Reading Elsewhere

• The New Yorker published a brief history on polling in US Politics a few years back worth reading.

• Clinton's 2016 data analytics team apparently worked using only a single model, codenamed Ada. From how people describe it, it sounds like a [Bayesian] MCMC model.1n+1 reported, By sampling a representative subset of all possible variations—the so-called Monte Carlo method of quantitative analysis—Ada would produce a set of outcomes. Edward Tenner claims, in his book The Efficiency Paradox: What Big Data Can't Do, that he had a computer scientist examine the source code of Ada, The software, Ada, could generate hundreds of thousands of simulations, but according to a computer scientist who studied the program, it was flawed by the assumptions built into it by Kriegel and his staff, biases that kept it away from countering a loss of support. (pg 67)

• Bloomberg Politics had a pre-election 2016 review of forecasts (which I learned of from New York magazine article).