My Twitter account is just getting started. Follow me here!

UPDATE: The Washington Post mega-poll is not predictive in a state-by-state election

UPDATE: Nate Silver analyzed these types of polls yesterday, where a pollster will call people in all fifty states and then release results from each state based on demographic data. Unsurprisingly, his analysis agreed with mine, that a simple demographic filter does nothing to effectively determine who will actually vote in each of the fifty states.

He pointed out that the method has not been empirically tested and that the demographic weighting is too slight and not reflective of actual turnout. Further, he noted that identifying where voters live based on IP address is a flawed process, certain to place some respondents in the wrong state. The process produces weird results in states, results like we note below in Texas and Mississippi. He concludes that these state-by-state subsamples from a fifty-state poll are not the same as actual polls directed specifically at one state at a time.

The Washington Post commissioned a poll of all registered voters and decided to draw conclusions about who holds an electoral advantage from the results. The problem for the Washington Post, and they undoubtedly know this, is that polls of all registered voters always heavily favor Democrats and skew the results in such a way that they are ineffective in determining a likely winner. To compound this problem, when applying this registered voter approach to fifty states and then applying a simple demographic model as a filter, as opposed to a well-tested historically based voter turnout model, you really get bad results.

The Survey Monkey poll weighted the results to match "demographic characteristics of registered voters in each state," rather than typical voter turnout models. Elections are not decided by a demographically representative pool of registered voters. If that were the case, the Democrats would win every national election because there are more registered Democrats. The problem for Democrats is that they have a big problem getting their voters to actually turn out and vote, and that is true every time.

Republicans do better in off-year elections because their voters turn out, even when the national focus is not on politics. Democrats do better in presidential years because so much attention is paid to politics. But even in a presidential year, only 60% of the potential electorate will turn out and vote.

The Washington Post's pollster is not filtering for who is likely to vote. What this method of polling achieves is a built-in bump for any Democrat, usually a big one. Nate Silver demonstrates the statistical bias towards Democrats in both presidential and non-presidential election years.

One need not delve into the statistics to see the invalidity of this type of poll for determining who will win fifty state elections. One need only look at the results to find that the poll is not predictive.

More Analysis from the Ref

**Latest battleground numbers show the race tied on Labor day

Mistake? CBS Pollster Confuses its Registered Voter Poll with a Likely Voter Poll, Boosts Clinton Number

With Virginia now Tied, Trump is Rapidly Closing Battleground Gap

Trump is Closer Than You Think

Donald Trump is Not Trailing by Much in this Election, and Actually May be Ahead

The Washington Post poll found that Trump is tied with Hillary in Mississippi and Texas. There is no chance that Hillary will win either state. There simply are not enough NeverTrumpers in the GOP to accomplish this task. But in the imaginary world the Washington Post created, that could happen.

This poll does accomplish one thing, however, something WaPo would not have wanted to demonstrate. Because Trump is running even or ahead in a number of normally Democratic states even when the sample is skewed to favor Democrats, it demonstrates that Trump has real strength in those states.

The Ref's Battleground Index

Clinton +0.24 Points in All Ref Battlegrounds

There are now four realisitic scenarios where Trump can win without Pennsylvania.

Registered voter polls, as opposed to likely voter polls, always favor Democrats


Please follow my new account on Twitter for daily updates on the election!