My Predictions for the 2012 Presidential Election

“Every revolution evaporates and leaves behind only the slime of a new bureaucracy” – Franz Kafka

The election is only a ten days away, and the predictions, claims of momentum, and hand-wringing are all over the place. So, I figured I’d join in, with some predictions (sort of) of my own. I was chatting with possiblywrong the other day and we discussed how predictions are only useful if they are accurate and made before everybody else has figured it out. So, I’ll attempt to get these on the books before the crowd sees what is happening.

First, let me reiterate my baseline position. I don’t like either Romney or Obama. Of the two, I suspect Romney is the lesser of two evils in the short term, and the long term is hard to predict. Obama fundamentally lacks leadership skills. Agree or disagree with their policies, but Bush 43, Clinton, and Reagan all had leadership abilities that far surpass Barack Obama (Bush 41 did too, but I find it hard to mention him in the same sentence as the others).

Executive Summary of Predictions …

This  election has turned. Romney is ahead, perhaps narrowly, but ahead. Who wins and loses will hinge on turnout and ground game and all that sort of thing – but Romney is ahead and will likely win. (How likely? who knows? … but I’ll offer a guess.)

Mitt Romney will win the popular vote. I call it Romney +1.5 at this point (I actually like +1.7 better, but will stay conservative at +1.5). Winning the popular vote is not the same as winning the electoral college vote, but Romney will win the electoral college vote too.

Real Clear Politics currently has the solids, likely, and leaners at Obama 201 – Romney 191, and I think they have all of those states right. The 11 remaining “toss-up” states are Colorado, Iowa, Florida, Michigan, Nevada, New Hampshire, North Carolina, Ohio, Pennsylvania, Virginia, and Wisconsin. Of these states, Romney will win Florida (29), North Carolina (15), Virginia (13), Colorado (9), and Ohio (18) to get over the 270 mark. Now 270 is all you need, but Romney will also win Nevada (6) and Iowa (6) and eek out a win in New Hampshire (4) to get to a healthy 291 electoral votes. Obama will hold on in Wisconsin, Michigan, and Pennsylvania … but they’ll be “too close to call” for longer than you might suspect. (Not Michigan – but the other two.)

Popular Vote Prediction …

I have a few graphs here to describe my popular vote prediction. First, I show the Real Clear Politics average of polls (plotted versus “days to the election”) from 2004 (Bush) and 2008 (Obama). The plot gives the RCP averages normalized to show percentage of the vote among the two primary candidates. (That is, we take out the undecideds or third party votes and renormalize.) Only the eventual winner is shown (the loser will be 100% minus winner), with Bush 2004 in teal and Obama 2008 in magenta. (I also offer some commentary on what the pundits were saying at various points in the race.)

The square at the end is the actual outcome of the popular vote (again, renormalized to compare only the two primary candidates).  On the next chart, we zoom in on the last 60 days to the election – and offer some more commentary. We also add the current standings for Obama and Romney (RCP average renormalized to 100%).Several things jump out from this chart. In both 2004 and 2008 there were big moves in the polls in the month of September leading into the early part of October. But once we’re mid-way into October, things stabilized and hovered near the final result (squares at the right). 2012 has thus far been no different, with the last big move being a pro-Romney shift at -35 days closing out  around -27 days. After that, it’s pretty steady. In fact, the prediction on “today” in each of the prior two races has been pretty dang good. By that calculus alone I’d make the race Romney +1.05. (As I’m writing, the average of polls is Romney 47.9 – Obama 46.9, which turns into 50.527% to 49.473% upon normalization.)

To that we add that the trend over the next 10 days will likely be slightly pro-Romney. First, the current RCP average is only as “low” for Romney because of older polls that have Obama up. Those polls will be replaced soon (hopefully by polls from the same outfit so we can see trends for fixed methodology). Of the polls that have any data from the last week, only the Investors Business Daily poll has Obama ahead (47-45), with the other tracking polls showing Romney +5 (Gallup) and Romney +4 (Rasmussen). If we only consider “recent” polls (polling in the last week), the RCP average shifts to 48.3 – 46.2, or a 2.3% (after normalization) Romney advantage. Now, I don’t note this to cherry pick the polls, just to say that my inclination is that the RCP average (which I’m using as a statistic) will trend more Romney over the next day or two. Will it then reverse course? Possibly.

