) were driving the effect (which two ROI's and which age-group). In order to explore these results further, each significant effect was selected and the values imported. We used the "calculator" viewing to explore the effects further. In the calculator viewing, we selected the same six regressors (group 1, group2, group3, score1, score2, score3) and the same contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1]. Then, we scrolled down the Predictor Variables to find the significant FCs. When one was spotted, in order to determine which age-group was responsible for the effect, we selected an age-group and their score (Group 1, score 1) and entered the contrast [0 1] and repeated it for group2 and group3. We recorded the stats for the age-group that was driving the significant effect. I know the model (the f-test) accounts for multiple comparisons, but do we have to do additional controls, at this stage, for these pairwise comparisons?
We would deeply appreciate your input on this. Many thanks.
Best,
Diana.
Originally posted by Diana Parvinchi:
Thank you very much for all your help so far. I just wanted to confirm whether we should incorporate additional steps to control for multiple comparisons in the calculator viewing or not. Just to refresh your memory, we are examining the correlation between FC and symptom severity in children with ASD. At the second level analysis (ROI to ROI), we selected six regressors (3 age-groups and their scores: group1, group2, group3, score1, score2, score3) from the Subject Effects. Then, we entered the [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the Between-Subjects contrast. We selected all ROI's (36) from the Seeds/Sources and used the default eye(N) as the between-sources contrast. Several ROI's were revealed to be significant. However, we could not determine which factor(s) were driving the effect (which two ROI's and which age-group). In order to explore these results further, each significant effect was selected and the values imported. We used the "calculator" viewing to explore the effects further. In the calculator viewing, we selected the same six regressors (group 1, group2, group3, score1, score2, score3) and the same contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1]. Then, we scrolled down the Predictor Variables to find the significant FCs. When one was spotted, in order to determine which age-group was responsible for the effect, we selected an age-group and their score (Group 1, score 1) and entered the contrast [0 1] and repeated it for group2 and group3. We recorded the stats for the age-group that was driving the significant effect. I know the model (the f-test) accounts for multiple comparisons, but do we have to do additional controls, at this stage, for these pairwise comparisons?
We would deeply appreciate your input on this. Many thanks.
Best,
Diana.
Originally posted by Diana Parvinchi:
Hi Alfonso,
That's great to know - we did not know that. Thank you very much for letting us know.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
That's great to know - we did not know that. Thank you very much for letting us know.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
Excellent, happy to help! And just for reference, another alternative approach to scrubbing a subject is to enter in the AllSubjects and/or the appropriate Group# regressor the value NaN for that subject (instead of 0). That makes sure that any analysis that includes any of these AllSubjects or Group# regressors will remove that subject (even if the other variables/covariates included in that analysis have not been updated and still contain non-zero values for this scrubbed subject)
Best
Alfonso
Originally posted by Diana Parvinchi:
Excellent, happy to help! And just for reference, another alternative approach to scrubbing a subject is to enter in the AllSubjects and/or the appropriate Group# regressor the value NaN for that subject (instead of 0). That makes sure that any analysis that includes any of these AllSubjects or Group# regressors will remove that subject (even if the other variables/covariates included in that analysis have not been updated and still contain non-zero values for this scrubbed subject)
Best
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
You were correct! We had scrubbed a subject at a later point and had not incorporate that change in the corresponding clinical scores in Conn. We have corrected it and R2 is now displayed for group 2 as well. We deeply appreciate you looking into this and resolving it promptly. I'm glad I did not continue with the analysis and wrote to you. Many many thanks!
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
You were correct! We had scrubbed a subject at a later point and had not incorporate that change in the corresponding clinical scores in Conn. We have corrected it and R2 is now displayed for group 2 as well. We deeply appreciate you looking into this and resolving it promptly. I'm glad I did not continue with the analysis and wrote to you. Many many thanks!
