help
489 Subscribers 
help > second-level analysis in Conn
Showing 26-43 of 43 posts
Oct 16, 2015 02:10 AM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
Hi Diana,
My guess would be that perhaps you used the 'import values' on an analysis that was only looking at Group1 subjects. If that was the case then the imported covariate will contain the connectivity values only for those subjects (and the rest of subjects are filled with NaN's; you may double-check this by looking in Setup.CovariatesSecondLevel and seeing whether the extracted covariates contain NaN values for other subjects). If that was the case then delete those partial second-level covariates, and simply go back to the Second-level results ROI-to-ROI tab with your original analysis, select the target regions of interest (e.g. the ones showing a significant effect in these analyses), then change the subject-effects list to select only the 'AllSubjects' effect before using 'import values' to now extract the connectivity values for all subjects instead.
Let me know if that seems to fix the missing-R2 issue.
Best
Alfonso
Originally posted by Diana Parvinchi:
My guess would be that perhaps you used the 'import values' on an analysis that was only looking at Group1 subjects. If that was the case then the imported covariate will contain the connectivity values only for those subjects (and the rest of subjects are filled with NaN's; you may double-check this by looking in Setup.CovariatesSecondLevel and seeing whether the extracted covariates contain NaN values for other subjects). If that was the case then delete those partial second-level covariates, and simply go back to the Second-level results ROI-to-ROI tab with your original analysis, select the target regions of interest (e.g. the ones showing a significant effect in these analyses), then change the subject-effects list to select only the 'AllSubjects' effect before using 'import values' to now extract the connectivity values for all subjects instead.
Let me know if that seems to fix the missing-R2 issue.
Best
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
This is a follow up question to our discussion below. I've selected all of the seed ROIs in the "sources" list and performed an F-test across all of our seeds. In order to explore each significant effect further (to determine where the effect is coming from), I select each significant effect from the analysis results and imported the values. Then, I use the option "calculator" to explore the effects further. In the calculator window, I select my group and their score (e.g. group 1, score 1), from the predictor variables, and enter the contrast [0 1] in the Between-subjects contrast and select a single pair of ROI from the list of outcome variables. A set of stats and a regression graph is provided here. Above the graph the R2 value is also displayed. However, for some reason, the R2 value is not displayed for my group 2, consistently. I'm not sure why and hoping that you could help me get this value. How can I get Conn to display the R2 for my group 2 as well?
Thank you very much for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
This is a follow up question to our discussion below. I've selected all of the seed ROIs in the "sources" list and performed an F-test across all of our seeds. In order to explore each significant effect further (to determine where the effect is coming from), I select each significant effect from the analysis results and imported the values. Then, I use the option "calculator" to explore the effects further. In the calculator window, I select my group and their score (e.g. group 1, score 1), from the predictor variables, and enter the contrast [0 1] in the Between-subjects contrast and select a single pair of ROI from the list of outcome variables. A set of stats and a regression graph is provided here. Above the graph the R2 value is also displayed. However, for some reason, the R2 value is not displayed for my group 2, consistently. I'm not sure why and hoping that you could help me get this value. How can I get Conn to display the R2 for my group 2 as well?
Thank you very much for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
Hi Alfonso,
That's very helpful. Thank you very much! We deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
That's very helpful. Thank you very much! We deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
Some thoughts on your questions below
Best
Alfonso
Originally posted by Diana Parvinchi:
No, the "analysis results" table multiple-comparison correction only corrects at the analysis-level (i.e. for that particular analysis -with a particular seed ROI and particular symptom score- it corrects across multiple target ROIs). Since you are performing multiple analyses (e.g. you are repeating this procedure for multiple seed ROIs) you would typically need to apply an additional correction to take that into account (unless each repeated analysis can be reasonably framed as a separate individual experimental hypothesis). Typically, in addition to simply using a more conservative Bonferroni-corrected set of individual thresholds, there are a couple of more-senstive ways to correct across multiple seed ROIs:
a) if the set of seed ROIs is not too large, you may simply perform an F-test across all of your seeds: i.e. repeat your analyses, but instead of selecting a single seed ROI in the "sources" list select now all of the seed ROIs that you want to test simultaneously and leave the default eye(N) between-sources contrast. That will test for connectivity associations with clinical scores across any of the selected seed ROIs
or b) alternatively click on "results explorer", then select in the "sources" list all of your seed ROIs. You will be now looking at the matrix of effects between all selected seed ROIs and all selected target ROIs, and you have multiple ways to threshold the results here while properly correcting for multiple comparisons (e.g. using seed-level thresholds as well as connection-level thresholds, using network-based-statistics, etc.)
2) Given that the symptoms severity scores are not standardized, meaning a score of "0" indicates the absence of that symptom, not the average, is our interpretation (e.g. functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour in our third group) correct in the attached graph?
Yes, the centering of the symptom score variables does not affect the interpretation of the sign of the association (between connectivity values and symptom scores). What changes with centering is the interpretation of the y-intercepts (the first three bars in your plot). In your case it means that the connectivity between those regions is typically (in the absence of symptoms) positive, and that this positive connectivity decreases with severity of symptoms.
3) When we see a significant effect in one of our three age-groups, how can I run pairwise comparisons (following an interaction effect) to report the p-value associated with that particular effect? For example, in the attached graph, there's significant correlation between connectivity and repetitive behaviour only in our third group, how can I determine the p-value for this particular effect?
To look at the association between connectivity and symptom scores only in your third group, simply select the 'Group3' and 'Group3_scores' effects in the "subject-effects" list and enter a [0 1] between-subjects contrast.
Hope this helps
Alfonso
Some thoughts on your questions below
Best
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
Thank you very much for all your help. I have three questions and deeply appreciate your help in confirming these issues before we submit the results for publication purposes. Just to refresh your mind, we examined the correlation between functional connectivity and core symptoms of autism spectrum disorder in three age-groups of participants. This is how I analyzed the data - followed by my three questions:
I selected 6 regressors (three age-groups and their corresponding scores on a given clinical measure) from the "subject effects" list in the 2nd-level results (ROI to ROI), entered the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the "between-groups contrasts", then selected a single ROI from our list of ROIs and ran the contrast. I repeated this process for each ROI in our list (for each symptom - total of 5 symptoms).
1) Are the results shown in the "analysis results" table corrected for multiple comparisons, given the route I took explained above?
Thank you very much for all your help. I have three questions and deeply appreciate your help in confirming these issues before we submit the results for publication purposes. Just to refresh your mind, we examined the correlation between functional connectivity and core symptoms of autism spectrum disorder in three age-groups of participants. This is how I analyzed the data - followed by my three questions:
I selected 6 regressors (three age-groups and their corresponding scores on a given clinical measure) from the "subject effects" list in the 2nd-level results (ROI to ROI), entered the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the "between-groups contrasts", then selected a single ROI from our list of ROIs and ran the contrast. I repeated this process for each ROI in our list (for each symptom - total of 5 symptoms).
