His point about polling missing in the opposite direction of conventional wisdom is nonscientific garbage.
I said this earlier, but if an error is truly an error it can’t be predicted in advance, otherwise it would and should be accounted for. True errors are random. Having the same or predictable bias is just bad science.