help
486 Subscribers 
help > second-level analysis in Conn
Showing 1-25 of 43 posts
Jul 31, 2015 08:07 PM | Diana P
second-level analysis in Conn
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.
Aug 3, 2015 05:08 AM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
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.
Aug 10, 2015 01:08 PM | Diana Parvinchi - McMaster University
second-level analysis in Conn
> (Hi Alfonso,
Thank you very much for your reply. That was very helpful. We have another questions for you. When we select �any effect� from the options provided under the �Between-subjects contrast� we find more significant effects than when �main effect� or simple contrasts are selected. What type of analysis does it ( �any effect�) run? Is it an f-test? If so, why does it produce more effects?
> )
Thank you very much for your reply. That was very helpful. We have another questions for you. When we select �any effect� from the options provided under the �Between-subjects contrast� we find more significant effects than when �main effect� or simple contrasts are selected. What type of analysis does it ( �any effect�) run? Is it an f-test? If so, why does it produce more effects?
> )
Aug 10, 2015 06:08 PM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
Hi Diana,
The 'All effects' contrast is an F-test using an 'eye(N)' contrast (where N is the number of effects selected in your 'subject effects' list). This looks at the significance of the entire second-level model (which typically includes main connectivity effects as well as any covariates entered) and it can be understood as an OR conjunction of each individual subject-effect covariate entered into your model. The particular interpretation depends on the specific subject-effects selected, for example if you are selecting 'Group1', 'Group2', and 'Group3', then the 'all effects' contrast will be [1 0 0; 0 1 0; 0 0 1] which tests an OR conjunction of the individual [1 0 0], [0 1 0], and [0 0 1] contrasts, or, in other words, those areas where the connectivity with your seed regions is different from zero within any of your three subject groups.
Hope this helps
Alfonso
Originally posted by Diana Parvinchi:
The 'All effects' contrast is an F-test using an 'eye(N)' contrast (where N is the number of effects selected in your 'subject effects' list). This looks at the significance of the entire second-level model (which typically includes main connectivity effects as well as any covariates entered) and it can be understood as an OR conjunction of each individual subject-effect covariate entered into your model. The particular interpretation depends on the specific subject-effects selected, for example if you are selecting 'Group1', 'Group2', and 'Group3', then the 'all effects' contrast will be [1 0 0; 0 1 0; 0 0 1] which tests an OR conjunction of the individual [1 0 0], [0 1 0], and [0 0 1] contrasts, or, in other words, those areas where the connectivity with your seed regions is different from zero within any of your three subject groups.
Hope this helps
Alfonso
Originally posted by Diana Parvinchi:
> (Hi Alfonso,
Thank you very much for your reply. That was very helpful. We have another questions for you. When we select �any effect� from the options provided under the �Between-subjects contrast� we find more significant effects than when �main effect� or simple contrasts are selected. What type of analysis does it ( �any effect�) run? Is it an f-test? If so, why does it produce more effects?
> )
Thank you very much for your reply. That was very helpful. We have another questions for you. When we select �any effect� from the options provided under the �Between-subjects contrast� we find more significant effects than when �main effect� or simple contrasts are selected. What type of analysis does it ( �any effect�) run? Is it an f-test? If so, why does it produce more effects?
> )
Aug 12, 2015 04:08 PM | Diana Parvinchi - McMaster University
second-level analysis in Conn
> (Hi Alfonso,
I followed your instructions, thank you very much. I am currently looking at the third option you listed - " 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". I have also selected 3 ROIs from the �seed/source� section. I�m new to Conn and not sure how/where to go to read the results. The table that pops out titled �connectivity values�, when all variables are selected, shows connectivity of each ROI to the one selected in the �Analysis results� for each group and it also lists the same information for the symptoms scores. In other words, it is treating the symptom scores for each group as additional independent groups even though these are within group variables. I�m looking for the correlation between the symptom score and connectivity within each group and whether it differs significantly across the groups. Where can I find/produce this information? I would deeply appreciate your help. Many thanks.
Best,
Diana.
> )
I followed your instructions, thank you very much. I am currently looking at the third option you listed - " 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". I have also selected 3 ROIs from the �seed/source� section. I�m new to Conn and not sure how/where to go to read the results. The table that pops out titled �connectivity values�, when all variables are selected, shows connectivity of each ROI to the one selected in the �Analysis results� for each group and it also lists the same information for the symptoms scores. In other words, it is treating the symptom scores for each group as additional independent groups even though these are within group variables. I�m looking for the correlation between the symptom score and connectivity within each group and whether it differs significantly across the groups. Where can I find/produce this information? I would deeply appreciate your help. Many thanks.
Best,
Diana.
> )
Aug 12, 2015 07:08 PM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
Hi Diana,
When you select the 'Group1', 'Group2', 'Group3', 'Scores1', 'Scores2', and 'Scores3' subject effects (in this order) and enter a contrast [0 0 0 1 -1 0; 0 0 0 0 1 -1], and select a single source/seed ROI, the results listed in the 'Analysis results' table (I am assuming here you are using ROI-to-ROI analyses) shows you which areas show significant differences between groups in their association between symptom scores and connectivity. If you find a significant result there and you select that target ROI in the 'analysis results' table you will get a popup window showing you the model regressors and associated confidence intervals. In this case there are six regressors, the first three represent the average connectivity within each of your three groups (I am assuming here that the 'Scores*' variables are centered within each group; otherwise these three regressors represent instead the estimated connectivity level at the zero-level of your score covariates within each group), and the next three represent the level/direction of association (regression coefficients) between symptom scores and connectivity within each of your three groups (see note below if in this display the bar sizes and errorbars for the last three effects appear too small to see). Looking at the sign and size of those 6 regressors helps you interpret the found differences. In particular the statistical test (the F/p values reported in the 'anaysis results' table) correspond to differences between the last three regressor values displayed in this barplot (differences between the regressor coefficients between symptom scores and connectivity within each group), so you expect to see "some"difference between those three values if you have selected a target ROI that shows a significant effect. Also the sign of those three regressors represents the direction of the association between symptom scores and connectivity within each group (e.g. if the first value is positive and the second negative that means that increases in symptom scores are associated with increases in connectivity in the first group, but they are associated with decreases in connectivity in the second group). If in doubt please send me a printout of this display and I will be happy to ellaborate how to interpret those results. Last, if you want to further explore/display the found effects, select the 'import values' option and that will create a new second-level covariate containing the connectivity between the source and target ROI for each subject. You may then go to Tools.Calculator to further display and analyze those values and their relationship with symptom scores in order to gain a better understanding of the found effects.