There’s more bad statistics for Obama though if we keep digging. There is a slight (not -1, but near -.3) negative relationship between the number of undecideds in a poll and the Romney lead. That is, the polls that have Obama up are the ones with the most undecideds. The ones with Romney up are the ones with the least undecideds. Want to take a guess at which way the undecideds are breaking?

Finally, I don’t see the driver toward Obama in the final 10 days. He’s the incumbent – folks have had four years to make a decision on him. The large Romney bump in late September and early October was people making a decision on Romney. Benghazi won’t change any votes – but it might change turnout (to the worse for Obama). The 2008 election was the “dreamer” election, with the hope that Obama would be the golden one. In 2012 the people with real jobs and real worries about taking care of their families are going to show up in far bigger numbers (and early voting results have indicated as much).

Fun with Electoral College Maps …

Now that we’ve laid out our call of Romney +1.5 in the popular vote, let’s consider the electoral vote for a bit. I’ve gone back and pulled out the results for the last three presidential elections (2000, 2004, and 2008). Below I’ve color-coded the electoral map based on how states behaved relative to the national popular vote. A state is noted as “solid” for a candidate if it voted more toward that candidate’s party than did the nation at large by at least two points in each of the previous three elections. A state is noted as “likely” if it always voted more toward a party, but dropped below the two point threshold at least once. A state is noted “leaning” if it split between the parties (relative to the national popular vote) – and is given the color of which party won two of the three elections on a relative basis.

(Thanks again to Real Clear Politics for the interface to create such a map)

Now, if you’re a Romney supporter you’ll be pleased to know that 281 electoral votes have gone Republican more often than Democrat relative to the national popular vote. That means if the trend holds and Romney wins the popular vote (which I think he will) then Romney’s in pretty good shape. But that “if the trend holds” line is serious.

Now, let’s consider a few “toss-up” states. First, the obvious one that is still listed as toss-up on a lot of maps but is solid red here – North Carolina. In 2008 the nation voted +7.2 for Obama, while North Carolina was +0.3 for Obama, leaning 6.9 points Republican. In 2004 the nation was +2.4 for Bush and North Carolina was +12.4 for Bush, leaning 10 points Republican. In 2000 the nation was -0.5 points for Bush and North Carolina was +12.8 points for Bush, leaning 13.3 points Republican. Now, you will point out that there is a trend here – the Republican margins are shrinking. Yes, but the slope doesn’t get us anywhere near zero yet. The notion that Romney can be leading in the popular vote (or trailing by less than 3!) and North Carolina might go Democrat is, umm, silly. I won’t say those “North Carolina is a toss-up” folks are smoking crack, but there smoking some sort of controlled substance in their respective pipes. Now, I actually think the trend will hold, and North Carolina will be less Republican leaning than in 2008 (perhaps +4 Republican). But, given the +1.5 Romney popular vote win, North Carolina will be +5.5 and solidly red.

And what about Florida? Florida leaned Republican from 2000 through 2008 by 0.5, 2.6, and 4.4 points. Always leaning Republican and trending moreso. A 50-50 national vote makes Florida solidly Red, using this (arguably weak) statistic.

Then there’s Ohio. Ohio leaned +4 Republican in 2000, 0.3 Democrat in 2004, and +2.6 Republican  in 2008. (That’s right, when Obama carried Ohio in 2008 he did so by 4.6 points, versus a 7.2 point national edge.) Now I know that there are some bad feelings towards Republicans in Ohio over some local and state level scandals, but the state still leans red by my reading and Romney winning popular vote likely makes Ohio red this time around too. (Even with the massively pro-Democrat turnout model of 2008 the demographics of Ohio were still more red than the nation.)

More interesting, I think, is Virginia. From 2000 to 2008 Virginia leaned Republican every election, by 8.6, 5.8, and 0.9. That 0.9 in 2008 is interesting. Two factors to consider here. First, the DC suburbs have continued to grow, which are more Democrat than the rural parts of the state. Secondly, the turnout model in 2008 was decidedly different than 2000 and 2004 and better matched a less-Republican-leaning state. Obama will likely need both of those factors to weigh in this time if he wants to win Virginia again … but the turnout model won’t even be close to 2008. Oh, I suspect the Democrats are fired up for Obama (but not as much as last time) – but the Tea Party Republicans are also fired up to get rid of Obama. A turnout similar to 2010 would put VA solidly red. I don’t think that will happen (Obama wasn’t on the ballot in 2010), but his 2008 turnout advantage will evaporate and Virginia will go back red.

