help
help > RE: second-level analysis in Conn
Apr 18, 2016 04:04 PM | Diana Parvinchi - McMaster University
RE: second-level analysis in Conn
Hi Alfonso,
We have analyzed our data in Conn and have interesting results. I am following up with you on one particular effect. We originally examined the correlation between FC and symptom severities in three age groups of cohorts with autism spectrum disorder (ASD). One particular effect is showing that FC in superior frontal gyrus (with fusiform gyrus) is positively correlated with social impairment severities in our age group2. This means that as FC between these two regions increases, social impairments in this group increase as well. I would like to determine if the FC between these two regions is an inhibitory or an excitatory connection - negatively or positively correlated. At this point, the direction of the correlation has to do with FC and symptom severity. How do I determine if the FC itself between these two regions is positive or negative?
Many thanks for your help,
Diana.
Originally posted by Alfonso Nieto-Castanon:
We have analyzed our data in Conn and have interesting results. I am following up with you on one particular effect. We originally examined the correlation between FC and symptom severities in three age groups of cohorts with autism spectrum disorder (ASD). One particular effect is showing that FC in superior frontal gyrus (with fusiform gyrus) is positively correlated with social impairment severities in our age group2. This means that as FC between these two regions increases, social impairments in this group increase as well. I would like to determine if the FC between these two regions is an inhibitory or an excitatory connection - negatively or positively correlated. At this point, the direction of the correlation has to do with FC and symptom severity. How do I determine if the FC itself between these two regions is positive or negative?
Many thanks for your help,
Diana.
Originally posted by Alfonso Nieto-Castanon:
Hi
Diana,
You may compute the association between symptom scores and connectivity within each of your three subject groups using the following procedure:
1) create three second-level covariates (Group1, Group2, and Group3) indicating your three subject groups (e.g. Group1 contains 1's for the subject in the first group and 0's for everyone else)
2) create three second-level covariates indicating the symptom scores within each of your three groups (e.g. Scores1 will contain the symptom scores for subjects in Group1 and 0's for everyone else; if you already have a second-level covariate named 'scores' containing all of the subjects' scores you may enter in the 'values' field of the new 'Scores1' covariate "Group1.*scores" -without the quotes- to create this new 'Scores1' covariate)
3) in the second-level analyses enter any of the following:
a) select 'AllSubjects' and 'scores' in the between-subject effects list and enter a between-subjects contrast [0 1] to look at the association between symptom scores and connectivity across all of your subjects (jointly across the three groups, disregarding group info)
b) select 'Group1', 'Group2', 'Group3', 'scores', and enter a contrast [0 0 0 1] to look at the association between symptom scores and connectivity across all of your subjects (jointly across the three groups) after discounting potential differences in average connectivity between your groups
c) select 'Group1' and 'Scores1' and enter a contrast [0 1] to look at the association between symptom scores and connectivity within your first group only
d) select 'Group1', 'Group2', 'Scores1', 'Scores', and enter a contrast [0 0 -1 1] to look at the difference between Group1 and Group2 in their association between symptom scores and connectivity
e) select 'Group1', 'Group2', 'Group3', 'Scores1', 'Scores2', 'Scores3', and enter a contrast [0 0 0 1 -1 0; 0 0 0 0 1 -1] to look at any differences between your groups in their association between symptom scores and connectivity
Regarding your second question, the beta values reported in the results table are the effect sizes of the chosen contrast. Mathematically if B is the matrix of regression coefficients of your second-level model, C is your between-subjects contrast, and D is your between-conditions contrast, then the effect sizes are C*B*D' (note: if any of your between-subjects or between-conditions contrast is a matrix instead of a vector then C*B*D' is a vector and the results table reports the norm of this vector). The interpretation of these betas depends on the chosen model and contrasts. For example, in the example (a) above, the effect size of the chosen contrast represents the association / regression-coefficient between symptom scores and functional connectivity, and its units are "increases in functional connectivity -fisher transformed correlation coefficients- associated with each unit increase in symptom scores". When looking at main connectivity effects (e.g. if you select 'AllSubjects' and enter a [1] contrast) then the reported effect sizes represent average functional connectivity -fisher transformed coefficients- among the tested subjects). Let me know if you would like me to further clarify any of the above.