Hope this helps
Alfonso
note: If the scale of the first three regressors (in units of fisher-transformed correlation coefficients) and the last three regressors (in units of fisher-transformed coefficients divided per unit change in symptom scores) looks wildly different in the bar display I woud suggest to divide your 'Score1' to 'Score3' variables by a constant factor (i.e. divide all symptom score numbers by 100, or divide all symptom score numbers by the standard deviation of the score variables across all groups jointly). This will not change the analyses, it just makes the scale of those effects a bit more comparable for display purposes.
Originally posted by Diana Parvinchi:
When you select the 'Group1', 'Group2', 'Group3', 'Scores1', 'Scores2', and 'Scores3' subject effects (in this order) and enter a contrast [0 0 0 1 -1 0; 0 0 0 0 1 -1], and select a single source/seed ROI, the results listed in the 'Analysis results' table (I am assuming here you are using ROI-to-ROI analyses) shows you which areas show significant differences between groups in their association between symptom scores and connectivity. If you find a significant result there and you select that target ROI in the 'analysis results' table you will get a popup window showing you the model regressors and associated confidence intervals. In this case there are six regressors, the first three represent the average connectivity within each of your three groups (I am assuming here that the 'Scores*' variables are centered within each group; otherwise these three regressors represent instead the estimated connectivity level at the zero-level of your score covariates within each group), and the next three represent the level/direction of association (regression coefficients) between symptom scores and connectivity within each of your three groups (see note below if in this display the bar sizes and errorbars for the last three effects appear too small to see). Looking at the sign and size of those 6 regressors helps you interpret the found differences. In particular the statistical test (the F/p values reported in the 'anaysis results' table) correspond to differences between the last three regressor values displayed in this barplot (differences between the regressor coefficients between symptom scores and connectivity within each group), so you expect to see "some"difference between those three values if you have selected a target ROI that shows a significant effect. Also the sign of those three regressors represents the direction of the association between symptom scores and connectivity within each group (e.g. if the first value is positive and the second negative that means that increases in symptom scores are associated with increases in connectivity in the first group, but they are associated with decreases in connectivity in the second group). If in doubt please send me a printout of this display and I will be happy to ellaborate how to interpret those results. Last, if you want to further explore/display the found effects, select the 'import values' option and that will create a new second-level covariate containing the connectivity between the source and target ROI for each subject. You may then go to Tools.Calculator to further display and analyze those values and their relationship with symptom scores in order to gain a better understanding of the found effects.
Hope this helps
Alfonso
note: If the scale of the first three regressors (in units of fisher-transformed correlation coefficients) and the last three regressors (in units of fisher-transformed coefficients divided per unit change in symptom scores) looks wildly different in the bar display I woud suggest to divide your 'Score1' to 'Score3' variables by a constant factor (i.e. divide all symptom score numbers by 100, or divide all symptom score numbers by the standard deviation of the score variables across all groups jointly). This will not change the analyses, it just makes the scale of those effects a bit more comparable for display purposes.
Originally posted by Diana Parvinchi:
> (Hi Alfonso,
I followed your instructions, thank you very much. I am currently looking at the third option you listed - " 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". I have also selected 3 ROIs from the �seed/source� section. I�m new to Conn and not sure how/where to go to read the results. The table that pops out titled �connectivity values�, when all variables are selected, shows connectivity of each ROI to the one selected in the �Analysis results� for each group and it also lists the same information for the symptoms scores. In other words, it is treating the symptom scores for each group as additional independent groups even though these are within group variables. I�m looking for the correlation between the symptom score and connectivity within each group and whether it differs significantly across the groups. Where can I find/produce this information? I would deeply appreciate your help. Many thanks.
Best,
Diana.
> )
I followed your instructions, thank you very much. I am currently looking at the third option you listed - " 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". I have also selected 3 ROIs from the �seed/source� section. I�m new to Conn and not sure how/where to go to read the results. The table that pops out titled �connectivity values�, when all variables are selected, shows connectivity of each ROI to the one selected in the �Analysis results� for each group and it also lists the same information for the symptoms scores. In other words, it is treating the symptom scores for each group as additional independent groups even though these are within group variables. I�m looking for the correlation between the symptom score and connectivity within each group and whether it differs significantly across the groups. Where can I find/produce this information? I would deeply appreciate your help. Many thanks.
Best,
Diana.
> )
Aug 12, 2015 09:08 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
> (
Hi Alfonzo,
Thanks a lot for your help! I have attached three screen shots of different tables to get your input on each. In all the tables, the first three regressors are the three age-groups (age-group1: 3-7 years; age-group2: 8-12; age-group3: 13-20 years) and the last three regressors are scores on a clinical measure (social interpersonal relationships) for each group. I used the between-subjects contrast you specified previously [0 0 0 1 -1 0; 0 0 0 0 1 -1] for two of the tables and I used the any-effect contrast from the options listed under the Between-subjects contrast for one table. Could you please provide your interpretation of these tables. The table with the any-effect is labels as such and I�m curious to get your input on that result - not sure what this contrast mean within this context.
