LAST weekend I ate a cricket on national television. Based on my statistical analysis of presidential polls, I made a bet that if Donald J. Trump won more than 240 Electoral College votes, I would chow down. Like millions of voters in both parties, I was surprised by the outcome. As a consolation, I’d like to learn what went wrong — and to figure out how pollsters might do better when they puzzle over our polarized electorate.
Data-based websites, from Facebook on down, have a responsibility to convey accurate information. In this regard, I owe an apology to readers of the Princeton Election Consortium, which I publish. My primary purpose was to show people where to put their campaigning energies by revealing which races were on a razor’s edge. I advised my readers to focus on close Senate races in states where the presidential race was also close. But I also reported an extremely high probability that Hillary Clinton would win, which was published by The New York Times alongside its own model.