Figure 4 shows numerical simulation of the expected score increase for different firstlevel variance configurations. As before, the expected increase is independent of the secondlevel effect size but will depend on the first level variance configuration for group 1 and group 2 as well as the different secondlevel variances and . The added flexibility of the heteroscedastic model is important for a variety of real FMRI experiment where the two groups naturally will have different variance configurations, e.g. studies of patients vs. nonpatients. Once again, significant changes in score (e.g. ) can be seen over a large set of configurations.