I also have a couple of other questions:
1) If we�re interests in connectivity within a brain network (e.g. default mode), could I select several ROIs corresponding to the regions making up that network and run the same contrasts?
2) could we make changes to these tables?
3) Are the error bars in the tables Standard Deviations or Standard Errors - all very large in our case?
Many thanks for your help.
Best,
Diana.
> )
Hi Alfonzo,
Thanks a lot for your help! I have attached three screen shots of different tables to get your input on each. In all the tables, the first three regressors are the three age-groups (age-group1: 3-7 years; age-group2: 8-12; age-group3: 13-20 years) and the last three regressors are scores on a clinical measure (social interpersonal relationships) for each group. I used the between-subjects contrast you specified previously [0 0 0 1 -1 0; 0 0 0 0 1 -1] for two of the tables and I used the any-effect contrast from the options listed under the Between-subjects contrast for one table. Could you please provide your interpretation of these tables. The table with the any-effect is labels as such and I�m curious to get your input on that result - not sure what this contrast mean within this context.
I also have a couple of other questions:
1) If we�re interests in connectivity within a brain network (e.g. default mode), could I select several ROIs corresponding to the regions making up that network and run the same contrasts?
2) could we make changes to these tables?
3) Are the error bars in the tables Standard Deviations or Standard Errors - all very large in our case?
Many thanks for your help.
Best,
Diana.
> )
Aug 13, 2015 02:08 AM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
Hi Diana,
For some reason the attachments do not seem to have made it into your post. Could you please re-send those?
Thanks
Alfonso
Originally posted by Diana Parvinchi:
For some reason the attachments do not seem to have made it into your post. Could you please re-send those?
Thanks
Alfonso
Originally posted by Diana Parvinchi:
> (
Hi Alfonzo,
Thanks a lot for your help! I have attached three screen shots of different tables to get your input on each. In all the tables, the first three regressors are the three age-groups (age-group1: 3-7 years; age-group2: 8-12; age-group3: 13-20 years) and the last three regressors are scores on a clinical measure (social interpersonal relationships) for each group. I used the between-subjects contrast you specified previously [0 0 0 1 -1 0; 0 0 0 0 1 -1] for two of the tables and I used the any-effect contrast from the options listed under the Between-subjects contrast for one table. Could you please provide your interpretation of these tables. The table with the any-effect is labels as such and I�m curious to get your input on that result - not sure what this contrast mean within this context.
I also have a couple of other questions:
1) If we�re interests in connectivity within a brain network (e.g. default mode), could I select several ROIs corresponding to the regions making up that network and run the same contrasts?
2) could we make changes to these tables?
3) Are the error bars in the tables Standard Deviations or Standard Errors - all very large in our case?
Many thanks for your help.
Best,
Diana.
> )
Hi Alfonzo,
Thanks a lot for your help! I have attached three screen shots of different tables to get your input on each. In all the tables, the first three regressors are the three age-groups (age-group1: 3-7 years; age-group2: 8-12; age-group3: 13-20 years) and the last three regressors are scores on a clinical measure (social interpersonal relationships) for each group. I used the between-subjects contrast you specified previously [0 0 0 1 -1 0; 0 0 0 0 1 -1] for two of the tables and I used the any-effect contrast from the options listed under the Between-subjects contrast for one table. Could you please provide your interpretation of these tables. The table with the any-effect is labels as such and I�m curious to get your input on that result - not sure what this contrast mean within this context.
I also have a couple of other questions:
1) If we�re interests in connectivity within a brain network (e.g. default mode), could I select several ROIs corresponding to the regions making up that network and run the same contrasts?
2) could we make changes to these tables?
3) Are the error bars in the tables Standard Deviations or Standard Errors - all very large in our case?
Many thanks for your help.
Best,
Diana.
> )
Aug 14, 2015 02:08 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Hi Alfonzo,
I'm sorry about that - not sure what happened. I have logged into the system and attaching the files from the website instead of e-mail. I hope this method will work. Thanks again for all your help!!!
I'm sorry about that - not sure what happened. I have logged into the system and attaching the files from the website instead of e-mail. I hope this method will work. Thanks again for all your help!!!
Aug 18, 2015 09:08 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Hi Alfonzo,
Thanks again for your help. I am wondering if you've had a chance to take a look at our attachments. I re-sent you three screen shots of different tables to get your input on each. In all the tables, the first three regressors are the three age-groups (age-group1: 3-7 years; age-group2: 8-12; age-group3: 13-20 years) and the last three regressors are scores on a clinical measure (social interpersonal relationships) for each group. I used the between-subjects contrast you specified previously [0 0 0 1 -1 0; 0 0 0 0 1 -1] for two of the tables and I used the any-effect contrast from the options listed under the Between-subjects contrast for one table. Could you please provide your interpretation of these tables? The table with the any-effect is labels as such and I'm very curious to get your input on that - not sure what this contrast means in this context.
I also have a couple of other questions:
1) If we're interested in connectivity within a network (e.g. default mode), could I select several ROIs corresponding to the regions making up that network and run the same contrasts?
2) could we make changes to the tables?
3) Are the error bars in the tables Standard Deviations or Standard Errors - all very large in our case?
Many many thanks for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
Thanks again for your help. I am wondering if you've had a chance to take a look at our attachments. I re-sent you three screen shots of different tables to get your input on each. In all the tables, the first three regressors are the three age-groups (age-group1: 3-7 years; age-group2: 8-12; age-group3: 13-20 years) and the last three regressors are scores on a clinical measure (social interpersonal relationships) for each group. I used the between-subjects contrast you specified previously [0 0 0 1 -1 0; 0 0 0 0 1 -1] for two of the tables and I used the any-effect contrast from the options listed under the Between-subjects contrast for one table. Could you please provide your interpretation of these tables? The table with the any-effect is labels as such and I'm very curious to get your input on that - not sure what this contrast means in this context.