And what about Wisconsin? Reports have Romney and Ryan moving in attempting to swing it to red. Well, in 2000 Wisconsin leaned Republican by 0.3 points. But in 2004 and 2008 it leaned Democrat by 2.8 points and 6.7 points, respectively. The Republicans believe that the Scott Walker revolution is enough to push the state into the red column. I doubt it. I suspect it will stay blue. Though, a strong finish by Romney will make it “too close to call” until enough other states have been called to give him the election.

Nevada is a wild card, leaning slightly Republican in 2000 (4.0) and 2004 (0.2) but strongly Democrat in 2008 (5.3). The reason I say it’s a wild card is the heavy Mormon population, which might turn out in greater numbers than usual and vote more Republican (Harry Reid – a liberal Mormon – manages to win there … surely some of his Mormon supporters will flip to vote for fellow Mormon Romney). I have no real way to quantify this effect, but I’ve gone out on a limb to predict that Nevada will flip to Romney.

Then there are Colorado and Iowa, both leaning slightly Democrat (and trending slightly Democrat) but by small enough margins to put it in range. Time will tell, but I’ve called them for Romney. [Update 10:17 PM, 27 October – Des Moines Register has just endorsed Mitt Romney. For reference, the past nine Register endorsements have gone to Democrats: Obama, Kerry, Gore, Clinton, Clinton, Dukakis, Mondale, Carter, Carter. If they’ve turned on Obama it is exceedingly telling – solidifies my Iowa-for-Romney call.]

“Mr. Silver, Meet Mr. Gingrich” …

(In this business it’s better to be right than to be happy)

Back during the 2012 Republican primary process a political commentator noted just how silly the Newt Gingrich campaign had become. Not only was he losing, but his behavior was actually hurting his future earnings potential. Nobody would really want to hire him to speak as “former Speaker of the House” and architect of the “Contract with America” … he was becoming angry (and crazy) old Newt.

If my predictions are even close to right then Nate Silver over at may be approaching Gingrich’s dangerous proposition. For fun, I have overlaid Silver’s predictions for the popular vote count (note, I also converted these to percentage out of the two major candidates).

Holy divergence Batman! The lighter hues are the Silver estimates. In the September time frame they were right on top of the RCP average. Then, as the race turned decidedly toward Romney, the Silver model followed with a lag – settling at about a point to a point and a half bias. OK, maybe it’s not a bias, maybe it’s a “house effect” as Silver likes to call it.

Now if I were a mathematician … oh, wait, I am a mathematician. Well, as a mathematician I’d look at a plot like that and be quite concerned that the model had diverged from the mean of the distribution. Then I’d pull every thread to figure out why my model was wrong, or to figure out why my model was right but the general average of estimates was wrong.

Ahhh, I hear the complaints now. “But the state polls say” stop it. Spare me your feeble attempts. First, we’re talking about the popular vote. But if you want to consider state polls then let’s do just that. Below I show the errors in the RCP averages of state polls for the 2004 and 2008 elections in the 11 swing states. That is, the RCP margin minus the true margin in the election.

You read that correctly – the standard deviation in the error amongst the RCP average of state polls is 3 points. Let’s tuck this away under the “state polls are useless” file. (“But we know how to pick the right state polls to give a good answer” stop it, you’re being silly.)

Mr. Silver, you’ve got a good thing going here. You’re blog has been picked up by the New York Times. You’ve got a book on the market. Don’t throw it all away on this crazy dream of a second Obama term (and handily at that).

(Two side notes are of use here. First, if Silver is right and the RCP averages are WAY off after being right in the past – or the RCP average trends to match him – then he’ll be seen as a genius and get to do this for a long time … and get paid fairly well. Second, even if he’s wrong, it likely won’t matter. It has not mattered in the past when liberal alarmists have been wrong about any number of issues; they still somehow maintain credibility. Why should politics be any different?)

