Thanks to Tailcalled for catching this—it didn't materially affect the
results (up to random noise) because I'm comparing distributions
centered at 1 with others centered at zero (and anything minus zero is
the thing), but geez, this was embarrassing!