Yes, the Oracles pooped in FL a couple of days ago, which means there's been time for the pundits (and pollsters?) to do the usual post mortem to explain how they missed Gillum. It appears to be along the same line as 2016/Trump: ineffective stratification. Long time reader knows that one of the mantra of this quant is 'stratification is critical and it is haaahhd'. Getting to the moon was easier.
Given the typical size of such samples, ~1000, accuracy depends of either of two factors: the population is largely homogeneous with respect to the issue being measured, or strata/subsamples have to reflect both defining characteristics of subpopulations and their weights. Miss either, much less both, and your results don't hit the bullseye. The Manchurian President was, inadvertently, right. The pollsters didn't oversample the cities, but rather treated ex-urban areas in those three states as homogeneous to the cities. And, before 2016, the data on which the stratification was based that was an accurate criterion. Some people, voters, were fooled by lying. Some people, pollsters, were fooled by historical data.
It's worth mentioning that The Great Recession happened just because the Smartest Guys in the Room believed that explosive growth in house buying and thence prices was not only sustainable, but an organic growth. If something is too good to be true, it is. In the housing case, the rules of the game were changed, mostly by the Smartest Guys in the Room to their advantage. They were the piper and they played the tune. The rest of us paid the price.
So, with all that, here's one self-congratulatory missive.
A Change Research poll released last week showed Andrew Gillum winning the Florida Democratic primary with 33% of the vote — making it the only pollster to predict Gillum's stunning victory last night.
And, in a footnote, they tell us how they did it (well, not with trumpets blaring, naturally):
Post-stratification was done on age, gender, ethnicity, region, and 2016 Democratic primary vote.
Know your voter.