I also have a couple of other questions:
1) If we're interested in connectivity within a network (e.g. default mode), could I select several ROIs corresponding to the regions making up that network and run the same contrasts?
2) could we make changes to the tables?
3) Are the error bars in the tables Standard Deviations or Standard Errors - all very large in our case?
Many many thanks for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
Hi Alfonzo,
I'm sorry about that - not sure what happened. I have logged into the system and attaching the files from the website instead of e-mail. I hope this method will work. Thanks again for all your help!!!
I'm sorry about that - not sure what happened. I have logged into the system and attaching the files from the website instead of e-mail. I hope this method will work. Thanks again for all your help!!!
Aug 19, 2015 01:08 AM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
Hi Diana,
Your previous post seems to contain only one attachment (named DifferencesContrasts_ASD1_ASD2_ASD3_Social_Interpersonal.png and containing a figure labeled "connectivity between SPL l and tolTG l"). The NITRC forum only allows you to attach a single file to a post, could you please either zip your multiple screenshots into a single compressed file and attach that one, or send me by mail the original screenshot files?
In the meantime, regarding your other questions:
1) yes, you may simply select all of the ROIs typically associated with that network and either enter a between-sources contrast "eye(N)" (where N is the number of ROIs; e.g. [1 0 0;0 1 0;0 0 1] for three ROIs), which will test your effect across any of the regions in the network instead of on a single region, or more typically enter a between-sources contrast "ones(1,N)/N" (e.g. [1/3 1/3 1/3] for three ROIs) which will test your effect when looking at the average connectivity with all of the regions in the network instead of a single region. note: the between-subject effects / contrasts would be just the same as before, you would only be modifying the selected sources and the associated between-sources contrast
2) I am not entirely clear if this is what you mean here, but you can right-click on any results table (or ctrl-click if on a Mac) and export that as a text file, as a .csv spreadsheet, or as a .mat Matlab file. If, instead, you mean modifying the barplot displays, you could type the following in Matlab command-line "set(gcf,'menubar','figure')" (without the double-quotes) and then use the "Edit" options there to modify these displays, or you could use the "import values" button on the main CONN gui to get the individual connectivity values for each subject if you prefer to analyze those separately and create your own displays, for example.
3) The error bars are 90% confidence intervals for the parameter estimates (those are typically larger than standard errors and smaller than standard deviations)
Hope this helps
Alfonso
Originally posted by Diana Parvinchi:
Your previous post seems to contain only one attachment (named DifferencesContrasts_ASD1_ASD2_ASD3_Social_Interpersonal.png and containing a figure labeled "connectivity between SPL l and tolTG l"). The NITRC forum only allows you to attach a single file to a post, could you please either zip your multiple screenshots into a single compressed file and attach that one, or send me by mail the original screenshot files?
In the meantime, regarding your other questions:
1) yes, you may simply select all of the ROIs typically associated with that network and either enter a between-sources contrast "eye(N)" (where N is the number of ROIs; e.g. [1 0 0;0 1 0;0 0 1] for three ROIs), which will test your effect across any of the regions in the network instead of on a single region, or more typically enter a between-sources contrast "ones(1,N)/N" (e.g. [1/3 1/3 1/3] for three ROIs) which will test your effect when looking at the average connectivity with all of the regions in the network instead of a single region. note: the between-subject effects / contrasts would be just the same as before, you would only be modifying the selected sources and the associated between-sources contrast
2) I am not entirely clear if this is what you mean here, but you can right-click on any results table (or ctrl-click if on a Mac) and export that as a text file, as a .csv spreadsheet, or as a .mat Matlab file. If, instead, you mean modifying the barplot displays, you could type the following in Matlab command-line "set(gcf,'menubar','figure')" (without the double-quotes) and then use the "Edit" options there to modify these displays, or you could use the "import values" button on the main CONN gui to get the individual connectivity values for each subject if you prefer to analyze those separately and create your own displays, for example.
3) The error bars are 90% confidence intervals for the parameter estimates (those are typically larger than standard errors and smaller than standard deviations)
Hope this helps
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonzo,
Thanks again for your help. I am wondering if you've had a chance to take a look at our attachments. I re-sent you three screen shots of different tables to get your input on each. In all the tables, the first three regressors are the three age-groups (age-group1: 3-7 years; age-group2: 8-12; age-group3: 13-20 years) and the last three regressors are scores on a clinical measure (social interpersonal relationships) for each group. I used the between-subjects contrast you specified previously [0 0 0 1 -1 0; 0 0 0 0 1 -1] for two of the tables and I used the any-effect contrast from the options listed under the Between-subjects contrast for one table. Could you please provide your interpretation of these tables? The table with the any-effect is labels as such and I'm very curious to get your input on that - not sure what this contrast means in this context.
I also have a couple of other questions:
1) If we're interested in connectivity within a network (e.g. default mode), could I select several ROIs corresponding to the regions making up that network and run the same contrasts?
2) could we make changes to the tables?
3) Are the error bars in the tables Standard Deviations or Standard Errors - all very large in our case?
Many many thanks for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
Thanks again for your help. I am wondering if you've had a chance to take a look at our attachments. I re-sent you three screen shots of different tables to get your input on each. In all the tables, the first three regressors are the three age-groups (age-group1: 3-7 years; age-group2: 8-12; age-group3: 13-20 years) and the last three regressors are scores on a clinical measure (social interpersonal relationships) for each group. I used the between-subjects contrast you specified previously [0 0 0 1 -1 0; 0 0 0 0 1 -1] for two of the tables and I used the any-effect contrast from the options listed under the Between-subjects contrast for one table. Could you please provide your interpretation of these tables? The table with the any-effect is labels as such and I'm very curious to get your input on that - not sure what this contrast means in this context.
