Looks like they have managed to trigger just about every cognitive bias that we have, all in one paragraph – never mind error margins and the effects of attitudinal versus behavioural data in polling situations.