His source data for California is almost certainly wrong. If you go look at it in the csv its got a bunch of repeats and obvious gorp.
The data assumptions in the model itself is questionable; as someone pointed out below, he's basically modeling the generating process involved in giving people tests (put a different way: the spread of test availability). Which is pretty useless. For myself I wouldn't have gone through the trouble he did to make it more sciencey looking; exponential smoothing and sqrt(n) would give you virtually the same picture.
The data assumptions in the model itself is questionable; as someone pointed out below, he's basically modeling the generating process involved in giving people tests (put a different way: the spread of test availability). Which is pretty useless. For myself I wouldn't have gone through the trouble he did to make it more sciencey looking; exponential smoothing and sqrt(n) would give you virtually the same picture.