Using the cognitive construct networks for example, one can rank order tests by how much a particular network contributes to performance.

This provides a biologically based approach for evaluating existing tests - such as the standardized tests that have been around for more than 60 years. And it provide a brain-based metric for developing better tests that interrogate specific systems.

We show that composite scores yield better brain-behavior models (using the same fMRI data) if their subscores all rely on the same proportional networks - but yield degraded models if the subscores tap different systems - Suggesting a framework for forming improved composites.

This approach also provides a brain-based method for combining scores from different tests to better interrogate specific networks of interest. The approach is general in the sense that any networks of interest can be predefined and used for analyses.

A cool aspect of this in developing new external test measures, is that once you have a cohort of subjects with imaging data, the imaging doesn't need to be repeated - different tests can be developed and modeled using the same subjects and their original imaging data.