For instance, on 11/5, Huffington Post was taking Nate Silver to task for "unskewing the polls," seemingly to create an image of a race that was closer than it "really is." [1]

For their own part, 538's forecast did have Clinton going into the election with a margin equivalent to polling error. [2] But that doesn't explain why the New York Times' Upshot, for instance, had Hillary at an 85% chance of winning on the eve of the election. [3]

(We should note also that the Trump "upset" was not just about white males without college education turning out in greater numbers than expected. Clinton shocked by underperforming 2012 Obama among black and Latino voters. [4])

So my question is: is the Trump "upset" an indictment all of big data as a tool for election prediction (or perhaps the social sciences at large)? Or do we just say that our models were largely miscalibrated and that the relative accuracy of 538's, for instance, shows that we simply need to learn from our mistakes to move forward using data in politics?

For some initial conversation on this topic, see this thread in the Trump victory HN post [5].

[0]: I found this one particularly insightful http://www.nytimes.com/2016/11/09/business/media/media-trump-clinton.html [1]: http://www.huffingtonpost.com/entry/nate-silver-election-forecast_us_581e1c33e4b0d9ce6fbc6f7f? [2]: https://twitter.com/NateSilver538/status/794571867449421829 [3]: http://www.nytimes.com/interactive/2016/upshot/presidential-polls-forecast.html [4]: http://fivethirtyeight.com/live-blog/2016-election-results-coverage/?lpup=12607239#livepress-update-12607239 [5]: https://news.ycombinator.com/item?id=12907961https://news.ycombinator.com/item?id=12907961