[color=#000000]Hi Marissa,[/color]
[color=#000000]Typically, in models that use functional connectivity as the dependent measure (y), an interaction between two continuous measures (e.g. BehavioralMeasureA and BehavioralMeasureB) can be probed directly using a model of the form:[/color]
[i][color=#000000]y ~ AllSubjects + BehavioralMeasureA + BehavioralMeasureB + BehavioralMeasureA*BehavioralMeasureB[/color][/i]
[color=#000000]with a between-subjects contrast [0 0 0 1]. A significant interaction in these analyses can be interpreted as responding to the research question:[/color]
a) "does the correlation/association between BehavioralMeasureA and functional connectivity depend on the level of BehavioralMeasureB?"
or, equivalently:
b) "does the correlation/association between BehavioralMeasureB and functional connectivity depend on the level of BehavioralMeasureA?"
Some times researchers find it simpler/better to first discretize one of the variables (e.g. create two groups: B_High and B_Low) and then test the model:
[i]y ~ B_High + B_Low + B_High*BehavioralMeasureA + B_Low*BehavioralMeasureA[/i]
with a between-subjects contrast [0 0 -1 1], or equivalently a model of the form:
[i]y ~ AllSubjects + B_High + BehavioralMeasureA + B_High*BehavioralMeasureA[/i]
with a between-subjects contrast [0 0 0 1]
which (both are identical) may be interpreted as responding to the similar research question:
a) "does the correlation/association between BehavioralMeasureA and functional connectivity differ between B_Low and B_High subjects?"
One of the advantages of the latter approach is that the results can be displayed by plotting, for example, the correlation between functional connectivity values and BehavioralMeasureA within each individual significant cluster, separately for B_High and B_Low subjects. Looking at the range of connectivity values in these plots often helps properly interpret the results
Let me know if this helps clarify
Best
Alfonso
ps. under development for the next release is a new way to display the above form of "descriptive explanations" for many choices of second-level analyses in CONN's [i]results [/i]tab (as well as the automatic generation of a set of suggested second-level analyses and contrasts based on the 2nd-level covariates defined in your project). If you want to try that out, the "development release" can be downloaded from www.conn-toolbox.org/resources/source (feedback/suggestions are welcome!)
[i]Originally posted by marissa_g:[/i][quote]Hello,
To be more specific about this question following the contrasts I ran below to probe a significant moderation in seed-to-voxel analysis, what does it mean theoretically for example to find significant positive functional connectivity in seed-to-voxel analysis at High levels of BehavioralMeasureA, but no significant functional connectivity at Low levels of BehavioralMeasureA or at High and Low levels of BehavioralMeasureB? I am trying to ensure that I am interpreting seed-to-voxel moderation probes appropriately since the dependent variable is significant functional connectivity in different regions and not necessarily a number value (i.e., making it difficult to plot the moderation probes). Is it appropriate to say that the moderation between BehavioralMeasureA and BehavioralMeasureB is contingent on high levels of BehavioralMeasureA? Is there a way to display this graphically?
Thanks,
Marissa[/quote]
[color=#000000]Typically, in models that use functional connectivity as the dependent measure (y), an interaction between two continuous measures (e.g. BehavioralMeasureA and BehavioralMeasureB) can be probed directly using a model of the form:[/color]
[i][color=#000000]y ~ AllSubjects + BehavioralMeasureA + BehavioralMeasureB + BehavioralMeasureA*BehavioralMeasureB[/color][/i]
[color=#000000]with a between-subjects contrast [0 0 0 1]. A significant interaction in these analyses can be interpreted as responding to the research question:[/color]
a) "does the correlation/association between BehavioralMeasureA and functional connectivity depend on the level of BehavioralMeasureB?"
or, equivalently:
b) "does the correlation/association between BehavioralMeasureB and functional connectivity depend on the level of BehavioralMeasureA?"
Some times researchers find it simpler/better to first discretize one of the variables (e.g. create two groups: B_High and B_Low) and then test the model:
[i]y ~ B_High + B_Low + B_High*BehavioralMeasureA + B_Low*BehavioralMeasureA[/i]
with a between-subjects contrast [0 0 -1 1], or equivalently a model of the form:
[i]y ~ AllSubjects + B_High + BehavioralMeasureA + B_High*BehavioralMeasureA[/i]
with a between-subjects contrast [0 0 0 1]
which (both are identical) may be interpreted as responding to the similar research question:
a) "does the correlation/association between BehavioralMeasureA and functional connectivity differ between B_Low and B_High subjects?"
One of the advantages of the latter approach is that the results can be displayed by plotting, for example, the correlation between functional connectivity values and BehavioralMeasureA within each individual significant cluster, separately for B_High and B_Low subjects. Looking at the range of connectivity values in these plots often helps properly interpret the results
Let me know if this helps clarify
Best
Alfonso
ps. under development for the next release is a new way to display the above form of "descriptive explanations" for many choices of second-level analyses in CONN's [i]results [/i]tab (as well as the automatic generation of a set of suggested second-level analyses and contrasts based on the 2nd-level covariates defined in your project). If you want to try that out, the "development release" can be downloaded from www.conn-toolbox.org/resources/source (feedback/suggestions are welcome!)
[i]Originally posted by marissa_g:[/i][quote]Hello,
To be more specific about this question following the contrasts I ran below to probe a significant moderation in seed-to-voxel analysis, what does it mean theoretically for example to find significant positive functional connectivity in seed-to-voxel analysis at High levels of BehavioralMeasureA, but no significant functional connectivity at Low levels of BehavioralMeasureA or at High and Low levels of BehavioralMeasureB? I am trying to ensure that I am interpreting seed-to-voxel moderation probes appropriately since the dependent variable is significant functional connectivity in different regions and not necessarily a number value (i.e., making it difficult to plot the moderation probes). Is it appropriate to say that the moderation between BehavioralMeasureA and BehavioralMeasureB is contingent on high levels of BehavioralMeasureA? Is there a way to display this graphically?
Thanks,
Marissa[/quote]