EVIDENCE FOR DIMINISHING RETURNS WITH INCREASING GROUP SIZE IN CORTICOCEREBELLAR CONNECTIVITY DATA

Increasing group size is often considered a solution to low statistical power, but the benefits may vary with the systems being studied. Here, we study the impact of increasing group size on replicability in the corticocerebellar system. We investigated the connectivity of 411 cerebellar cortical locations using resting-state fMRI. We tested the impact of increasing group size on two measures of replicability. We separated ~70,000,000 connectivity values into replicable and non-replicable categories, repeating the analysis with 10 group sizes. Replicability did not increase appreciably for group sizes over 450.
We processed fMRI data from 500 CamCAN participants (CONN Toolbox). First-level analyses estimated connectivity between 411 equidistant (6mm) cerebellar seeds and ~174,000 extra-cerebellar grey matter targets, generating 205,500 unique beta images. To test replicability, we computed 12,330 independent second-level analyses using split-half resampling across 10 matched groups (N=50 to 500, steps of 50, p<0.05). We constructed 411 x ~174,000 matrices populated with T-statistics. Replicability was tested by calculating: (i) proportions of supra-threshold replicable connections and (ii) linear regressions (R²).
The two replicability measures changed similarly with group size, plateauing between 450 and 500 participants (C). Proportions of replicable connectivity values did not exceed 42%. Replicable connectivity T-values were higher than non-replicable ones (B). While most areas in the cerebral cortex are functionally connected with few cerebellar cortical locations, operating as modules, some communicate with large proportions of cerebellar cortex, suggesting global influence (A), challenging the traditional view of purely modular corticocerebellar organisation.