I also have a couple of other questions:
1) If we're interested in connectivity within a network (e.g. default mode), could I select several ROIs corresponding to the regions making up that network and run the same contrasts?
2) could we make changes to the tables?
3) Are the error bars in the tables Standard Deviations or Standard Errors - all very large in our case?
Many many thanks for your help.
Best,
Diana.
Originally posted by Diana Parvinchi:
Hi Alfonzo,
I'm sorry about that - not sure what happened. I have logged into the system and attaching the files from the website instead of e-mail. I hope this method will work. Thanks again for all your help!!!
I'm sorry about that - not sure what happened. I have logged into the system and attaching the files from the website instead of e-mail. I hope this method will work. Thanks again for all your help!!!
Aug 31, 2015 02:08 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
> (Hi Alfonso,
Thank you very much for your detailed reply! I found it very helpful and appreciate it very much. I�m attaching the files via e-mail and hope that the attachments will reach you. Please let me know how you would interpret these two graphs. Many thanks.
> )
Thank you very much for your detailed reply! I found it very helpful and appreciate it very much. I�m attaching the files via e-mail and hope that the attachments will reach you. Please let me know how you would interpret these two graphs. Many thanks.
> )
Aug 31, 2015 04:08 PM | Diana Parvinchi - McMaster University
second-level analysis in Conn
> ( Hi Alfonso,
> As you suggested, I selected my groups and then scores in the �Between-subjects contrast�. Namely, I selected age-group1, age-group2, age-group3, age-group1-scores, age-group2-scores, age-group3-scores. then, I selected �any effect of interest and one ROI (Medial Prefrontal Cortex). I have attached a copy of one of the tables which shows all six regressors and connectivity between MPFC and Paracingulate Gyrus left. How would you interpret this graph? Many many thanks for your help!
Best,
Diana.
> )
> As you suggested, I selected my groups and then scores in the �Between-subjects contrast�. Namely, I selected age-group1, age-group2, age-group3, age-group1-scores, age-group2-scores, age-group3-scores. then, I selected �any effect of interest and one ROI (Medial Prefrontal Cortex). I have attached a copy of one of the tables which shows all six regressors and connectivity between MPFC and Paracingulate Gyrus left. How would you interpret this graph? Many many thanks for your help!
Best,
Diana.
> )
Aug 31, 2015 07:08 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Originally posted by Diana Parvinchi:
> ( Hi Alfonso,
> As you suggested, I selected my groups and then scores in the �Between-subjects contrast�. Namely, I selected age-group1, age-group2, age-group3, age-group1-scores, age-group2-scores, age-group3-scores. then, I selected �any effect of interest and one ROI (Medial Prefrontal Cortex). I have attached a copy of one of the tables which shows all six regressors and connectivity between MPFC and Paracingulate Gyrus left. How would you interpret this graph? Many many thanks for your help!
Best,
Diana.
> )
> As you suggested, I selected my groups and then scores in the �Between-subjects contrast�. Namely, I selected age-group1, age-group2, age-group3, age-group1-scores, age-group2-scores, age-group3-scores. then, I selected �any effect of interest and one ROI (Medial Prefrontal Cortex). I have attached a copy of one of the tables which shows all six regressors and connectivity between MPFC and Paracingulate Gyrus left. How would you interpret this graph? Many many thanks for your help!
Best,
Diana.
> )
Sep 3, 2015 01:09 AM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
Hi Diana,
From the anyEffect_Groups_socialCope.png figure, and for the selected source/target ROIs displayed there, there does not appear to be any significant association between functional connectivity strength and behavioral scores within any of your three subject groups (this is indicated by the last three effects error-bars crossing the y=0 axis). You may test whether that is the case using the between-subjects contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] (this tests whether there is any significant association between connectivity and behavioral scores in any of the three groups). If you want to focus on those regions that show significant associations with behavioral scores in any of the three groups (the main effect of the behavioral score factor) use the contrast above to test specifically for these effects. If you want to focus on those regions that show significantly different associations with behavioral scores between your three groups (the interaction between group and behavioral scores) use instead the contrast [0 0 0 1 -1 0;0 0 0 0 1 -1].
Hope this helps
Alfonso
Originally posted by Diana Parvinchi:
From the anyEffect_Groups_socialCope.png figure, and for the selected source/target ROIs displayed there, there does not appear to be any significant association between functional connectivity strength and behavioral scores within any of your three subject groups (this is indicated by the last three effects error-bars crossing the y=0 axis). You may test whether that is the case using the between-subjects contrast [0 0 0 1 0 0;0 0 0 0 1 0; 0 0 0 0 0 1] (this tests whether there is any significant association between connectivity and behavioral scores in any of the three groups). If you want to focus on those regions that show significant associations with behavioral scores in any of the three groups (the main effect of the behavioral score factor) use the contrast above to test specifically for these effects. If you want to focus on those regions that show significantly different associations with behavioral scores between your three groups (the interaction between group and behavioral scores) use instead the contrast [0 0 0 1 -1 0;0 0 0 0 1 -1].
Hope this helps
Alfonso
Originally posted by Diana Parvinchi:
Originally posted by Diana Parvinchi:
> ( Hi Alfonso,
> As you suggested, I selected my groups and then scores in the �Between-subjects contrast�. Namely, I selected age-group1, age-group2, age-group3, age-group1-scores, age-group2-scores, age-group3-scores. then, I selected �any effect of interest and one ROI (Medial Prefrontal Cortex). I have attached a copy of one of the tables which shows all six regressors and connectivity between MPFC and Paracingulate Gyrus left. How would you interpret this graph? Many many thanks for your help!
Best,
Diana.
