Hi CONN team, 

I’m reporting results from a seed-to-voxel task-based functional connectivity analysis in CONN (not gPPI).

At the second-level (Patients > Controls, Right-Hand Draw condition), the results table lists t(42), uncorrected p, and FDR-corrected p, and the associated bar plot is labeled “Effect sizes".

1. What metric the “effect size” label refers to — are the bar values the mean Fisher Z correlation for each group, or some standardized statistic?

2. For the contrast itself (Patients > Controls), in addition to showing the bar plot, what is the most accurate way to report or compute an effect size in the manuscript? given that the displayed values in the bar plot are per-group for each cluster rather than per-contrast?

Thank you, 

Samah

Hi Samah

Regarding your first question I am assuming your are referring to the 'Plot effects' button in the results explorer window? If so, the actual meaning of the 'effect size' values displayed in these plots depends on your specific second-level model. For an analysis like yours looking at the differene between two groups (Patients > Controls) the effect-size values represent the average Fisher-transformed correlations within each group separately (while in a more general scenario 'effect sizes' represent the regressor coefficients of your second-level GLM model).

And regarding your second question, in a between-group comparison the "contrast effect-size" would typically be reporting the difference in connectivity strength between the two groups, which can be computed directly as the difference between the two displayed effect-sizes (the connectivity within each group), so I would recommend reporting the original (per-group) effect-sizes as they are equally or more informative that simply reporting their difference. 

Hope this helps

Alfonso

Originally posted by Samah Gassass :

Hi CONN team, 

I’m reporting results from a seed-to-voxel task-based functional connectivity analysis in CONN (not gPPI).

At the second-level (Patients > Controls, Right-Hand Draw condition), the results table lists t(42), uncorrected p, and FDR-corrected p, and the associated bar plot is labeled “Effect sizes".

1. What metric the “effect size” label refers to — are the bar values the mean Fisher Z correlation for each group, or some standardized statistic?

2. For the contrast itself (Patients > Controls), in addition to showing the bar plot, what is the most accurate way to report or compute an effect size in the manuscript? given that the displayed values in the bar plot are per-group for each cluster rather than per-contrast?

Thank you, 

Samah