1) Are the results shown in the "analysis results" table corrected for multiple comparisons, given the route I took explained above?
No, the "analysis results" table multiple-comparison correction only corrects at the analysis-level (i.e. for that particular analysis -with a particular seed ROI and particular symptom score- it corrects across multiple target ROIs). Since you are performing multiple analyses (e.g. you are repeating this procedure for multiple seed ROIs) you would typically need to apply an additional correction to take that into account (unless each repeated analysis can be reasonably framed as a separate individual experimental hypothesis). Typically, in addition to simply using a more conservative Bonferroni-corrected set of individual thresholds, there are a couple of more-senstive ways to correct across multiple seed ROIs:
a) if the set of seed ROIs is not too large, you may simply perform an F-test across all of your seeds: i.e. repeat your analyses, but instead of selecting a single seed ROI in the "sources" list select now all of the seed ROIs that you want to test simultaneously and leave the default eye(N) between-sources contrast. That will test for connectivity associations with clinical scores across any of the selected seed ROIs
or b) alternatively click on "results explorer", then select in the "sources" list all of your seed ROIs. You will be now looking at the matrix of effects between all selected seed ROIs and all selected target ROIs, and you have multiple ways to threshold the results here while properly correcting for multiple comparisons (e.g. using seed-level thresholds as well as connection-level thresholds, using network-based-statistics, etc.)
2) Given that the symptoms severity scores are not standardized, meaning a score of "0" indicates the absence of that symptom, not the average, is our interpretation (e.g. functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour in our third group) correct in the attached graph?
Yes, the centering of the symptom score variables does not affect the interpretation of the sign of the association (between connectivity values and symptom scores). What changes with centering is the interpretation of the y-intercepts (the first three bars in your plot). In your case it means that the connectivity between those regions is typically (in the absence of symptoms) positive, and that this positive connectivity decreases with severity of symptoms.
3) When we see a significant effect in one of our three age-groups, how can I run pairwise comparisons (following an interaction effect) to report the p-value associated with that particular effect? For example, in the attached graph, there's significant correlation between connectivity and repetitive behaviour only in our third group, how can I determine the p-value for this particular effect?
To look at the association between connectivity and symptom scores only in your third group, simply select the 'Group3' and 'Group3_scores' effects in the "subject-effects" list and enter a [0 1] between-subjects contrast.
Hope this helps
Alfonso
Oct 16, 2015 03:10 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
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.
Originally posted by Alfonso Nieto-Castanon:
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.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
My guess would be that perhaps you used the 'import values' on an analysis that was only looking at Group1 subjects. If that was the case then the imported covariate will contain the connectivity values only for those subjects (and the rest of subjects are filled with NaN's; you may double-check this by looking in Setup.CovariatesSecondLevel and seeing whether the extracted covariates contain NaN values for other subjects). If that was the case then delete those partial second-level covariates, and simply go back to the Second-level results ROI-to-ROI tab with your original analysis, select the target regions of interest (e.g. the ones showing a significant effect in these analyses), then change the subject-effects list to select only the 'AllSubjects' effect before using 'import values' to now extract the connectivity values for all subjects instead.
Let me know if that seems to fix the missing-R2 issue.
Best
Alfonso
Originally posted by Diana Parvinchi:
My guess would be that perhaps you used the 'import values' on an analysis that was only looking at Group1 subjects. If that was the case then the imported covariate will contain the connectivity values only for those subjects (and the rest of subjects are filled with NaN's; you may double-check this by looking in Setup.CovariatesSecondLevel and seeing whether the extracted covariates contain NaN values for other subjects). If that was the case then delete those partial second-level covariates, and simply go back to the Second-level results ROI-to-ROI tab with your original analysis, select the target regions of interest (e.g. the ones showing a significant effect in these analyses), then change the subject-effects list to select only the 'AllSubjects' effect before using 'import values' to now extract the connectivity values for all subjects instead.
Let me know if that seems to fix the missing-R2 issue.
Best
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
This is a follow up question to our discussion below. I've selected all of the seed ROIs in the "sources" list and performed an F-test across all of our seeds. In order to explore each significant effect further (to determine where the effect is coming from), I select each significant effect from the analysis results and imported the values. Then, I use the option "calculator" to explore the effects further. In the calculator window, I select my group and their score (e.g. group 1, score 1), from the predictor variables, and enter the contrast [0 1] in the Between-subjects contrast and select a single pair of ROI from the list of outcome variables. A set of stats and a regression graph is provided here. Above the graph the R2 value is also displayed. However, for some reason, the R2 value is not displayed for my group 2, consistently. I'm not sure why and hoping that you could help me get this value. How can I get Conn to display the R2 for my group 2 as well?
Thank you very much for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
This is a follow up question to our discussion below. I've selected all of the seed ROIs in the "sources" list and performed an F-test across all of our seeds. In order to explore each significant effect further (to determine where the effect is coming from), I select each significant effect from the analysis results and imported the values. Then, I use the option "calculator" to explore the effects further. In the calculator window, I select my group and their score (e.g. group 1, score 1), from the predictor variables, and enter the contrast [0 1] in the Between-subjects contrast and select a single pair of ROI from the list of outcome variables. A set of stats and a regression graph is provided here. Above the graph the R2 value is also displayed. However, for some reason, the R2 value is not displayed for my group 2, consistently. I'm not sure why and hoping that you could help me get this value. How can I get Conn to display the R2 for my group 2 as well?
Thank you very much for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
Hi Alfonso,
That's very helpful. Thank you very much! We deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
That's very helpful. Thank you very much! We deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
Some thoughts on your questions below
Best
Alfonso
Originally posted by Diana Parvinchi:
No, the "analysis results" table multiple-comparison correction only corrects at the analysis-level (i.e. for that particular analysis -with a particular seed ROI and particular symptom score- it corrects across multiple target ROIs). Since you are performing multiple analyses (e.g. you are repeating this procedure for multiple seed ROIs) you would typically need to apply an additional correction to take that into account (unless each repeated analysis can be reasonably framed as a separate individual experimental hypothesis). Typically, in addition to simply using a more conservative Bonferroni-corrected set of individual thresholds, there are a couple of more-senstive ways to correct across multiple seed ROIs:
a) if the set of seed ROIs is not too large, you may simply perform an F-test across all of your seeds: i.e. repeat your analyses, but instead of selecting a single seed ROI in the "sources" list select now all of the seed ROIs that you want to test simultaneously and leave the default eye(N) between-sources contrast. That will test for connectivity associations with clinical scores across any of the selected seed ROIs
or b) alternatively click on "results explorer", then select in the "sources" list all of your seed ROIs. You will be now looking at the matrix of effects between all selected seed ROIs and all selected target ROIs, and you have multiple ways to threshold the results here while properly correcting for multiple comparisons (e.g. using seed-level thresholds as well as connection-level thresholds, using network-based-statistics, etc.)