> )
> As you suggested, I selected my groups and then scores in the �Between-subjects contrast�. Namely, I selected age-group1, age-group2, age-group3, age-group1-scores, age-group2-scores, age-group3-scores. then, I selected �any effect of interest and one ROI (Medial Prefrontal Cortex). I have attached a copy of one of the tables which shows all six regressors and connectivity between MPFC and Paracingulate Gyrus left. How would you interpret this graph? Many many thanks for your help!
Best,
Diana.
> )
Sep 4, 2015 06:09 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Hi Alfonso,
Many thanks for your reply - very helpful! I have attached several graphs and would love to hear your interpretation of these results. We are starting to understand how to read these graphs. I also have two questions. 1) how can we determine where an effect is coming from? For example, the last three regressors are the scores of the three groups on a specific measure. We can read the direction of association between symptoms severity and connectivity via these three regressor. Based on your previous response, the bars with the error bars not crossing the y=0 axis are the significant associations. Using this observation, If we observe one significant effect, among the three, and would like to report this finding in a manuscript for publication purposes, where should we look for the statistical info (e.g. p-values)? 2) Also, how can we modify these graphs (e.g. the x-axis labels)?
Best,
Diana.
Many thanks for your reply - very helpful! I have attached several graphs and would love to hear your interpretation of these results. We are starting to understand how to read these graphs. I also have two questions. 1) how can we determine where an effect is coming from? For example, the last three regressors are the scores of the three groups on a specific measure. We can read the direction of association between symptoms severity and connectivity via these three regressor. Based on your previous response, the bars with the error bars not crossing the y=0 axis are the significant associations. Using this observation, If we observe one significant effect, among the three, and would like to report this finding in a manuscript for publication purposes, where should we look for the statistical info (e.g. p-values)? 2) Also, how can we modify these graphs (e.g. the x-axis labels)?
Best,
Diana.
Sep 10, 2015 05:09 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Hi Alfonso,
This is just a friendly reminder. We would deeply appreciate a reply to our previous message. Many many thanks for your continued support!
Best,
Diana.
Originally posted by Diana Parvinchi:
This is just a friendly reminder. We would deeply appreciate a reply to our previous message. Many many thanks for your continued support!
Best,
Diana.
Originally posted by Diana Parvinchi:
Hi Alfonso,
Many thanks for your reply - very helpful! I have attached several graphs and would love to hear your interpretation of these results. We are starting to understand how to read these graphs. I also have two questions. 1) how can we determine where an effect is coming from? For example, the last three regressors are the scores of the three groups on a specific measure. We can read the direction of association between symptoms severity and connectivity via these three regressor. Based on your previous response, the bars with the error bars not crossing the y=0 axis are the significant associations. Using this observation, If we observe one significant effect, among the three, and would like to report this finding in a manuscript for publication purposes, where should we look for the statistical info (e.g. p-values)? 2) Also, how can we modify these graphs (e.g. the x-axis labels)?
Best,
Diana.
Many thanks for your reply - very helpful! I have attached several graphs and would love to hear your interpretation of these results. We are starting to understand how to read these graphs. I also have two questions. 1) how can we determine where an effect is coming from? For example, the last three regressors are the scores of the three groups on a specific measure. We can read the direction of association between symptoms severity and connectivity via these three regressor. Based on your previous response, the bars with the error bars not crossing the y=0 axis are the significant associations. Using this observation, If we observe one significant effect, among the three, and would like to report this finding in a manuscript for publication purposes, where should we look for the statistical info (e.g. p-values)? 2) Also, how can we modify these graphs (e.g. the x-axis labels)?
Best,
Diana.
Sep 24, 2015 06:09 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Hi Alfonso,
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.
Sep 24, 2015 06:09 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
*my interpretation (not interoperation) - autocorrect does
not help sometimes. My apologies!
Originally posted by Diana Parvinchi:
Originally posted by Diana Parvinchi:
Hi Alfonso,
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.
Sep 25, 2015 02:09 PM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
Hi Diana,
Yes, your interpretation is perfectly correct. Connectivity between right Intracalcarine and right FO cortex appears negatively correlated with symptom scores (i.e. higher connectivity between these regions in children with lower symptom scores). In addition there also appear to be an interaction with age, with stronger associations with symptom scores (and perhaps also higher average connectivity as well, see point below) in older children (age group 3) compared to younger children.
There is only one subtlety here that affects the interpretation of the first three bars. Could you please let me know whether the symptom-score covariates are centered or not? If they are centered, did you center them individually within each age-group or globally across all subjects?; and if they are not centered, is a symptom-score equal to zero meaningul/interpretable? This affects the interpretation of the sign and effect-size of the first three bars in your plot -in other words, whether on average the connectivity between right ICC and right FO is positive or negative/anticorrelated-, which in turn affects the correct interpretation of the associations with symptom scores -e.g. higher connectivity in children with lower symptoms may reflect stronger/more-positive connectivity in these children or weaker/less-negative conectivity in these children-. Generally the effect-sizes in your first three bars represent the average level of connectivity within each age-group at the zero level of your symptom score covariate. If the symptom score covariates are centered globally (subtracted the average across all of your subjects) then that zero-level represents the average symptom-level across all of your subjects (and the same level across the three age-groups), if they are centered separately (subtracted the average within each age-group separately) then that zero-level represents the average symptom-level within each age-group separately, and if they are not centered (they represent "raw" symptom scores) then that zero-level represents an actual zero-value of your original symptom variables (again the same level across the three age-groups).
Thanks
Alfonso
Originally posted by Diana Parvinchi:
Yes, your interpretation is perfectly correct. Connectivity between right Intracalcarine and right FO cortex appears negatively correlated with symptom scores (i.e. higher connectivity between these regions in children with lower symptom scores). In addition there also appear to be an interaction with age, with stronger associations with symptom scores (and perhaps also higher average connectivity as well, see point below) in older children (age group 3) compared to younger children.
