Hi Annie,

Some thoughts on your questions below
Best
Alfonso
Originally posted by Annie Möller:
The "eye(5)" (i.e. [1 0 0 0 0; 0 1 0 0 0; 0 0 1 0 0; 0 0 0 1 0; 0 0 0 0 1]) contrast performs an OR conjunction across the 5 individual components (i.e. looking at significant effects within any of the selected components). Because looking at the significant effects for each of the 5 individual components separately (e.g. each at a p<.05 level) can lead to a multiple-comparison increase in the chance of false positives, the "eye(5)" contrast is inherently more conservative than each individual contrast, but it will be more sensitive than, for example, performing the 5 individual tests (one for each component) using an explicit bonferroni correction to deal with the multiple-comparison problem. Typically you would want to define a priori how many components you are going to test. The trade-off in this descission is that: a) including too few components might miss some effect occurring in the higher-order components; and b) including too many components might decrease your analysis sensitivity/power beyond reasonable levels (so you will not have a high likelihood of finding an effect even if it is present). The typical rule-of-thumb is to select between N/5 and N/10 components (where N is the number of subjects in your study), but there is nothing absolute about this choice since the actual optimum in the trade-off above depends on many factors, and not just the number of subjects by number of components ratio. CONN will extract by default N/5 components but it is up to the user to determine how many components to extract, and mong those, how many to include in your second-level test.  

Yes, if you perform ROI-to-ROI analyses, define your between-subjects and between-condition test, and click on 'results explorer', you can then restrict the analyses to all of the atlas ROIs, for example, and then click on 'select all sources' to look at any effects across the entire ROI-to-ROI connectivity matrix. While these are a great deal of individual tests (one for each pair of ROIs), you can use for example NBS (network based statistics) to deal with the associated multiple comparisons problem while still providing reasonable sensitivity to find any effects across the entire ROI-to-ROI connectivity matrix.

Typically you would want to enter each separate cluster as an individual seed. This will be handled automatically by CONN when you use the *.ROIs.img file generated by "export mask" in the results explorer (that single file contains all of your individual clusters as separate ROIs, so when you enter that file in CONN it will interpret it as an "atlas" file containing multiple ROIs)

In the "seed-to-voxel" or "voxel-to-voxel" results explorer, when you click on the "display values" or "import values" buttons that will give you by default the individual contrast effect-sizes of your second-level model. That means that, for a three-group [-1 1 0; 0 -1 1] between-subjects contrast, and a [-1 1] between-conditions contrast, you will be getting, for each cluster, two values, each representing a row of your between-subjects F contrast (the first value represents the difference between groupB and groupA in the difference between conditionB and conditionA connectivity, and the second value represents the difference between groupC and groupB in the difference betwen conditionB and conditionA connectivity). That is likely hard to interpret, so the recommendation when you want to instead look at the individual group and condition effects within each cluster is to do the following:

 1) go back to the main CONN gui and create there a secondary analysis just like the original analysis but now leaving the default "eye(N)" contrasts in the between-subjects and between-conditions contrast fields (e.g. in your case select your seed ROI or voxel-to-voxel measure, then select all three groups and leave the default [1 0 0;0 1 0; 0 0 1] contrast, and select both conditionA and conditionB and leave the default [1 0;0 1] contrast). Then click on 'results explorer' just to have this analysis run (and look in the Matlab command window to learn in what folder those analysis results have been stored; other than this you can simpy disregard/close the new results-explorer window to avoid confusion).

 2) go back to your original analysis 'results explorer' window, click on 'display values' or 'import values', leave the default 'extract from these clusters' but now change the option that reads something like 'extract effects from this analyses' to 'extract effects from other analysis (select SPM.mat)' and select there the SPM.mat file generated in step (1) above

That will display or import now six values for each cluster, each representing the effect-sizes within each of your three subject groups and for each of your two conditions.

Hope this helps
Alfonso