2) Given that the symptoms severity scores are not standardized, meaning a score of "0" indicates the absence of that symptom, not the average, is our interpretation (e.g. functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour in our third group) correct in the attached graph?
Yes, the centering of the symptom score variables does not affect the interpretation of the sign of the association (between connectivity values and symptom scores). What changes with centering is the interpretation of the y-intercepts (the first three bars in your plot). In your case it means that the connectivity between those regions is typically (in the absence of symptoms) positive, and that this positive connectivity decreases with severity of symptoms.
3) When we see a significant effect in one of our three age-groups, how can I run pairwise comparisons (following an interaction effect) to report the p-value associated with that particular effect? For example, in the attached graph, there's significant correlation between connectivity and repetitive behaviour only in our third group, how can I determine the p-value for this particular effect?
To look at the association between connectivity and symptom scores only in your third group, simply select the 'Group3' and 'Group3_scores' effects in the "subject-effects" list and enter a [0 1] between-subjects contrast.
Hope this helps
Alfonso
Some thoughts on your questions below
Best
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
Thank you very much for all your help. I have three questions and deeply appreciate your help in confirming these issues before we submit the results for publication purposes. Just to refresh your mind, we examined the correlation between functional connectivity and core symptoms of autism spectrum disorder in three age-groups of participants. This is how I analyzed the data - followed by my three questions:
I selected 6 regressors (three age-groups and their corresponding scores on a given clinical measure) from the "subject effects" list in the 2nd-level results (ROI to ROI), entered the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the "between-groups contrasts", then selected a single ROI from our list of ROIs and ran the contrast. I repeated this process for each ROI in our list (for each symptom - total of 5 symptoms).
1) Are the results shown in the "analysis results" table corrected for multiple comparisons, given the route I took explained above?
Thank you very much for all your help. I have three questions and deeply appreciate your help in confirming these issues before we submit the results for publication purposes. Just to refresh your mind, we examined the correlation between functional connectivity and core symptoms of autism spectrum disorder in three age-groups of participants. This is how I analyzed the data - followed by my three questions:
I selected 6 regressors (three age-groups and their corresponding scores on a given clinical measure) from the "subject effects" list in the 2nd-level results (ROI to ROI), entered the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the "between-groups contrasts", then selected a single ROI from our list of ROIs and ran the contrast. I repeated this process for each ROI in our list (for each symptom - total of 5 symptoms).
1) Are the results shown in the "analysis results" table corrected for multiple comparisons, given the route I took explained above?
No, the "analysis results" table multiple-comparison correction only corrects at the analysis-level (i.e. for that particular analysis -with a particular seed ROI and particular symptom score- it corrects across multiple target ROIs). Since you are performing multiple analyses (e.g. you are repeating this procedure for multiple seed ROIs) you would typically need to apply an additional correction to take that into account (unless each repeated analysis can be reasonably framed as a separate individual experimental hypothesis). Typically, in addition to simply using a more conservative Bonferroni-corrected set of individual thresholds, there are a couple of more-senstive ways to correct across multiple seed ROIs:
a) if the set of seed ROIs is not too large, you may simply perform an F-test across all of your seeds: i.e. repeat your analyses, but instead of selecting a single seed ROI in the "sources" list select now all of the seed ROIs that you want to test simultaneously and leave the default eye(N) between-sources contrast. That will test for connectivity associations with clinical scores across any of the selected seed ROIs
or b) alternatively click on "results explorer", then select in the "sources" list all of your seed ROIs. You will be now looking at the matrix of effects between all selected seed ROIs and all selected target ROIs, and you have multiple ways to threshold the results here while properly correcting for multiple comparisons (e.g. using seed-level thresholds as well as connection-level thresholds, using network-based-statistics, etc.)
2) Given that the symptoms severity scores are not standardized, meaning a score of "0" indicates the absence of that symptom, not the average, is our interpretation (e.g. functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour in our third group) correct in the attached graph?
Yes, the centering of the symptom score variables does not affect the interpretation of the sign of the association (between connectivity values and symptom scores). What changes with centering is the interpretation of the y-intercepts (the first three bars in your plot). In your case it means that the connectivity between those regions is typically (in the absence of symptoms) positive, and that this positive connectivity decreases with severity of symptoms.
3) When we see a significant effect in one of our three age-groups, how can I run pairwise comparisons (following an interaction effect) to report the p-value associated with that particular effect? For example, in the attached graph, there's significant correlation between connectivity and repetitive behaviour only in our third group, how can I determine the p-value for this particular effect?
To look at the association between connectivity and symptom scores only in your third group, simply select the 'Group3' and 'Group3_scores' effects in the "subject-effects" list and enter a [0 1] between-subjects contrast.
Hope this helps
Alfonso
Oct 16, 2015 04:10 PM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
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.
Originally posted by Alfonso Nieto-Castanon:
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.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
My guess would be that perhaps you used the 'import values' on an analysis that was only looking at Group1 subjects. If that was the case then the imported covariate will contain the connectivity values only for those subjects (and the rest of subjects are filled with NaN's; you may double-check this by looking in Setup.CovariatesSecondLevel and seeing whether the extracted covariates contain NaN values for other subjects). If that was the case then delete those partial second-level covariates, and simply go back to the Second-level results ROI-to-ROI tab with your original analysis, select the target regions of interest (e.g. the ones showing a significant effect in these analyses), then change the subject-effects list to select only the 'AllSubjects' effect before using 'import values' to now extract the connectivity values for all subjects instead.
Let me know if that seems to fix the missing-R2 issue.
Best
Alfonso
Originally posted by Diana Parvinchi:
My guess would be that perhaps you used the 'import values' on an analysis that was only looking at Group1 subjects. If that was the case then the imported covariate will contain the connectivity values only for those subjects (and the rest of subjects are filled with NaN's; you may double-check this by looking in Setup.CovariatesSecondLevel and seeing whether the extracted covariates contain NaN values for other subjects). If that was the case then delete those partial second-level covariates, and simply go back to the Second-level results ROI-to-ROI tab with your original analysis, select the target regions of interest (e.g. the ones showing a significant effect in these analyses), then change the subject-effects list to select only the 'AllSubjects' effect before using 'import values' to now extract the connectivity values for all subjects instead.
Let me know if that seems to fix the missing-R2 issue.
Best
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
This is a follow up question to our discussion below. I've selected all of the seed ROIs in the "sources" list and performed an F-test across all of our seeds. In order to explore each significant effect further (to determine where the effect is coming from), I select each significant effect from the analysis results and imported the values. Then, I use the option "calculator" to explore the effects further. In the calculator window, I select my group and their score (e.g. group 1, score 1), from the predictor variables, and enter the contrast [0 1] in the Between-subjects contrast and select a single pair of ROI from the list of outcome variables. A set of stats and a regression graph is provided here. Above the graph the R2 value is also displayed. However, for some reason, the R2 value is not displayed for my group 2, consistently. I'm not sure why and hoping that you could help me get this value. How can I get Conn to display the R2 for my group 2 as well?