Watching the Mega Trends …

We’ve noted this in prior posts but it seems worth note here. There are two strong trends playing out against each other here, and there is a razor thin margin (perhaps) over which one will hold on. The stronger trend is that “new party incumbents” don’t lose. Over the past 120 years or so the White House has changed parties 10 times, giving 10 elections where the “new party incumbent” could have been thrown out. This group has only lost one election – Jimmy Carter in 1980.

The weaker trend is that this country does not re-elect Democrats … not often anyway. Bill Clinton won re-election in 1996. Jimmy Carter lost in 1980. Lyndon Johnson was so hated that he did not even win his party’s nomination in 1968. The same goes for Harry Truman who didn’t even get the nomination in 1952. That’s right, before Clinton you have to go back to Franklin D. Roosevelt and WWII to find a Democrat incumbent who won re-election.

These two trends oppose one another, and one will be proven stronger in 10 days. My suspicion is that the weaker trend will prevail. This nation is center-right and gets real uncomfortable when faced with actual liberal policies and liberals in control.

This entry was posted in Uncategorized. Bookmark the permalink.

4 Responses to My Predictions for the 2012 Presidential Election

  1. You may have already seen this, but I thought you might find it interesting: a couple of CU poli-sci professors are also predicting a big Romney win (link here). What is interesting is that (1) they appear to use primarily *economic* data more than (or even instead of, I’m not sure) polling data in their model, and (2) their model back-predicts every election since 1980 correctly, including the 2000 popular/electoral vote swap.

    (Without an actual paper that I could find describing the model in more detail, this does not sound terribly impressive to me. I am reasonably confident that I can find coefficients to draw a curve through what amounts to a measly *eight data points* that yields either desired 2012 outcome.)

    I think that I am apparently one of those “silly” readers making a “feeble attempt” at not dismissing state polls entirely. I am pretty slow in understanding, because I still don’t get it. Specifically, do you have any of the raw data (either PM me or provide a link if it’s easily accessible somewhere) from which you prepared the histogram of state poll averages from the past two elections? I ask because I think the distribution of error in *actual percentage of vote* is not really important in determining the allocation of electoral votes… at least, in any state except for NE and ME. I wonder, of those 22 samples, how many were simply *on the wrong side of 50%*, which is all that really matters. For example, if instead of the RCP averages, if we used Silver’s per-state predictions, for which I imagine we only have 2008 data, I wonder what his distribution of error would be across those same states… compared with the fact that he predicted all but one state correctly, from a “right side of 50%” perspective.

    I guess I am still struggling with the argument that a state poll (i.e., a sample of a smaller set) is useless, but a national poll (i.e., a similarly-sized sample but of a much larger set) can be trusted… not only to predict popular vote despite a smaller relative sample size, but even to predict the behavior of a *subset* of the population being sampled (i.e. voters within a single state).

    • nomasir says:

      I’ve seen the UC paper / predictions. I’m always watching out for another “Paul the Octopus” moment. That “we have the right model because we’ve been right a large number of times in the past, consecutively” – which is great, as long as (i) they have a reasonable story behind WHY the model is predictive and (ii) they made the calls before hand. But, they do take an interesting approach to the problem. Forget about polling and just use economic factors as the driver.

      I started pulling down the state polling data into a more presentable format / more useful response to your comment. The basic is this – state polls are demonstrably biased, unlike national polls. That is, we can point to a large number of instances where all of the state polls missed in the same direction – they all had a bias (sometimes significant). On the other hand, our few data points with national polls shows that they tend to miss both directions and follow something not-so-different from a normal distribution.

      Consider some 2008 state poll results (and I’ll put all the results in the post). Of the 11 current swing states, there were 58 state polls in the RCP final averages for those states. The standard deviation of those polling errors (predicted margin versus actual final state margin) was 3.49 points. When we take the RCP averages per state (that’s just over five polls a state on average – some more, some less) the new standard deviation of errors is 3.02. Now, if averaging 5 polls were just reducing noise we should have seen that 3.49 drop to something like 1.56. But it only came down to 3.02. Why? The state polls have state-level systemic biases that get washed over in the national polls.

