Nate Silver @NateSilver538
@conorsen Of course it's possible! But as a matter of practice, it's way, way easier to build overconfident models than under-confident ones, especially given the nature of the data in a presidential election forecast (low n, correlated errors). — PolitiTweet.org