Thank you very much for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
This is a follow up question to our discussion below. I've selected all of the seed ROIs in the "sources" list and performed an F-test across all of our seeds. In order to explore each significant effect further (to determine where the effect is coming from), I select each significant effect from the analysis results and imported the values. Then, I use the option "calculator" to explore the effects further. In the calculator window, I select my group and their score (e.g. group 1, score 1), from the predictor variables, and enter the contrast [0 1] in the Between-subjects contrast and select a single pair of ROI from the list of outcome variables. A set of stats and a regression graph is provided here. Above the graph the R2 value is also displayed. However, for some reason, the R2 value is not displayed for my group 2, consistently. I'm not sure why and hoping that you could help me get this value. How can I get Conn to display the R2 for my group 2 as well?
Thank you very much for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
Hi Alfonso,
That's very helpful. Thank you very much! We deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
That's very helpful. Thank you very much! We deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
Some thoughts on your questions below
Best
Alfonso
Originally posted by Diana Parvinchi:
No, the "analysis results" table multiple-comparison correction only corrects at the analysis-level (i.e. for that particular analysis -with a particular seed ROI and particular symptom score- it corrects across multiple target ROIs). Since you are performing multiple analyses (e.g. you are repeating this procedure for multiple seed ROIs) you would typically need to apply an additional correction to take that into account (unless each repeated analysis can be reasonably framed as a separate individual experimental hypothesis). Typically, in addition to simply using a more conservative Bonferroni-corrected set of individual thresholds, there are a couple of more-senstive ways to correct across multiple seed ROIs:
a) if the set of seed ROIs is not too large, you may simply perform an F-test across all of your seeds: i.e. repeat your analyses, but instead of selecting a single seed ROI in the "sources" list select now all of the seed ROIs that you want to test simultaneously and leave the default eye(N) between-sources contrast. That will test for connectivity associations with clinical scores across any of the selected seed ROIs
or b) alternatively click on "results explorer", then select in the "sources" list all of your seed ROIs. You will be now looking at the matrix of effects between all selected seed ROIs and all selected target ROIs, and you have multiple ways to threshold the results here while properly correcting for multiple comparisons (e.g. using seed-level thresholds as well as connection-level thresholds, using network-based-statistics, etc.)
2) Given that the symptoms severity scores are not standardized, meaning a score of "0" indicates the absence of that symptom, not the average, is our interpretation (e.g. functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour in our third group) correct in the attached graph?
Yes, the centering of the symptom score variables does not affect the interpretation of the sign of the association (between connectivity values and symptom scores). What changes with centering is the interpretation of the y-intercepts (the first three bars in your plot). In your case it means that the connectivity between those regions is typically (in the absence of symptoms) positive, and that this positive connectivity decreases with severity of symptoms.
3) When we see a significant effect in one of our three age-groups, how can I run pairwise comparisons (following an interaction effect) to report the p-value associated with that particular effect? For example, in the attached graph, there's significant correlation between connectivity and repetitive behaviour only in our third group, how can I determine the p-value for this particular effect?
To look at the association between connectivity and symptom scores only in your third group, simply select the 'Group3' and 'Group3_scores' effects in the "subject-effects" list and enter a [0 1] between-subjects contrast.
Hope this helps
Alfonso
Some thoughts on your questions below
Best
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
Thank you very much for all your help. I have three questions and deeply appreciate your help in confirming these issues before we submit the results for publication purposes. Just to refresh your mind, we examined the correlation between functional connectivity and core symptoms of autism spectrum disorder in three age-groups of participants. This is how I analyzed the data - followed by my three questions:
I selected 6 regressors (three age-groups and their corresponding scores on a given clinical measure) from the "subject effects" list in the 2nd-level results (ROI to ROI), entered the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the "between-groups contrasts", then selected a single ROI from our list of ROIs and ran the contrast. I repeated this process for each ROI in our list (for each symptom - total of 5 symptoms).
1) Are the results shown in the "analysis results" table corrected for multiple comparisons, given the route I took explained above?
Thank you very much for all your help. I have three questions and deeply appreciate your help in confirming these issues before we submit the results for publication purposes. Just to refresh your mind, we examined the correlation between functional connectivity and core symptoms of autism spectrum disorder in three age-groups of participants. This is how I analyzed the data - followed by my three questions:
I selected 6 regressors (three age-groups and their corresponding scores on a given clinical measure) from the "subject effects" list in the 2nd-level results (ROI to ROI), entered the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the "between-groups contrasts", then selected a single ROI from our list of ROIs and ran the contrast. I repeated this process for each ROI in our list (for each symptom - total of 5 symptoms).
1) Are the results shown in the "analysis results" table corrected for multiple comparisons, given the route I took explained above?
No, the "analysis results" table multiple-comparison correction only corrects at the analysis-level (i.e. for that particular analysis -with a particular seed ROI and particular symptom score- it corrects across multiple target ROIs). Since you are performing multiple analyses (e.g. you are repeating this procedure for multiple seed ROIs) you would typically need to apply an additional correction to take that into account (unless each repeated analysis can be reasonably framed as a separate individual experimental hypothesis). Typically, in addition to simply using a more conservative Bonferroni-corrected set of individual thresholds, there are a couple of more-senstive ways to correct across multiple seed ROIs:
a) if the set of seed ROIs is not too large, you may simply perform an F-test across all of your seeds: i.e. repeat your analyses, but instead of selecting a single seed ROI in the "sources" list select now all of the seed ROIs that you want to test simultaneously and leave the default eye(N) between-sources contrast. That will test for connectivity associations with clinical scores across any of the selected seed ROIs
or b) alternatively click on "results explorer", then select in the "sources" list all of your seed ROIs. You will be now looking at the matrix of effects between all selected seed ROIs and all selected target ROIs, and you have multiple ways to threshold the results here while properly correcting for multiple comparisons (e.g. using seed-level thresholds as well as connection-level thresholds, using network-based-statistics, etc.)
2) Given that the symptoms severity scores are not standardized, meaning a score of "0" indicates the absence of that symptom, not the average, is our interpretation (e.g. functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour in our third group) correct in the attached graph?
Yes, the centering of the symptom score variables does not affect the interpretation of the sign of the association (between connectivity values and symptom scores). What changes with centering is the interpretation of the y-intercepts (the first three bars in your plot). In your case it means that the connectivity between those regions is typically (in the absence of symptoms) positive, and that this positive connectivity decreases with severity of symptoms.
3) When we see a significant effect in one of our three age-groups, how can I run pairwise comparisons (following an interaction effect) to report the p-value associated with that particular effect? For example, in the attached graph, there's significant correlation between connectivity and repetitive behaviour only in our third group, how can I determine the p-value for this particular effect?
To look at the association between connectivity and symptom scores only in your third group, simply select the 'Group3' and 'Group3_scores' effects in the "subject-effects" list and enter a [0 1] between-subjects contrast.
