Hi

This might be an easy one. Thought someone could give me the equation used to plot ROI-ROI effect size (plot effect button on 2nd level analysis). So I have the z scores of say source A and seed B. What was done to get the bar plot? My qn was based on 18b but I've tested on 19c and got two plots one for each condition. In 18b was it an average across both conditions? Thank you.

Kevin

Hi Kevin,

If your second-level model is of the form:

Z = X*B + noise

where Z are the Fisher transformed correlations (across subjects and condtions), X is your design matrix, and B is the (unknown) matrix of group-level effects (estimated for example as B=X\Z); and your hypothesis is of the form:

C*B*M' = 0

where C is your between-subjects contrast vector/matrix, and M is your within-subjects (between-conditions) contrast vector/matrix; then the effect size barplots represent (in the newest versions of CONN) the elements b_ij of the matrix B (the estimated effects in your group-level model) and their associated 90% confidence intervals, for any i with at least one non-zero element along the i-th column of C, and any j with at least one non-zero element along the j-th column of M. For example, if you are evaluating the differences in connectivity between two groups of subjects using a two-sample t-test, then the bar-plots will represent the average connectivity values within each group of subjects. 

In older versions of CONN, the barplots would represent the contrast values C*B*M' directly instead of the relevant B regressors, so, more precisely sum_ij C_ni*B_ij*M*mj for each n-th row of C and each m-th row of M. For example, in the same case as above of a two-sample t-test, the bar-plots in older CONN releases will represent the difference in average connectivity values between the two groups. 

Hope this helps
Alfonso

Originally posted by Kevin T: