Is the methodology that bad? What's wrong with it?

From their test protocol (https://vtechworks.lib.vt.edu/server/api/core/bitstreams/5e7...), it looks like they simulate a fall (with a model head inside it) against a target at an oblique angle, at six different impact locations and two speeds each. They go through 4 of each helmet model for the rest.

It seems a lot better than nothing (which is what we had before them, at least outside of manufacturers own private tests). Their research was initially funded by the IIHS, the group that does the highway crash tests.

How would you like to see it improved?