Given that the NHSTA says that their current "mathematical model" accurately predicts rollovers I'd have to say that they probably have something right.
and the NHSTA is obviously correlating actual SUV rollovers statistics with predicted by their model statistics and saying the two agree pretty well.
So the NHSTA is in effect saying, regardless of what the current safety features on the existing SUVs say should or should not happen, actual driving experience tracks with our predictions.

I recall fo many years in the 90s that Detroit denied that their SUVs were more likey to rollover or be no more dangerous than other vehicles (e.g. cars) on the road.
I also recall that Detroit said that the (then) anti-rollover safety measures in their SUVs were effective and safe.

Statistics since then show otherwise - otherwise why is Detroit putting all these new safety features in their SUVs if the current safety features are "good enough"?

Those newer features may help prevent rollovers, but the proof of the pudding is in the actual driving statistics of SUV rollovers with new models with all the new features. And only for newer SUVs - the older SUV models without them will of course remain dangerous until they are scrapped.

Which is why the NHSTA is finalising a track test to simulate actual driving conditions SUVs and drivers face.

Only then will the truth or otherwise of Detroits claims to the effectiveness of the new features be known.