help
help > RE: second-level analysis in Conn
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.
> )