State-of-the-art neuroimaging data provides several ways today to construct "brain connectomics" based on structural and functional measures. Connection-wise statistical assessment suffers from the large amount of multiple comparisons and is suboptimal as it ignores the positive dependency structure that is present in such data. This aspect is taken into consideration by two-step methods that we developed recently. These two-step methods are adapted to brain connectivity statistical analysis because they guarantee the strong control of the type I error rate under weak assumptions (e.g., weaker than assumptions needed by SPM and network based statistic (NBS)). We present a practical example in which we compare the performance of the univariate and multivariate two-step methods to the standard methods (that do not consider data structure and positive dependence). The study consists in detecting topological nodal differences in functional connectivity between two mental states; i.e., resting state (RS) and movie watching (MW). The study do not only show the gain of the power of detecting real differences when using two-step methods because the information of the data structure is exploited and not ignored, but also reveals that functional connectivity is strongly increased during RS. In particular, the medial temporal gyrus (MTG) becomes more independently active when processing information during MW. However, we observe an opposite trend for nodal efficiency, which might indicate that although nodal strength is increased during RS, the average path length is decreased, as some nodes are not part of large RS networks.