Hope this helps
Alfonso
Oct 16, 2015 04:10 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
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.
Originally posted by Alfonso Nieto-Castanon:
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.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
My guess would be that perhaps you used the 'import values' on an analysis that was only looking at Group1 subjects. If that was the case then the imported covariate will contain the connectivity values only for those subjects (and the rest of subjects are filled with NaN's; you may double-check this by looking in Setup.CovariatesSecondLevel and seeing whether the extracted covariates contain NaN values for other subjects). If that was the case then delete those partial second-level covariates, and simply go back to the Second-level results ROI-to-ROI tab with your original analysis, select the target regions of interest (e.g. the ones showing a significant effect in these analyses), then change the subject-effects list to select only the 'AllSubjects' effect before using 'import values' to now extract the connectivity values for all subjects instead.
Let me know if that seems to fix the missing-R2 issue.
Best
Alfonso
Originally posted by Diana Parvinchi:
My guess would be that perhaps you used the 'import values' on an analysis that was only looking at Group1 subjects. If that was the case then the imported covariate will contain the connectivity values only for those subjects (and the rest of subjects are filled with NaN's; you may double-check this by looking in Setup.CovariatesSecondLevel and seeing whether the extracted covariates contain NaN values for other subjects). If that was the case then delete those partial second-level covariates, and simply go back to the Second-level results ROI-to-ROI tab with your original analysis, select the target regions of interest (e.g. the ones showing a significant effect in these analyses), then change the subject-effects list to select only the 'AllSubjects' effect before using 'import values' to now extract the connectivity values for all subjects instead.
Let me know if that seems to fix the missing-R2 issue.
Best
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
This is a follow up question to our discussion below. I've selected all of the seed ROIs in the "sources" list and performed an F-test across all of our seeds. In order to explore each significant effect further (to determine where the effect is coming from), I select each significant effect from the analysis results and imported the values. Then, I use the option "calculator" to explore the effects further. In the calculator window, I select my group and their score (e.g. group 1, score 1), from the predictor variables, and enter the contrast [0 1] in the Between-subjects contrast and select a single pair of ROI from the list of outcome variables. A set of stats and a regression graph is provided here. Above the graph the R2 value is also displayed. However, for some reason, the R2 value is not displayed for my group 2, consistently. I'm not sure why and hoping that you could help me get this value. How can I get Conn to display the R2 for my group 2 as well?
Thank you very much for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
This is a follow up question to our discussion below. I've selected all of the seed ROIs in the "sources" list and performed an F-test across all of our seeds. In order to explore each significant effect further (to determine where the effect is coming from), I select each significant effect from the analysis results and imported the values. Then, I use the option "calculator" to explore the effects further. In the calculator window, I select my group and their score (e.g. group 1, score 1), from the predictor variables, and enter the contrast [0 1] in the Between-subjects contrast and select a single pair of ROI from the list of outcome variables. A set of stats and a regression graph is provided here. Above the graph the R2 value is also displayed. However, for some reason, the R2 value is not displayed for my group 2, consistently. I'm not sure why and hoping that you could help me get this value. How can I get Conn to display the R2 for my group 2 as well?
Thank you very much for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
Hi Alfonso,
That's very helpful. Thank you very much! We deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
That's very helpful. Thank you very much! We deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
Some thoughts on your questions below
Best
Alfonso
Originally posted by Diana Parvinchi:
No, the "analysis results" table multiple-comparison correction only corrects at the analysis-level (i.e. for that particular analysis -with a particular seed ROI and particular symptom score- it corrects across multiple target ROIs). Since you are performing multiple analyses (e.g. you are repeating this procedure for multiple seed ROIs) you would typically need to apply an additional correction to take that into account (unless each repeated analysis can be reasonably framed as a separate individual experimental hypothesis). Typically, in addition to simply using a more conservative Bonferroni-corrected set of individual thresholds, there are a couple of more-senstive ways to correct across multiple seed ROIs:
a) if the set of seed ROIs is not too large, you may simply perform an F-test across all of your seeds: i.e. repeat your analyses, but instead of selecting a single seed ROI in the "sources" list select now all of the seed ROIs that you want to test simultaneously and leave the default eye(N) between-sources contrast. That will test for connectivity associations with clinical scores across any of the selected seed ROIs
or b) alternatively click on "results explorer", then select in the "sources" list all of your seed ROIs. You will be now looking at the matrix of effects between all selected seed ROIs and all selected target ROIs, and you have multiple ways to threshold the results here while properly correcting for multiple comparisons (e.g. using seed-level thresholds as well as connection-level thresholds, using network-based-statistics, etc.)
2) Given that the symptoms severity scores are not standardized, meaning a score of "0" indicates the absence of that symptom, not the average, is our interpretation (e.g. functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour in our third group) correct in the attached graph?
Yes, the centering of the symptom score variables does not affect the interpretation of the sign of the association (between connectivity values and symptom scores). What changes with centering is the interpretation of the y-intercepts (the first three bars in your plot). In your case it means that the connectivity between those regions is typically (in the absence of symptoms) positive, and that this positive connectivity decreases with severity of symptoms.
3) When we see a significant effect in one of our three age-groups, how can I run pairwise comparisons (following an interaction effect) to report the p-value associated with that particular effect? For example, in the attached graph, there's significant correlation between connectivity and repetitive behaviour only in our third group, how can I determine the p-value for this particular effect?
To look at the association between connectivity and symptom scores only in your third group, simply select the 'Group3' and 'Group3_scores' effects in the "subject-effects" list and enter a [0 1] between-subjects contrast.
Hope this helps
Alfonso
Some thoughts on your questions below
Best
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
Thank you very much for all your help. I have three questions and deeply appreciate your help in confirming these issues before we submit the results for publication purposes. Just to refresh your mind, we examined the correlation between functional connectivity and core symptoms of autism spectrum disorder in three age-groups of participants. This is how I analyzed the data - followed by my three questions:
I selected 6 regressors (three age-groups and their corresponding scores on a given clinical measure) from the "subject effects" list in the 2nd-level results (ROI to ROI), entered the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the "between-groups contrasts", then selected a single ROI from our list of ROIs and ran the contrast. I repeated this process for each ROI in our list (for each symptom - total of 5 symptoms).
1) Are the results shown in the "analysis results" table corrected for multiple comparisons, given the route I took explained above?
Thank you very much for all your help. I have three questions and deeply appreciate your help in confirming these issues before we submit the results for publication purposes. Just to refresh your mind, we examined the correlation between functional connectivity and core symptoms of autism spectrum disorder in three age-groups of participants. This is how I analyzed the data - followed by my three questions:
I selected 6 regressors (three age-groups and their corresponding scores on a given clinical measure) from the "subject effects" list in the 2nd-level results (ROI to ROI), entered the contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] in the "between-groups contrasts", then selected a single ROI from our list of ROIs and ran the contrast. I repeated this process for each ROI in our list (for each symptom - total of 5 symptoms).
