Quote (ThatAlex @ Jul 22 2015 02:39pm)
Don't completely dismiss the predictive power of polls even though we are still a year plus out.
That's precisely what everyone should do based upon its extensive record. There are literally hundreds of examples to show how it isn't predictive, and even in the small sample of U.S. presidentials there's an iron-clad list that speaks to its failings: Phil Gramm, Mario Cuomo, Gore, Gephardt, Guiliani, and Huckabee/Palin just to name a few.
Head-to-head polling just isn't predictive right now, period. The real use for polling this early in the cycle is to measure favorability and level of name recognition.
Quote
Take a look at what FiveThirtyEight and Nate Silver have done over the past couple of elections, for instance.
Uh, Nate Silver is the very first person to argue that polling so far before Election Day is not predictive. He repeats this in practically every column that he personally writes due to how his critics incorrectly evaluate his commentary. He agrees with the academic consensus that it's
informative, not
predictive. He in particular uses a model that is reliant on voluminous data, meaning its accuracy increases closer to Election Day. That's probably his main motivation for agreeing with the consensus, and not with your contention.
Quote
There is still great predictive power to be had from polling and statistics, even 15+ months out.
There's great predictive power to be had from polling, but not from 15 months before Election Day. There's just no getting around the consensus here. I actually come down closer to the side of "useful" in the great debate of whether or not early polling is useful or useless than most people, but even I can't conjure up a reason to conclude that it's useful because it's predictive because it just flat out isn't. It's real value this early comes from combining it with other evaluative metrics, specifically information gained by and shared between campaigns, to increase what's known about how viable a candidate is and where.
Sometimes polling itself, in the baseline it helps create through constant releases, is useful to help refute the contentions of individual polls. A recent case is the new head-to-head surveys from Quinnipiac in Iowa, Colorado, and Virginia that most lazy media personalities are predictably freaking out over.