Better samples, better correlation coefficient

Dr. Matt Mitchell of the Mercatus Center gives us this chart relating the reliance on volatile progressive taxes and a states bugdet gap.
Two complaints.  States are an odd statistical samples, they vary as much as geese vary from tigers.   Second, the linear regression has such a low correlation coefficient it is useless.
I have made this complaint about cross section state studies, they sucks because we have counties in California that have twice the budget of some states.