A new Florida poll of early voters has generated great consternation in the blogosphere, because it claims that 28 percent of Republicans have already voted for Hillary Clinton over Donald Trump. This means Trump only got 72 percent of his own party’s vote in a must-win state.
By itself, this news guarantees a general election loss. If this is true, it’s over for Trump, period. Pack it up boys and the last one out can turn off the lights at the Trump Tower “war room.”
But here’s what doesn’t make sense. The same poll showed Sen. Marco Rubio trouncing his Democratic opponent Rep. Patrick Murphy by 6 points, 49 to 43 percent. This means we’ve got a major split ticket problem in Florida. But if that’s true, it affects way more than just Florida.
FiveThirtyEight’s “polls plus” model uses a “voter elasticity” factor by state, based on past data, to determine its election forecast. The factor is intended to account for how much a state tends to wander from the national poll numbers. The more “elastic” a state, the more “swing voters” are available in that state–either demographically, or by party.
Florida has a bit of elasticity–the factor is 1.05, meaning that for every point in the national polls, Florida will swing 1.05 points. Compare this to New Hampshire and Rhode Island, the two swingiest states at 1.28 and 1.29 respectively, and Florida is much closer to the baseline.
That means the TargetSmart/William & Mary Poll has blown the model by predicting a massive shift in party allegiance and demographics away from Trump and to Clinton. FiveThirtyEight’s model also accounts for states by demographics, with similar states voting similarly. Florida is so important in this part of the model because it’s considered the number one “tipping state” with a nearly 16 percent chance of “tipping the election.”
So if Florida is off by 6 or 7 points–an enormous difference in a state that’s divided on a razor edge (FiveThirtyEight has Trump’s chance of winning Florida at 52.3 percent for “polls plus” and 49.4 percent for “polls only.”), the national election forecasts are likely equally trashed. It’s that close.
I actually trust FiveThirtyEight’s methodology. Nate Silver has accounted for all the variables and done it very transparently. The problem is the data. If the polls and pollster ratings used to weight FiveThirtyEight (and RealClearPolitics, which relies on most of the same polls–RCP has Trump up by a mere 1 point) are that far off from the actual voting results, we’re in trouble.
Based on the evidence, it would appear that the TargetSmart poll is wrong, skewed, or somehow demographically invalid. It simply doesn’t make sense that Florida Republican voters would be so far off the established polling for Trump but right on the nose for Rubio. But it’s worth repeating: if this poll is correct, Trump is finished.