The two ideas are joined together in the URL, not in the piece. He's talking about why he thinks the polling was so off from a data-guy's viewpoint--not that it had anything to do with the result. I don't know anything about polling models, but I found his remedy interesting (also fits with the way he approaches baseball stats, which is basically not to decide x and y are irrelevant and discard them):
Junk it. It doesn't work. Deal with people you don't know. Use 25 different models. Go to a shopping mall and set up a booth where you hand out candy bars to anybody who'll fill out a poll for you. Walk down the street and stop every seventh person you see and ask. Put up buckets with pictures of Trump and Hillary and ask random people to drop a quarter in one or the other and count them up. Do it a hundred different ways and see if you can figure out 900 more. Then you get a broader understanding rather than a narrow understanding.
― clemenza, Saturday, 12 November 2016 04:40 (seven years ago) link