      Here are a few instances:
      PA – each of the 7 polls missed “right” (called it more Republican than actual outcome)
      NV – each of the 6 polls missed “right”
      MI – each of the 4 polls missed “right”
      IA – each of the 3 polls missed “left” BADLY (4.5, 5.5, and 7.5 points)
      CO – each of the 4 polls missed “right” (closest was 2 points)

      These type of all-one-way misses point to severe weaknesses in the state polling data – from a predictive standpoint. In 2004 the results were quite similar. WI (4 polls), PA (6 polls), NC (4 polls), NH (4 polls), NV (3 polls) all missed in a party-bias direction. In the case of NV with 3 polls it was again the case that they all missed BADLY (4.8, 5.8, 7.8).

      My point is, again, that if the national polls get it about right (average is nearly spot on and distribution is clean about the actual result) then the national pollsters are getting the turnout behaviors and electorate sentiment right. The state level polls are not.

      The problem of course is that I’ve only pulled together the data for recent elections at the national level (or state level, for that matter). Would be useful to dig back into the past more. But, one suspects that polling is better now than it was then (easy out if the trend doesn’t hold).

      • “These type of all-one-way misses point to severe weaknesses in the state polling data – from a predictive standpoint.”

        How are *any* of these specific examples “severe weaknesses”? I may have not clearly expressed my point, because this seems to side-step that point entirely. In Maine or Nebraska, I can see how it might be useful to consider the accuracy of a state poll measured as a difference between predicted and actual percentage of vote for a candidate.(*)

        But in any other state, the important prediction is not the percentage, it’s the electoral vote allocation. And the EV is winner-take-all, so all that matters is whether the predicted percentage is a majority. And in every one of the quoted examples from 2008, both the average and even every single one of the individual polls predicted the outcome correctly. Put more simply, none of those “biased state polls” in the comment reply were wrong!

        (What a difference between predicted and true percentage *does* affect is the *confidence*, as opposed to the *accuracy*, of the prediction of the EV allocation. In your extreme case of Iowa, for example, I wouldn’t have lost a bet on Obama winning, but I would have been willing to take riskier odds than I should have.)

        (*) But even in these two special cases, I think it’s still “winner take all” by district, as opposed to allocating in proportion to state-wide “popular” vote, but I’m not sure, and not motivated to look it up :).

      • nomasir says:

        “And not motivated to look it up” … don’t get me started on the cool statistics that are worth a look but I just don’t have time to do it.

        I think your last parenthetical is where I’m coming from. The fact the polls “got it right” but were off on the margin is meaningful to me. In NFL terms it means that the spread was wrong (yes, we’re conflating a few things here, but bear with me). If the distribution of outcomes is actually random with 14 point sigma about the spread – BUT I can discern a bias in the spread, then I am more than willing to wager a little money for the improved probability of winning. Even if we go to the “moneyline” where a 9 point favorite has a 74% chance of winning. If I find out, with my impressive modeling skills (on the catwalk) that the actual spread should be 5 points – or a 64% chance the favorite wins – then I gladly bet the underdog. My expected return is 38%, which is huge, even though I’m betting the underdog. (Of course, I want a lot of trials to bring the noise down … and I’ve left out any “vigorish” in the payouts.)

        Similarly, if we are estimating probability that a state goes left of right based on an average of polls and a spread of uncertainty, then an indication that the uncertainty is larger than advertised is quite meaningful. Consider Ohio, for instance, where the current RCP average is Obama +1.9. If I believe that the uncertainty (1 sigma) around this is 2 points then I give Obama an 83% chance of winning. But if I believe the uncertainty is 4 points it comes down to a 68% chance – which is a betable proposition.

        OK, I’m meandering in rather trivial examples. My point is that the predictive value of the estimate is more than just “right/wrong” precisely because we have a rather small sample size. It would be more meaningful if the distribution of errors followed a discernible (random) behavior. BECAUSE we can point to quite a few instances of “all polls were wrong in the same direction” (and in some cases wrong by a lot) then a lead in the state poll even if taken as an average of polls, must be assumed to be drawn from a distribution with a large random bias. That the bias is unknown (and will continue to be) means we can’t call the state soundly this way or that – but we can say that the outcomes have greater uncertainty. If that greater uncertainty becomes big enough (as I believe it is) then other measures become useful as predictors. Which is where my “model” stands today. (I use “scare quotes” because there’s no actual code for the algorithm – but there is an algorithm.)

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s