There is only one subtlety here that affects the interpretation of the first three bars. Could you please let me know whether the symptom-score covariates are centered or not? If they are centered, did you center them individually within each age-group or globally across all subjects?; and if they are not centered, is a symptom-score equal to zero meaningul/interpretable? This affects the interpretation of the sign and effect-size of the first three bars in your plot -in other words, whether on average the connectivity between right ICC and right FO is positive or negative/anticorrelated-, which in turn affects the correct interpretation of the associations with symptom scores -e.g. higher connectivity in children with lower symptoms may reflect stronger/more-positive connectivity in these children or weaker/less-negative conectivity in these children-. Generally the effect-sizes in your first three bars represent the average level of connectivity within each age-group at the zero level of your symptom score covariate. If the symptom score covariates are centered globally (subtracted the average across all of your subjects) then that zero-level represents the average symptom-level across all of your subjects (and the same level across the three age-groups), if they are centered separately (subtracted the average within each age-group separately) then that zero-level represents the average symptom-level within each age-group separately, and if they are not centered (they represent "raw" symptom scores) then that zero-level represents an actual zero-value of your original symptom variables (again the same level across the three age-groups).
Thanks
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.
Sep 29, 2015 07:09 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Hi Alfonso,
Thank you very much for your reply, as always, we found it very helpful. To answer your question, a repetitive score of "0" here would mean an absence of this symptom - the higher the score the greater the symptom severity. Given this point, I think you're confirming that we are correct in interpreting this finding as showing a negative correlation - as connectivity between these two ROI's increases, repetitive symptom severity decreases. My question is if we want to report this finding (and other similar ones), where can I get the statistical information? Should we take the information from the "Analysis results"? In other words, what is equivelent to running pair-wise comparisons to identify where a significant effect is coming from? In this case, we can see that this effect is significant only in the third age-group from the graph, but what is the p-value corresponding to it? Thanks again for your help.
Cheers,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Thank you very much for your reply, as always, we found it very helpful. To answer your question, a repetitive score of "0" here would mean an absence of this symptom - the higher the score the greater the symptom severity. Given this point, I think you're confirming that we are correct in interpreting this finding as showing a negative correlation - as connectivity between these two ROI's increases, repetitive symptom severity decreases. My question is if we want to report this finding (and other similar ones), where can I get the statistical information? Should we take the information from the "Analysis results"? In other words, what is equivelent to running pair-wise comparisons to identify where a significant effect is coming from? In this case, we can see that this effect is significant only in the third age-group from the graph, but what is the p-value corresponding to it? Thanks again for your help.
Cheers,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
Yes, your interpretation is perfectly correct. Connectivity between right Intracalcarine and right FO cortex appears negatively correlated with symptom scores (i.e. higher connectivity between these regions in children with lower symptom scores). In addition there also appear to be an interaction with age, with stronger associations with symptom scores (and perhaps also higher average connectivity as well, see point below) in older children (age group 3) compared to younger children.
There is only one subtlety here that affects the interpretation of the first three bars. Could you please let me know whether the symptom-score covariates are centered or not? If they are centered, did you center them individually within each age-group or globally across all subjects?; and if they are not centered, is a symptom-score equal to zero meaningul/interpretable? This affects the interpretation of the sign and effect-size of the first three bars in your plot -in other words, whether on average the connectivity between right ICC and right FO is positive or negative/anticorrelated-, which in turn affects the correct interpretation of the associations with symptom scores -e.g. higher connectivity in children with lower symptoms may reflect stronger/more-positive connectivity in these children or weaker/less-negative conectivity in these children-. Generally the effect-sizes in your first three bars represent the average level of connectivity within each age-group at the zero level of your symptom score covariate. If the symptom score covariates are centered globally (subtracted the average across all of your subjects) then that zero-level represents the average symptom-level across all of your subjects (and the same level across the three age-groups), if they are centered separately (subtracted the average within each age-group separately) then that zero-level represents the average symptom-level within each age-group separately, and if they are not centered (they represent "raw" symptom scores) then that zero-level represents an actual zero-value of your original symptom variables (again the same level across the three age-groups).
Thanks
Alfonso
Originally posted by Diana Parvinchi:
Yes, your interpretation is perfectly correct. Connectivity between right Intracalcarine and right FO cortex appears negatively correlated with symptom scores (i.e. higher connectivity between these regions in children with lower symptom scores). In addition there also appear to be an interaction with age, with stronger associations with symptom scores (and perhaps also higher average connectivity as well, see point below) in older children (age group 3) compared to younger children.
There is only one subtlety here that affects the interpretation of the first three bars. Could you please let me know whether the symptom-score covariates are centered or not? If they are centered, did you center them individually within each age-group or globally across all subjects?; and if they are not centered, is a symptom-score equal to zero meaningul/interpretable? This affects the interpretation of the sign and effect-size of the first three bars in your plot -in other words, whether on average the connectivity between right ICC and right FO is positive or negative/anticorrelated-, which in turn affects the correct interpretation of the associations with symptom scores -e.g. higher connectivity in children with lower symptoms may reflect stronger/more-positive connectivity in these children or weaker/less-negative conectivity in these children-. Generally the effect-sizes in your first three bars represent the average level of connectivity within each age-group at the zero level of your symptom score covariate. If the symptom score covariates are centered globally (subtracted the average across all of your subjects) then that zero-level represents the average symptom-level across all of your subjects (and the same level across the three age-groups), if they are centered separately (subtracted the average within each age-group separately) then that zero-level represents the average symptom-level within each age-group separately, and if they are not centered (they represent "raw" symptom scores) then that zero-level represents an actual zero-value of your original symptom variables (again the same level across the three age-groups).
