It turns out that none of these conclusions stand up to statistical scrutiny. Instead of voters in these places changing their party registration, the Catalist data suggest that L2 is likely using current polling and recent election data to generate a change in party modeling — and the AP is misinterpreting that as the registration figure.

That is pretty concerning, as far as telling people the truth goes, since the AP is the one reporting the trends, not L2. And this dovetails with something broader about the state of statistical political analysis today: The people doing the data work seem to be to be, on average, smart straight shooters — but then you have ideologues, partisans and click-hungry reporters with their own preconceptions injecting what is otherwise pretty sound analysis with a lot of bias.