June 14, 2007
Mathematical Formula Predicts Romney Win?
This is the second installment of Justin Hart’s series on campaigns from www.mymanmitt.com.
In our continuing quest to predict the outcome of the 2008 race I submit the second in a series of posts examining how the academic world perceives primary elections.
Last week I examined recent computer models predicting that the GOP nominee will be known earlier than the Democratic nominee because of the type of primaries utilized by each party. I surmised that this phenomenon currently bodes well for Mitt Romney who is leading in New Hampshire and Iowa.
This week we examine momentum in more specific terms
—————————————————————————–
Momentum is a powerful thing. Will the traditional model of front-loading wins in primary states, the so-called New Hampshire effect, hold true for 2008?
In his 1983 study, M. Malbin reported that George Bush and Ronald Reagan allocated ? of their respective total 1980 campaign budgets to early primary states. A 1987 study reported that the New Hampshire election grabbed 20% of the 1984 season’s news coverage. You might also recall Howard Dean’s burnout because he spent so much money up front.
This pattern has been well established for over two decades. In short, early primary states have historically garnered a disproportionate amount of attention. But for good reason.
Tilman Klumpp and Mattius Polbor in their 2005 paper “Primaries and the New Hampshire Effect” (pdf file) describe this interesting phenomenon: “The outcome of the very first primary election creates an asymmetry between ex-ante symmetric candidates which endogenously facilitates momentum in later districts.”
Endogenous is a great Scrabble word meaning: “no apparent external cause.” In other words, the momentum factor of early primary wins is a very real and hard rule of elections.
They go on to note:
Although our analysis compares mainly two extreme cases completely sequential elections versus completely simultaneous elections , the distinct results of the sequential case basically apply to a mixed temporal structure as well, as long as it involves some sequential elements at the early stages. One can argue that such an intermediate system is closer to the modern primary races, in which there are dates (such as “Super Tuesday”) when several states vote simultaneously.
Nevertheless, even in this case, some primary states vote in sequence at the very beginning of the nomination process. We show that this is enough to generate (and sometimes even amplify) the momentum effect and the spending pattern that arise in a completely sequential system.
Put a babel fish in your ear and read that paragraph out loud: “Regardless of the adjective you use to describe that infamous Tuesday, as long as New Hampshire and Iowa come up first, a candidate has a better than even chance to build some serious kick-butt momentum.”
Another 1005 study by L. Keele was more explicit:
We find that, while front-loading has significantly shortened the primary season, it has not altered the effect of finishing strong in Iowa and New Hampshire.
But for Klumpp and Polbor this is not just fun theoretical musing. The goal of their paper is to recommend campaign expenditure models. At one point they examine the comparison to sporting tournaments:
[I]n sporting tournaments it is often desirable to induce contestants to spend a maximal amount of effort, or to induce an effort allocation that increases the chance of a close contest, as this enhances the excitement level the tournament generates. For primary elections, on the other hand, we are interested in finding a design that minimizes wasteful campaign expenditures and avoids long, close battles as these will be very costly.
Channeling the Powell doctrine this adds up to: “use overwhelming force to win before it even gets close.”
OK now this might seem absurd (it is academia after all) but Klumpp and Polbor go on to identify an actually mathematical formula to predict how much money and effort must be expended in early primaries to win outright. Here’s part of their equation:
I know. That’s sick and wrong. Calling John Derbyshire!
They bring the model back to earth with a pretty straight forward table showing who won New Hampshire and what percentage of primaries they won in Feb/Mar and Apr/May.
They conclude from this table:
One implication of our model is that winning the first primary makes it more likely to win the nomination. Out of the ten races in Table 4, six were such that the winner of the New Hampshire primary was the eventual nominee. Although this seems hardly indicative of the existence of momentum effects described in this paper, one needs to keep in mind that in most of these races there were more than two candidates in the New Hampshire primary.