1) Are the results shown in the "analysis results" table corrected for multiple comparisons, given the route I took explained above?
No, the "analysis results" table multiple-comparison correction only corrects at the analysis-level (i.e. for that particular analysis -with a particular seed ROI and particular symptom score- it corrects across multiple target ROIs). Since you are performing multiple analyses (e.g. you are repeating this procedure for multiple seed ROIs) you would typically need to apply an additional correction to take that into account (unless each repeated analysis can be reasonably framed as a separate individual experimental hypothesis). Typically, in addition to simply using a more conservative Bonferroni-corrected set of individual thresholds, there are a couple of more-senstive ways to correct across multiple seed ROIs:
a) if the set of seed ROIs is not too large, you may simply perform an F-test across all of your seeds: i.e. repeat your analyses, but instead of selecting a single seed ROI in the "sources" list select now all of the seed ROIs that you want to test simultaneously and leave the default eye(N) between-sources contrast. That will test for connectivity associations with clinical scores across any of the selected seed ROIs
or b) alternatively click on "results explorer", then select in the "sources" list all of your seed ROIs. You will be now looking at the matrix of effects between all selected seed ROIs and all selected target ROIs, and you have multiple ways to threshold the results here while properly correcting for multiple comparisons (e.g. using seed-level thresholds as well as connection-level thresholds, using network-based-statistics, etc.)
2) Given that the symptoms severity scores are not standardized, meaning a score of "0" indicates the absence of that symptom, not the average, is our interpretation (e.g. functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour in our third group) correct in the attached graph?
Yes, the centering of the symptom score variables does not affect the interpretation of the sign of the association (between connectivity values and symptom scores). What changes with centering is the interpretation of the y-intercepts (the first three bars in your plot). In your case it means that the connectivity between those regions is typically (in the absence of symptoms) positive, and that this positive connectivity decreases with severity of symptoms.
3) When we see a significant effect in one of our three age-groups, how can I run pairwise comparisons (following an interaction effect) to report the p-value associated with that particular effect? For example, in the attached graph, there's significant correlation between connectivity and repetitive behaviour only in our third group, how can I determine the p-value for this particular effect?
To look at the association between connectivity and symptom scores only in your third group, simply select the 'Group3' and 'Group3_scores' effects in the "subject-effects" list and enter a [0 1] between-subjects contrast.
Hope this helps
Alfonso
Oct 16, 2015 04:10 PM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
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. That could be a potential reason why Calculator would not identify your model as a proper regression equation and would refuse to show you the associated R2 values. 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. That could be a potential reason why Calculator would not identify your model as a proper regression equation and would refuse to show you the associated R2 values. 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.
Oct 16, 2015 05:10 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
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.
Oct 16, 2015 07:10 PM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
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.
Oct 19, 2015 04:10 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
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.
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.
Apr 18, 2016 04:04 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Hi Alfonso,
We have analyzed our data in Conn and have interesting results. I am following up with you on one particular effect. We originally examined the correlation between FC and symptom severities in three age groups of cohorts with autism spectrum disorder (ASD). One particular effect is showing that FC in superior frontal gyrus (with fusiform gyrus) is positively correlated with social impairment severities in our age group2. This means that as FC between these two regions increases, social impairments in this group increase as well. I would like to determine if the FC between these two regions is an inhibitory or an excitatory connection - negatively or positively correlated. At this point, the direction of the correlation has to do with FC and symptom severity. How do I determine if the FC itself between these two regions is positive or negative?
Many thanks for your help,
Diana.
Originally posted by Alfonso Nieto-Castanon:
We have analyzed our data in Conn and have interesting results. I am following up with you on one particular effect. We originally examined the correlation between FC and symptom severities in three age groups of cohorts with autism spectrum disorder (ASD). One particular effect is showing that FC in superior frontal gyrus (with fusiform gyrus) is positively correlated with social impairment severities in our age group2. This means that as FC between these two regions increases, social impairments in this group increase as well. I would like to determine if the FC between these two regions is an inhibitory or an excitatory connection - negatively or positively correlated. At this point, the direction of the correlation has to do with FC and symptom severity. How do I determine if the FC itself between these two regions is positive or negative?
Many thanks for your help,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
You may compute the association between symptom scores and connectivity within each of your three subject groups using the following procedure:
1) create three second-level covariates (Group1, Group2, and Group3) indicating your three subject groups (e.g. Group1 contains 1's for the subject in the first group and 0's for everyone else)
2) create three second-level covariates indicating the symptom scores within each of your three groups (e.g. Scores1 will contain the symptom scores for subjects in Group1 and 0's for everyone else; if you already have a second-level covariate named 'scores' containing all of the subjects' scores you may enter in the 'values' field of the new 'Scores1' covariate "Group1.*scores" -without the quotes- to create this new 'Scores1' covariate)
3) in the second-level analyses enter any of the following:
a) select 'AllSubjects' and 'scores' in the between-subject effects list and enter a between-subjects contrast [0 1] to look at the association between symptom scores and connectivity across all of your subjects (jointly across the three groups, disregarding group info)
b) select 'Group1', 'Group2', 'Group3', 'scores', and enter a contrast [0 0 0 1] to look at the association between symptom scores and connectivity across all of your subjects (jointly across the three groups) after discounting potential differences in average connectivity between your groups
c) select 'Group1' and 'Scores1' and enter a contrast [0 1] to look at the association between symptom scores and connectivity within your first group only
d) select 'Group1', 'Group2', 'Scores1', 'Scores', and enter a contrast [0 0 -1 1] to look at the difference between Group1 and Group2 in their association between symptom scores and connectivity
e) select 'Group1', 'Group2', 'Group3', 'Scores1', 'Scores2', 'Scores3', and enter a contrast [0 0 0 1 -1 0; 0 0 0 0 1 -1] to look at any differences between your groups in their association between symptom scores and connectivity
Regarding your second question, the beta values reported in the results table are the effect sizes of the chosen contrast. Mathematically if B is the matrix of regression coefficients of your second-level model, C is your between-subjects contrast, and D is your between-conditions contrast, then the effect sizes are C*B*D' (note: if any of your between-subjects or between-conditions contrast is a matrix instead of a vector then C*B*D' is a vector and the results table reports the norm of this vector). The interpretation of these betas depends on the chosen model and contrasts. For example, in the example (a) above, the effect size of the chosen contrast represents the association / regression-coefficient between symptom scores and functional connectivity, and its units are "increases in functional connectivity -fisher transformed correlation coefficients- associated with each unit increase in symptom scores". When looking at main connectivity effects (e.g. if you select 'AllSubjects' and enter a [1] contrast) then the reported effect sizes represent average functional connectivity -fisher transformed coefficients- among the tested subjects). Let me know if you would like me to further clarify any of the above.
