Resting state fMRI consists of continuous neural-activity recordings over a period of several minutes without structured experimental manipulation. These measurements are typically summarized into a correlation matrix between activity in p predetermined brain-regions (p between 90 and 500). Neurologists are interested in identifying localized differences in correlation across patient populations, e.g. disease versus control. However, running a univariate test for each region-pair separately requires strong multiplicity corrections, making it hard to detect differences. In this work, we reparametrize the matrix of differences between populations as p main effects representing change for each region and ~p^2/2 “interactions”. We propose a likelihood based method that accounts for the dependence between the parameters, between correlation coefficients, and within the time-series. Adjusted intervals are derived for the main effects and for the second-order effects that are selected. In simulations we show that our method increases power to detect changes. In examples, we detect regions of difference in Amnesia patients compared to healthy controls.