And <any kind of science or pseudoscience> != <scientific objectivity>. The idea of scientific objectivity is "forming conclusions based only on independently testable measurements." That seems applicable here.
If political polling data gives you information about the outcome of an election -- if it can make your probability estimate more accurate than that of someone who hasn't seen any such data -- then I don't see what's so all-important about independent testability. Evidence is evidence; the possibility that a poll is skewed merely weakens its evidentiary value, by increasing the probability that the poll would have those results if they're not true.