Their overall conclusion is also striking:
The winners of early districts is endogenously more likely to win later districts than the loser, not because voters react to performance in previous elections, but rather because of equilibrium candidate spending behavior. In addition to reproducing these stylized facts from primary races, our model also provides a rationale for why political parties have chosen a sequential organization of primary elections: First, it induces lower expected expenditures and higher expected rents than a simultaneous structure. Moreover, if one candidate has an ex-ante advantage over the other, either in terms of campaign effectiveness or in the number of assured districts, a sequential organization selects the stronger candidate with probability close to one, provided there are sufficiently many districts.
Now, I think you know what conclusion I’ll draw from this. Let’s just say, the guy currently spending tons of dough in NH and IA and earning double-digit margins in return will likely win the nomination: Mitt Romney.
(Please note, lest you think me outright biased, there are a few of characteristics coming up that favor candidates other than Mitt)
More to follow…
http://race42008.com/2007/06/14/mathematical-formula-predicts-romney-win/trackback/
June 14th, 2007 at 9:00 am
Mitt Romney = Howard Dean
Fred Thompson = Wes Clark
June 14th, 2007 at 9:24 am
So tell me, how do the writers of this paper deal with the obvious factors that skew the results. In effect, it treats victory in the primaries as an outcome that, apart from early victories (themselves seen as a product of “spending”), is essentially random.
It’s true that the home team wins the majority of basketball playoff series. It’s also true that the home team is almost always the team with the best record (a few seeding irregularities), usually confirming that it is in fact the better team. So the apparent home team advantage is in this case both a “home team” advantage and a “better team” advantage.
The better candidate will tend to have more money and tend to do well in elections, early or late. In addition, there are many other exogenous factors influencing or potentially influencing specific race outcomes and eventual outcomes. Finally, there are far too few data points either presented or EVER existent to establish a meaningful statistical probability model. If you accept that early victories are not wholly determinative, then, even if you are willing to make the huge assumption that other factors are not highly influential where not much more important, you have to accept that the relationship is, at strongest, probabilistic, and are left with only a handful of instances withi which to determine that probability.
It’s almost like a flipping a coin and determining that, on the basis of having hit heads 6 times and tails 3, that the odds of hitting heads are 2 to 1.
June 14th, 2007 at 9:36 am
Let me cite another paragraph from the Keele paper:
“We see that for a candidate that finishes outside the top three (in NH) the risk of ending their campaign is 85% higher. Alternatively, each percentage of the vote one picks up in New Hampshire reduces the risk of failure by 8%. The effect of Iowa is less clear cut. While finishing in the top three implies that the risk of failure is 67% lower, the estimate for the percentage of votes in Iowa is not estimated with enough precision to draw any inferences about its affect.”
This is hardly coin-flipping material. In pure historical terms there are statistical trends that speak volumes. I agree, this is not HARD science… but it is a branch of science.
June 14th, 2007 at 9:47 am
I’m still reading through the paper but so-far I’ve got to mention the paper assumes rationality on the part of actors (politicians, voters) that is not just a simplification but distortion of actual human behavior.
The method of rent-seeking analysis is quite ingenious though. I’ll have more later.
June 14th, 2007 at 10:03 am
From the paper:
“A sequential organization basically requires that candidates can observe the outcome of an election before they commit resources to further campaigns. If elections are separated by short time intervals only, it seems unlikely that this is still a practical possibility.”
This certainly seems possible.
Also more substantially Klumpp and Polborn find momentum as an endogenous factor (aka determined within the system). The endogeneity is based in rational candidate decision making processes based on game theory analysis.
The writers do not analyze the appropriateness of this theory (as they are unable to) but it seems to be at best a minority of the explanation for “momentum”.
In highly discrete long primary seasons such an effect would be large (thru candidate and fund raiser decision making). However in the modern primary season (since 1976) media effects have to be a predominant element of momentum.
Given that I’d suggest winning earlier caucuses/primaries has a small benefit regardless of media coverage. However the majority of the “benefit” of winning early is media based.
This means that the other candidates can play spoiler if they ruin a candidate’s media coverage by reducing the significance of early races.