Hope this helps
Alfonso
Originally posted by Diana P:
You may compute the association between symptom scores and connectivity within each of your three subject groups using the following procedure:
1) create three second-level covariates (Group1, Group2, and Group3) indicating your three subject groups (e.g. Group1 contains 1's for the subject in the first group and 0's for everyone else)
2) create three second-level covariates indicating the symptom scores within each of your three groups (e.g. Scores1 will contain the symptom scores for subjects in Group1 and 0's for everyone else; if you already have a second-level covariate named 'scores' containing all of the subjects' scores you may enter in the 'values' field of the new 'Scores1' covariate "Group1.*scores" -without the quotes- to create this new 'Scores1' covariate)
3) in the second-level analyses enter any of the following:
a) select 'AllSubjects' and 'scores' in the between-subject effects list and enter a between-subjects contrast [0 1] to look at the association between symptom scores and connectivity across all of your subjects (jointly across the three groups, disregarding group info)
b) select 'Group1', 'Group2', 'Group3', 'scores', and enter a contrast [0 0 0 1] to look at the association between symptom scores and connectivity across all of your subjects (jointly across the three groups) after discounting potential differences in average connectivity between your groups
c) select 'Group1' and 'Scores1' and enter a contrast [0 1] to look at the association between symptom scores and connectivity within your first group only
d) select 'Group1', 'Group2', 'Scores1', 'Scores', and enter a contrast [0 0 -1 1] to look at the difference between Group1 and Group2 in their association between symptom scores and connectivity
e) select 'Group1', 'Group2', 'Group3', 'Scores1', 'Scores2', 'Scores3', and enter a contrast [0 0 0 1 -1 0; 0 0 0 0 1 -1] to look at any differences between your groups in their association between symptom scores and connectivity
Regarding your second question, the beta values reported in the results table are the effect sizes of the chosen contrast. Mathematically if B is the matrix of regression coefficients of your second-level model, C is your between-subjects contrast, and D is your between-conditions contrast, then the effect sizes are C*B*D' (note: if any of your between-subjects or between-conditions contrast is a matrix instead of a vector then C*B*D' is a vector and the results table reports the norm of this vector). The interpretation of these betas depends on the chosen model and contrasts. For example, in the example (a) above, the effect size of the chosen contrast represents the association / regression-coefficient between symptom scores and functional connectivity, and its units are "increases in functional connectivity -fisher transformed correlation coefficients- associated with each unit increase in symptom scores". When looking at main connectivity effects (e.g. if you select 'AllSubjects' and enter a [1] contrast) then the reported effect sizes represent average functional connectivity -fisher transformed coefficients- among the tested subjects). Let me know if you would like me to further clarify any of the above.
Hope this helps
Alfonso
Originally posted by Diana P:
Hi,
I have two questions about Conn's second-level analysis and would appreciate your help. We are interested in examining the association between core symptoms of autism spectrum disorder (ASD) (repetitive behaviour, socialization, etc.) and connectivity in three age groups (3-7, 8-12, 13-20 years) of individuals diagnosed with ASD. How do I specify the contrasts in the second-level analysis to examine this question? My second question is what does "beta" in the Analysis results represent in the second level analysis? For example, if I have three groups selected in "between-subjects contrasts" and two Seeds selected in the "Between-sources contrasts", what does "beta" in the Analysis results represent? Many thanks.
I have two questions about Conn's second-level analysis and would appreciate your help. We are interested in examining the association between core symptoms of autism spectrum disorder (ASD) (repetitive behaviour, socialization, etc.) and connectivity in three age groups (3-7, 8-12, 13-20 years) of individuals diagnosed with ASD. How do I specify the contrasts in the second-level analysis to examine this question? My second question is what does "beta" in the Analysis results represent in the second level analysis? For example, if I have three groups selected in "between-subjects contrasts" and two Seeds selected in the "Between-sources contrasts", what does "beta" in the Analysis results represent? Many thanks.
May 5, 2016 06:05 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Hi Alfonso,
I have one question and hope that you can help us with this question. Our analysis has revealed various effects. One in particular is showing a positive correlation between FC and social impairment in children with ASD. I have attached a graph displaying this relationship. On the x-axis, it is plotting symptom severity and on the y-axis, it is showing the FC between two ROIs. This figure is displaying a positive correlation between FC and symptoms severity. Meaning, as FC between these two ROIs increases, social impairment increases as well. However the FC between the two ROIs, in this case, could be an inhibitory or an excitatory one (positive or negative). How can I determine if the correlation between these two ROIs is positive or negative? Could we look at the y-axis and use that information? I would deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Diana P:
I have one question and hope that you can help us with this question. Our analysis has revealed various effects. One in particular is showing a positive correlation between FC and social impairment in children with ASD. I have attached a graph displaying this relationship. On the x-axis, it is plotting symptom severity and on the y-axis, it is showing the FC between two ROIs. This figure is displaying a positive correlation between FC and symptoms severity. Meaning, as FC between these two ROIs increases, social impairment increases as well. However the FC between the two ROIs, in this case, could be an inhibitory or an excitatory one (positive or negative). How can I determine if the correlation between these two ROIs is positive or negative? Could we look at the y-axis and use that information? I would deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Diana P:
Hi,
I have two questions about Conn's second-level analysis and would appreciate your help. We are interested in examining the association between core symptoms of autism spectrum disorder (ASD) (repetitive behaviour, socialization, etc.) and connectivity in three age groups (3-7, 8-12, 13-20 years) of individuals diagnosed with ASD. How do I specify the contrasts in the second-level analysis to examine this question? My second question is what does "beta" in the Analysis results represent in the second level analysis? For example, if I have three groups selected in "between-subjects contrasts" and two Seeds selected in the "Between-sources contrasts", what does "beta" in the Analysis results represent? Many thanks.
I have two questions about Conn's second-level analysis and would appreciate your help. We are interested in examining the association between core symptoms of autism spectrum disorder (ASD) (repetitive behaviour, socialization, etc.) and connectivity in three age groups (3-7, 8-12, 13-20 years) of individuals diagnosed with ASD. How do I specify the contrasts in the second-level analysis to examine this question? My second question is what does "beta" in the Analysis results represent in the second level analysis? For example, if I have three groups selected in "between-subjects contrasts" and two Seeds selected in the "Between-sources contrasts", what does "beta" in the Analysis results represent? Many thanks.
May 9, 2016 06:05 PM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
Hi Diana,
Yes, exactly, the y-axis in this display informs you of the actual FC values between these two ROIs for each subject, so in this case I would say that these are regions that show no correlations or small anticorrelations in low-severity children while trending towards positive correlations as symptom severity increases.
Hope this helps
Alfonso
Originally posted by Diana Parvinchi:
Yes, exactly, the y-axis in this display informs you of the actual FC values between these two ROIs for each subject, so in this case I would say that these are regions that show no correlations or small anticorrelations in low-severity children while trending towards positive correlations as symptom severity increases.
