Fig. 5.

Results from the simulation. A: Distributions of single-subject posterior dimensionality estimates for a ground-truth dimensionality of 4, 8, or 12 and increasing noise levels. As noise increases, the estimates become less accurate and less certain, as indicated by the width of the distributions. For the highest noise level, the posterior distributions for all ground-truth dimensionalities overlap largely. B: Average reconstruction correlations for the different ground-truth dimensionalities and increasing noise levels. As the noise level increases, reconstruction correlations drop, and this effect is the same across the three different ground-truth dimensionalities.