Hope this helps
Alfonso
Originally posted by Diana P:
You may compute the association between symptom scores and connectivity within each of your three subject groups using the following procedure:
1) create three second-level covariates (Group1, Group2, and Group3) indicating your three subject groups (e.g. Group1 contains 1's for the subject in the first group and 0's for everyone else)
2) create three second-level covariates indicating the symptom scores within each of your three groups (e.g. Scores1 will contain the symptom scores for subjects in Group1 and 0's for everyone else; if you already have a second-level covariate named 'scores' containing all of the subjects' scores you may enter in the 'values' field of the new 'Scores1' covariate "Group1.*scores" -without the quotes- to create this new 'Scores1' covariate)
3) in the second-level analyses enter any of the following:
a) select 'AllSubjects' and 'scores' in the between-subject effects list and enter a between-subjects contrast [0 1] to look at the association between symptom scores and connectivity across all of your subjects (jointly across the three groups, disregarding group info)
b) select 'Group1', 'Group2', 'Group3', 'scores', and enter a contrast [0 0 0 1] to look at the association between symptom scores and connectivity across all of your subjects (jointly across the three groups) after discounting potential differences in average connectivity between your groups
c) select 'Group1' and 'Scores1' and enter a contrast [0 1] to look at the association between symptom scores and connectivity within your first group only
d) select 'Group1', 'Group2', 'Scores1', 'Scores', and enter a contrast [0 0 -1 1] to look at the difference between Group1 and Group2 in their association between symptom scores and connectivity
e) select 'Group1', 'Group2', 'Group3', 'Scores1', 'Scores2', 'Scores3', and enter a contrast [0 0 0 1 -1 0; 0 0 0 0 1 -1] to look at any differences between your groups in their association between symptom scores and connectivity
Regarding your second question, the beta values reported in the results table are the effect sizes of the chosen contrast. Mathematically if B is the matrix of regression coefficients of your second-level model, C is your between-subjects contrast, and D is your between-conditions contrast, then the effect sizes are C*B*D' (note: if any of your between-subjects or between-conditions contrast is a matrix instead of a vector then C*B*D' is a vector and the results table reports the norm of this vector). The interpretation of these betas depends on the chosen model and contrasts. For example, in the example (a) above, the effect size of the chosen contrast represents the association / regression-coefficient between symptom scores and functional connectivity, and its units are "increases in functional connectivity -fisher transformed correlation coefficients- associated with each unit increase in symptom scores". When looking at main connectivity effects (e.g. if you select 'AllSubjects' and enter a [1] contrast) then the reported effect sizes represent average functional connectivity -fisher transformed coefficients- among the tested subjects). Let me know if you would like me to further clarify any of the above.
Hope this helps
Alfonso
Originally posted by Diana P:
Hi,
I have two questions about Conn's second-level analysis and would appreciate your help. We are interested in examining the association between core symptoms of autism spectrum disorder (ASD) (repetitive behaviour, socialization, etc.) and connectivity in three age groups (3-7, 8-12, 13-20 years) of individuals diagnosed with ASD. How do I specify the contrasts in the second-level analysis to examine this question? My second question is what does "beta" in the Analysis results represent in the second level analysis? For example, if I have three groups selected in "between-subjects contrasts" and two Seeds selected in the "Between-sources contrasts", what does "beta" in the Analysis results represent? Many thanks.
I have two questions about Conn's second-level analysis and would appreciate your help. We are interested in examining the association between core symptoms of autism spectrum disorder (ASD) (repetitive behaviour, socialization, etc.) and connectivity in three age groups (3-7, 8-12, 13-20 years) of individuals diagnosed with ASD. How do I specify the contrasts in the second-level analysis to examine this question? My second question is what does "beta" in the Analysis results represent in the second level analysis? For example, if I have three groups selected in "between-subjects contrasts" and two Seeds selected in the "Between-sources contrasts", what does "beta" in the Analysis results represent? Many thanks.