I wouldnt say it is a flaw, really. The data in general is a good approximation of auto dependence. And any researcher who isn’t an idiot will see the same thing you did and simply discard the data in these counties as obvious outliers. Sure, we can imagine a more accurate metric for measuring auto dependency for the purposes of creating a very nice map for public consumption. But it your purpose is simply to conduct some statistical analysis, I don’t think this dataset is bad - or at least not a bad start.