Thanks
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.
Oct 1, 2015 05:10 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
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?
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?
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?
Many thanks again,
Diana.
Originally posted by Diana Parvinchi:
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?
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?
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?
Many thanks again,
Diana.
Originally posted by Diana Parvinchi:
Hi Alfonso,
Thank you very much for your reply, as always, we found it very helpful. To answer your question, a repetitive score of "0" here would mean an absence of this symptom - the higher the score the greater the symptom severity. Given this point, I think you're confirming that we are correct in interpreting this finding as showing a negative correlation - as connectivity between these two ROI's increases, repetitive symptom severity decreases. My question is if we want to report this finding (and other similar ones), where can I get the statistical information? Should we take the information from the "Analysis results"? In other words, what is equivelent to running pair-wise comparisons to identify where a significant effect is coming from? In this case, we can see that this effect is significant only in the third age-group from the graph, but what is the p-value corresponding to it? Thanks again for your help.
Cheers,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Thank you very much for your reply, as always, we found it very helpful. To answer your question, a repetitive score of "0" here would mean an absence of this symptom - the higher the score the greater the symptom severity. Given this point, I think you're confirming that we are correct in interpreting this finding as showing a negative correlation - as connectivity between these two ROI's increases, repetitive symptom severity decreases. My question is if we want to report this finding (and other similar ones), where can I get the statistical information? Should we take the information from the "Analysis results"? In other words, what is equivelent to running pair-wise comparisons to identify where a significant effect is coming from? In this case, we can see that this effect is significant only in the third age-group from the graph, but what is the p-value corresponding to it? Thanks again for your help.
Cheers,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
Yes, your interpretation is perfectly correct. Connectivity between right Intracalcarine and right FO cortex appears negatively correlated with symptom scores (i.e. higher connectivity between these regions in children with lower symptom scores). In addition there also appear to be an interaction with age, with stronger associations with symptom scores (and perhaps also higher average connectivity as well, see point below) in older children (age group 3) compared to younger children.
There is only one subtlety here that affects the interpretation of the first three bars. Could you please let me know whether the symptom-score covariates are centered or not? If they are centered, did you center them individually within each age-group or globally across all subjects?; and if they are not centered, is a symptom-score equal to zero meaningul/interpretable? This affects the interpretation of the sign and effect-size of the first three bars in your plot -in other words, whether on average the connectivity between right ICC and right FO is positive or negative/anticorrelated-, which in turn affects the correct interpretation of the associations with symptom scores -e.g. higher connectivity in children with lower symptoms may reflect stronger/more-positive connectivity in these children or weaker/less-negative conectivity in these children-. Generally the effect-sizes in your first three bars represent the average level of connectivity within each age-group at the zero level of your symptom score covariate. If the symptom score covariates are centered globally (subtracted the average across all of your subjects) then that zero-level represents the average symptom-level across all of your subjects (and the same level across the three age-groups), if they are centered separately (subtracted the average within each age-group separately) then that zero-level represents the average symptom-level within each age-group separately, and if they are not centered (they represent "raw" symptom scores) then that zero-level represents an actual zero-value of your original symptom variables (again the same level across the three age-groups).
Thanks
Alfonso
Originally posted by Diana Parvinchi:
Yes, your interpretation is perfectly correct. Connectivity between right Intracalcarine and right FO cortex appears negatively correlated with symptom scores (i.e. higher connectivity between these regions in children with lower symptom scores). In addition there also appear to be an interaction with age, with stronger associations with symptom scores (and perhaps also higher average connectivity as well, see point below) in older children (age group 3) compared to younger children.
There is only one subtlety here that affects the interpretation of the first three bars. Could you please let me know whether the symptom-score covariates are centered or not? If they are centered, did you center them individually within each age-group or globally across all subjects?; and if they are not centered, is a symptom-score equal to zero meaningul/interpretable? This affects the interpretation of the sign and effect-size of the first three bars in your plot -in other words, whether on average the connectivity between right ICC and right FO is positive or negative/anticorrelated-, which in turn affects the correct interpretation of the associations with symptom scores -e.g. higher connectivity in children with lower symptoms may reflect stronger/more-positive connectivity in these children or weaker/less-negative conectivity in these children-. Generally the effect-sizes in your first three bars represent the average level of connectivity within each age-group at the zero level of your symptom score covariate. If the symptom score covariates are centered globally (subtracted the average across all of your subjects) then that zero-level represents the average symptom-level across all of your subjects (and the same level across the three age-groups), if they are centered separately (subtracted the average within each age-group separately) then that zero-level represents the average symptom-level within each age-group separately, and if they are not centered (they represent "raw" symptom scores) then that zero-level represents an actual zero-value of your original symptom variables (again the same level across the three age-groups).
Thanks
Alfonso
Originally posted by Diana Parvinchi:
Hi Alfonso,
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.
Could I please get your input on a graph that I have attached. Just to remind you of our analysis, we are looking at the correlation between functional connectivity and symptoms severity in children with autism spectrum disorder (ASD). We have three age-groups of cohorts with ASD. I have attached a screen shot of a graph showing 6 regressors. The first 3 are the age-groups and the last three are the symptom scores within each group. My interoperation of this graph is that functional connectivity between right Intracalcarine and right Frontal Orbital Cortex is negatively correlated with severity of Repetitive Behaviour. Meaning, as connectivity between these two regions increases, repetitive behaviour decreases. Is this correct? Also, if we'd like to report this finding in a manuscript, how could we provide details of this effect (p-value, beta which are specific to this significant effect)? Many many thanks for your help!
Best,
Diana.
Oct 8, 2015 01:10 PM | Alfonso Nieto-Castanon - Boston University
RE: second-level analysis in Conn
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 8, 2015 05:10 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
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 15, 2015 04:10 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
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