How did the polls get it so wrong?
To be fair, Nate Silver of FiveThirtyEight.com was leery about some of the last polls, just days before the election -- especially when Clinton’s numbers narrowed to a 2.5 to 3.5 point lead, within the range of error.
Still, in the days before that -- the week leading up to the election -- many, including Silver, had posted daily predictive missives, giving Clinton a big chance of winning. Early on November 8, FiveThirtyEight gave Clinton a 71.4% chance of winning when culling all polling data.
Perhaps one bit of polling information that would have been good to get was that of Cambridge Analytica, the President-elect Trump data team.
Matt Oczkowski, director of product for the company, told Wired.com, he noticed a decrease in black turnout, an increase in Hispanic turnout, and an increase in turnout among those over 55.
Additionally, there was overwhelming turnout of voters in rural areas -- especially in Ohio, Michigan, Iowa and Wisconsin.
Oczkowski said: “The amount of disenfranchised voters who came out to vote in rural America has been significant.” He added: “This is not something that political intuition would tell you, but our models predicted most of these states correctly.”
Why didn’t other polls have that? Because polling samples were very incorrect, especially among “likely voters.”
Welcome to the new world of big data for elections, stuff that, on the whole, wasn’t at all that predictive. Even Trump wasn’t all that convinced that big data was needed for his campaign -- only hiring Cambridge Analytica late in the campaign in the summer.
If you were were a brand manager of a particular product/service that failed to sell with this kind of data, you would ask yourself: Why -- in the age of ever expanding information -- did my big data resources fail?
And if you are a TV news network looking at the next election, what important and valuable election poll content will you use in future?