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi Diana,
Could you also please double-check that the Group2 and Score2 variables are correctly defined? i.e. please check if there are any non-zero values in the Score2 variable for any subject who is not included in group 2 (for whom the Group2 variable has a zero value). One simple way to check this would be to create a new second-level covariate and enter in the 'values' field "~Group2 & Score2" (without the quotes and changing Group2 and Score2 to your actual second-level covariate names). If that new covariate contains any 1's then those subjects are not within group2 (they have a zero value in the Group2 variable) yet they have a non-zero Score2 value. If that does not seem to be the issue please send me your conn*.mat file and I will take a quick look to see if I can figure out the source of this issue.
Thanks
Alfonso
Originally posted by Diana Parvinchi:
Could you also please double-check that the Group2 and Score2 variables are correctly defined? i.e. please check if there are any non-zero values in the Score2 variable for any subject who is not included in group 2 (for whom the Group2 variable has a zero value). One simple way to check this would be to create a new second-level covariate and enter in the 'values' field "~Group2 & Score2" (without the quotes and changing Group2 and Score2 to your actual second-level covariate names). If that new covariate contains any 1's then those subjects are not within group2 (they have a zero value in the Group2 variable) yet they have a non-zero Score2 value. If that does not seem to be the issue please send me your conn*.mat file and I will take a quick look to see if I can figure out the source of this issue.
Thanks
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
Yes, in the Calculator display, I select my group 2 & score 2 from the predictor variables, enter the contrast [0 1] and select my FC variable as my outcome variable. Thanks very much for looking into this! Looking forward to hearing back from you.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Yes, in the Calculator display, I select my group 2 & score 2 from the predictor variables, enter the contrast [0 1] and select my FC variable as my outcome variable. Thanks very much for looking into this! Looking forward to hearing back from you.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
That is curious, the Calculator display will attempt to somehow "smartly" only display R2 values when they are relevant (i.e. when you have a regression-like equation) so perhaps the logic behind that code is not as "smart" as intended. I will double-check the code to see if I can figure out what might be causing this behavior; in the meantime could you please confirm that for your Group2 analyses you are selecting "Group2" and "Score2" as your predictor variables (contrast [0 1]) and your FC variable as your outcome variable?
Thanks
Alfonso
Originally posted by Diana Parvinchi:
That is curious, the Calculator display will attempt to somehow "smartly" only display R2 values when they are relevant (i.e. when you have a regression-like equation) so perhaps the logic behind that code is not as "smart" as intended. I will double-check the code to see if I can figure out what might be causing this behavior; in the meantime could you please confirm that for your Group2 analyses you are selecting "Group2" and "Score2" as your predictor variables (contrast [0 1]) and your FC variable as your outcome variable?
Thanks
Alfonso
Originally posted by Diana Parvinchi:
Hi
Alfonso,
Thank you very much for writing back. In the second level results window, I select my groups (group 1, group 2, group 3) and their scores (score1, score2, score3), enter the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1], select all of our ROIs (36) and leave the default eye (n) in the between sources contrast (this is to investigate association between FC and symptom severity in three age groups of cohorts with ASD). I don't enter "all subjects" in the Subject effects list, is that ok? I also looked in Setup.CovariatesSecondLevel and did not see NaN values for other subjects. Why do you think the R2 is missing in the calculator window for group 2 only (we get stats and the regression graph, but not R2 for this group)? Many many thanks for your help.
Best,
Diana.
Thank you very much for writing back. In the second level results window, I select my groups (group 1, group 2, group 3) and their scores (score1, score2, score3), enter the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1], select all of our ROIs (36) and leave the default eye (n) in the between sources contrast (this is to investigate association between FC and symptom severity in three age groups of cohorts with ASD). I don't enter "all subjects" in the Subject effects list, is that ok? I also looked in Setup.CovariatesSecondLevel and did not see NaN values for other subjects. Why do you think the R2 is missing in the calculator window for group 2 only (we get stats and the regression graph, but not R2 for this group)? Many many thanks for your help.
Best,
Diana.