I really like the approach, this is exactly what statistical tools are for. I like it all the way up to the very end, where I feel this report drops the ball, in trying to derive a hierarchy from the results.

Hierarchies don't exist. Nothing in nature is organized in trees (data structures), not even actual trees (the other kind). Organization, relationships, form directed graphs (or the continuous domain equivalent). Hierarchies are deceptively simple because they planarize, meaning you can draw them on paper without crossing lines, or introducing indirect references. But that's often as deceiving as it is helpful.

Think of the "tree of life" - there is no one tree of life, its shape depends on which aspects of phenotype or genotype you select and organize around. Think of corporations and governments - you can either accept there's a lot of "dotted lines" in the org chart, turning it into a dense DAG, or never understand why people do what they do. And so here, it's also quite apparent there are correlations cutting across the proposed hierarchy, clearly visible even on Figure 2 - the one that urges us to disregard the left-to-right ordering of the constructs.

I mean, the researchers are likely perfectly aware of this. My worry is that 90%+ of the insight of the proposed model will get lost in final planarization into a hierarchy that gets into the next DSM.