There are some interesting, and in some cases, potentially disturbing, things going on with the state by state numbers in the current election. Most of this has to do with third party candidates, and most of it with Gary Johnson.
First, I’ll note, that despite fears among liberals and progressives that a lot of Bernie Bots would flock to third party candidates and eschew Clinton, there is no strong evidence that Clinton is losing much to any third party candidates. However, in some states, especially those with libertarian tendencies, Gary Johnson is doing fairly well. And, this had been hurting Trump.
However, lately, there has been a shift backwards in at least one state, New Hampshire. Johnson supporters are abandoning Johnson and switching to Trump, as though they were trying to shore up his position there. This has brought the Trump-Clinton numbers to within the margin of error.
In other words, Libertarian White Males in the “Live Free or Die” state are flocking to Misogynist Racist Trump’s aid rather than “voting on principle” which is what, I assume, they were formally pretending to do. And, this could cost Clinton a couple of electoral votes if the trend continues.
Meanwhile, something like this may be happening in Virginia, but in the opposite direction, where Johnson appears to be getting a lot of Trump votes, maybe more as time goes on.
I don’t have time to do any of this right now, but when this is all over, it would be very interesting to look at the third party effects in this race.
OK, now on to the model. Let me explain the basic approach I take, which is different from other predictors (though 538 may have quietly adopted part of my approach for the general, as they’ve added something that looks a lot like my primary methods to their analysis).
Assume that all polls are good, and that all states are recently sampled with high quality polls with good methods and good samples.
OK, after you’ve stopped laughing, work with this assumption for a minutes. If this was the case, then you could use those polls to predict the electoral outcome, and unless the electoral outcome was really close, or something major went wrong, your prediction would be clear as two who won, and very close if not spot on as to how many electoral votes ultimately go to each candidate.
Now assume that we don’t have polls at all, but we have some numbers indicating how people in a given state are likely to vote (like, if they went for Romney, they are likely to go GOP) or numbers indicating how people will vote based on ethnicity (like, African Americans are not likely to vote for any of the candidates other than Clinton, or among whites there is a certain percentage of White Supremacists, so they’ll vote for Johnson or Trump, etc.) If these numbers are accurate, you can predict the state by state outcome.
We don’t have either of these, but we do have a little of each.
My method uses only a subset of polls, hopefully across a range of states (geographically, politically, etc.), that are taken by higher end polling agencies and recently. These are then combined with data on percentage of voters in that state that voted for Romney, and the classically defined ethnic breakdown for that state, to come up with a muliti-variable regression model. This model uses the percent of the vote that Trump gets out of Trump vs Clinton as the dependent variable, and a Romney number, and the ethnic breakdowns, as the independent variables.
I exclude some states that have recent data but that are beating to their own drums. In this case, Iowa is doing something different, and nobody understands it. Also, Virginia is doing something different and has not been analyses yet. So, even though I have recent data from those two states, they are excluded.
The polls need to be mostly or entirely after the famous “bus tape” and most are after the second presidential debate. These polls come from Utah, Wisconsin, Georgia, Missouri, Indiana, Texas, Alaska, Ohio, Colorado, New Hampshire, Florida, North Carolina, Nevada, Maine, Pennsylvania, Oregon, Michigan, and Washington.
So, good polls are assumed to be nearly perfect, and they show the relationship between available prior voting patterns and demographics and the likely outcome. Then, this model is applied to all states (even those with the good polls) to come up with a list of states and their corresponding “Trumposity”
The result of that analysis is this:
|District of Columbia||0.339691168|
You can now split the table at the 50-50% mark to decide which states will break for Clinton and which will break for Trump.
(Note: Alaska will always break for Trump. It is located near the 50-50 line because Alaska is a special snowflake state. Ignore it, just keep it red on any map, and that will do.)
The first map I want to show you is the map of states that are in the Clinton Camp that are a) most Clinton leaning in this analysis, and b) sufficient to get Clinton to 270:
I added Virginia and colored it light blue. The reason I did this is that Iowa is a presumed-Clinton state in this mode, but is in fact, polling for Trump, because people in Iowa seem to have a new goal in life: Pissing off the parties and the electorate sufficiently that nobody cares about them any more, and the Iowa Caucus is no longer allowed to take the prominent role it has for all these year. I predict that if Iowa breaks for Trump, in four years, the first contest will not be the Iowa Caucus.
By adding Virginia and thus potentially starting early on the process of regarding Iowa as irrelevant to electoral politics, we have a list of states that is clearly Clinton and sufficient to put the former first lady back in the White House but with a different job.
Now, let’s do the same thing for Trump. What states are required to put him past the 270 line?
In this case, I’ve colored pink the states that my model puts in the Clinton column but that are on the Trump-end of that part of the list (see table above), that are required to give Trump the election.
Ohio is actually possible. My model shows Ohio going to Clinton, but recent polling shows that Ohioans are more white supremacist than we might have thought. So may be Trump gets Ohio, but I don’t think he’g soing to get all those other pink states, or even any of them, likely.
Putting this a slightly different way, the solid Trump states (in my model) plus Ohio is still under 200 electoral points.
The current most likely outcome according to this model is this:
That would be an electoral blowout.
What happens if some of the more suspect states go backwards and vote for Trump? Iowa is threatening its own irrelevance, New Hampshire is acting strange, Ohio is polling towards Trump, and North Carolina, Arizona and Florida are close to the mid point. Change all of those states to Trump, and we get this nailbiter:
The difference between these last two maps is clearly going to be the focus of interest over the next several days.
Colored here in red, for Danger, not for Trump/GOP, are the states that need to be watched closely, for which we eagerly await new polling, because they are either close, near the middle, or acting strange over recent days:
That’s my story and I’m sticking to it. Until at least Tuesday or so.