June 14th, 2007 at 10:16 am
Econ,
I don’t understand you.
June 14th, 2007 at 10:19 am
Which part. I’ll try to be less technical.
June 14th, 2007 at 10:29 am
I think I got the last 4 paragarphs. But dumb it down for me. You sound good.
June 14th, 2007 at 10:43 am
Ok, from the quote above, the writers admit that primaries/caucuses have to be far enough apart so candidates can make decisions about how much they’ll spend (in campaign time or money) in the next race. So squishing all the primaries together eliminates the momentum effect they describe.
Also the writers assume that “momentum” is just the logical result of candidates making ‘good’ decisions with how to spend their time or money (losing candidates have less to gain from winning a single race the further they get behind).
The writers don’t actually try to prove their theory so the whole article is “Hey we’re guessing how momentum might work”.
It’s interesting stuff but it shouldn’t be used to predict anything.
What I find extremely weak is the idea that momentum isn’t caused by the media. While candidates would get an extra boost by winning even without media coverage, most of the momentum is caused by media coverage and its effect on voters.
The article is built around a concept that is probably a small part of how momentum in campaigns actually works.
One thing I’d like to add though is that this article has a lot to do with the Iowa Straw poll because of how much earlier it is than any other ‘race’. It would have had an important impact in boosting Romney and hurting the other guys. That’s why Rudy and McCain pulled out.
Hopefully that’s a little bit clearer.
June 14th, 2007 at 10:49 am
It is pretty tough to formulate an equation for any human behavior. It is fun to see them attempt such a thing.
Also,
I can’t ever for the life of me picture Romney shrieking as Dean did. Romney does not equal Howard Dean. Please give the man some as least some credit. He’s certainly smarter, competent, and in control a great deal more than Dean..
June 14th, 2007 at 11:00 am
cwpete,
Back when I was an undergraduate I formed an equation for Presidential general elections that correctly predicted the popular vote winner for each election since 1892. (Caveat: it was a regression equation that if I wrote out would be as long as 4 pages of single space type)
It’s not too hard to formulate an equation for human behavior, it’s just that equations are limited to what we can observe.
June 14th, 2007 at 11:06 am
I’d say it a non sequitur, and, what I mean by that, it’s a null set.
June 14th, 2007 at 11:15 am
uhhhh…. ok
June 14th, 2007 at 11:26 am
ROFLMAO. KevinP you get the quote of the day award.
June 14th, 2007 at 11:57 am
This is basically just a simple media analysis in window dressing.
Ignore the charts and graphs. This says: “Winning Iowa and New Hampshire is important!” That’s nothing new, everyone!
CK is wrong in #2 though, with his coin-flipping comparison. Coin-flipping is a random event; drawing some sort of prediction about which way it’s going to flip in any given time is a fallacy because of that. However, primaries are not random events; they are determined by the volition of individuals. In perfect circumstances, flipping a coin is not (let’s assume a virtual model for coin-flipping).
Either way, the model ignores the new factors — National Primary Day, mega-millions in fundraising like we’ve never seen before, etc.
The conclusion that this SHOULD HAVE reached is: “If someone wins New Hampshire, there is a sixty percent chance that they will become the nominee.” But even that’s incorrect, with the new factors, and where does Iowa play into this? And the competition is different throughout the different primaries, and –
Let’s just say this is basically a useless graph.
June 14th, 2007 at 12:02 pm
Nick #14: Don’t get me wrong - I think Romney should be the next President - with or without this type of analysis.
June 14th, 2007 at 12:23 pm
If anyone believes this, they can quintuple their money on Intrade, as Romney is trading at 20.
June 14th, 2007 at 12:54 pm
Kevin, I’m a hard-core Romney supporter, and hope he is the next President as well. I just thought your comment was hilarious because of all the attention it got on the blogosphere after the debate.
June 14th, 2007 at 3:13 pm
TLG#15: The point of the coin flipping analogy was to stress that you can’t assess probabilities based on such a limited number of data points, not that election victories are random or that campaigns are like coin tosses.