Hope this helps
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
I have one question and hope that you can help us with this question. Our analysis has revealed various effects. One in particular is showing a positive correlation between FC and social impairment in children with ASD. I have attached a graph displaying this relationship. On the x-axis, it is plotting symptom severity and on the y-axis, it is showing the FC between two ROIs. This figure is displaying a positive correlation between FC and symptoms severity. Meaning, as FC between these two ROIs increases, social impairment increases as well. However the FC between the two ROIs, in this case, could be an inhibitory or an excitatory one (positive or negative). How can I determine if the correlation between these two ROIs is positive or negative? Could we look at the y-axis and use that information? I would deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Diana P:
I have one question and hope that you can help us with this question. Our analysis has revealed various effects. One in particular is showing a positive correlation between FC and social impairment in children with ASD. I have attached a graph displaying this relationship. On the x-axis, it is plotting symptom severity and on the y-axis, it is showing the FC between two ROIs. This figure is displaying a positive correlation between FC and symptoms severity. Meaning, as FC between these two ROIs increases, social impairment increases as well. However the FC between the two ROIs, in this case, could be an inhibitory or an excitatory one (positive or negative). How can I determine if the correlation between these two ROIs is positive or negative? Could we look at the y-axis and use that information? I would deeply appreciate your help. Many thanks.
Best,
Diana.
Originally posted by Diana P:
Hi,
I have two questions about Conn's second-level analysis and would appreciate your help. We are interested in examining the association between core symptoms of autism spectrum disorder (ASD) (repetitive behaviour, socialization, etc.) and connectivity in three age groups (3-7, 8-12, 13-20 years) of individuals diagnosed with ASD. How do I specify the contrasts in the second-level analysis to examine this question? My second question is what does "beta" in the Analysis results represent in the second level analysis? For example, if I have three groups selected in "between-subjects contrasts" and two Seeds selected in the "Between-sources contrasts", what does "beta" in the Analysis results represent? Many thanks.
I have two questions about Conn's second-level analysis and would appreciate your help. We are interested in examining the association between core symptoms of autism spectrum disorder (ASD) (repetitive behaviour, socialization, etc.) and connectivity in three age groups (3-7, 8-12, 13-20 years) of individuals diagnosed with ASD. How do I specify the contrasts in the second-level analysis to examine this question? My second question is what does "beta" in the Analysis results represent in the second level analysis? For example, if I have three groups selected in "between-subjects contrasts" and two Seeds selected in the "Between-sources contrasts", what does "beta" in the Analysis results represent? Many thanks.
May 10, 2016 01:05 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Alfonso, many many thanks for your help - much appreciated.
Best,
Diana.
Best,
Diana.
Jun 14, 2017 10:06 AM | marta555
RE: second-level analysis in Conn
Dear Alfonso,
Does this method apply to ICA studies? (e.g. do I have to select all ICA I want to test simultaneously or one by one?)
Many thanks,
Marta
Does this method apply to ICA studies? (e.g. do I have to select all ICA I want to test simultaneously or one by one?)
Many thanks,
Marta
Jun 16, 2017 01:06 AM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
Dear Marta,
Yes, just as with multiple seeds/ROIs, in the case of multiple ICA components you can also simply select all of the ones you are interested in and leave the default 'eye(n)' contrast in order to perform an F-test across any of the selected ICA components. That said, you do not typically want to select too many components simultaneously (because that reduces very drastically the degrees of freedom in your analyses) so in ICA you typically first use the 'Summary display' to identify one or a few networks that you are interested in and then focus the rest of your analyses only on that or those few networks.
Hope this helps
Alfonso
Originally posted by marta555:
Yes, just as with multiple seeds/ROIs, in the case of multiple ICA components you can also simply select all of the ones you are interested in and leave the default 'eye(n)' contrast in order to perform an F-test across any of the selected ICA components. That said, you do not typically want to select too many components simultaneously (because that reduces very drastically the degrees of freedom in your analyses) so in ICA you typically first use the 'Summary display' to identify one or a few networks that you are interested in and then focus the rest of your analyses only on that or those few networks.
Hope this helps
Alfonso
Originally posted by marta555:
Dear Alfonso,
Does this method apply to ICA studies? (e.g. do I have to select all ICA I want to test simultaneously or one by one?)
Many thanks,
Marta
Does this method apply to ICA studies? (e.g. do I have to select all ICA I want to test simultaneously or one by one?)
Many thanks,
Marta
Aug 3, 2018 09:08 PM | Kulpreet Cheema - University of Alberta
RE: second-level analysis in Conn
Hello, I was wondering if someone can help me understand how to
interpret results when running the test to look for difference
between group1 and group2 in their association between symptom
scores and connectivity (see contrast below).
d) select 'Group1', 'Group2', 'Scores1', 'Scores', and enter a contrast [0 0 -1 1] to look at the difference between Group1 and Group2 in their association between symptom scores and connectivity
How would you interpret the effect size and t-value in this case? Thank you!
d) select 'Group1', 'Group2', 'Scores1', 'Scores', and enter a contrast [0 0 -1 1] to look at the difference between Group1 and Group2 in their association between symptom scores and connectivity
How would you interpret the effect size and t-value in this case? Thank you!
Aug 4, 2018 02:08 PM | Jeff Browndyke
RE: second-level analysis in Conn
Kulpreet,
I believe in this case what you are testing are differences in the slope of score association between groups, in which case the null would be no differences in between groups in the association slopes. The t-value and effect size would be the extent of any association slope differences.
Regards,
Jeff Browndyke
I believe in this case what you are testing are differences in the slope of score association between groups, in which case the null would be no differences in between groups in the association slopes. The t-value and effect size would be the extent of any association slope differences.
Regards,
Jeff Browndyke
Aug 7, 2018 09:08 PM | Kulpreet Cheema - University of Alberta
RE: second-level analysis in Conn
Thanks Dr. Browndyke, that helps a lot!
I have a follow-up question. For the same contrast above, I got a result that has significant p-value (p>0.05) for connectivity-score but the beta/effect size is -0. I don't understand how can the beta/effect size both be 0 and significant? I have attached the screenshot of the results.
For information on my study, I am testing for association between functional connectivity and reaction time (RT) scores in individuals with (DYS) and without dyslexia(CONTROL). For this contrast, I selected CONTROL, DYS, RT_SCORES_CONTROL, RT_SCORES_DYS and did 0, 0, 1, -1.
Thank you!
I have a follow-up question. For the same contrast above, I got a result that has significant p-value (p>0.05) for connectivity-score but the beta/effect size is -0. I don't understand how can the beta/effect size both be 0 and significant? I have attached the screenshot of the results.
For information on my study, I am testing for association between functional connectivity and reaction time (RT) scores in individuals with (DYS) and without dyslexia(CONTROL). For this contrast, I selected CONTROL, DYS, RT_SCORES_CONTROL, RT_SCORES_DYS and did 0, 0, 1, -1.
Thank you!