Election ’08: Markets and Models

It may be surprising to learn that one of the leading scholars studying U.S. politics is in fact a Swedish economist. But the advantage of this unusual state of affairs is that during my recent trip to Stockholm, I had a chance to catch up with David Strömberg.

David and I spent an interesting afternoon exploring data from both political prediction markets, and his own econometric model for forecasting elections. The Strömberg model is, in my view, the leading quantitative election-forecasting model — both parsimonious and sophisticated. By building up forecasts state-by-state based on a slew of economic and political factors (details here), his model can not only predict who will win, but also the likelihood that each state will be pivotal. In turn, this makes David’s approach ideal for campaigns trying to figure out where to direct campaign resources.

We have written up our findings for my latest WSJ column.

The headline: Prediction markets currently give the Democrats a 63 percent chance of winning; the Stromberg model puts the odds at 65 percent, reflecting the fact that “the Democrats are likely to win 51 percent of the popular vote, and to win the Electoral College 294-244.”

The reason: “In turn, this Democratic edge reflects the combination of a weak economy and incumbency effects in the wake of an unpopular Bush presidency, which are likely to disadvantage Sen. McCain. Weak local conditions and liberal swings in the electorate suggest that Democrats are likely to do well in key states, such as Ohio, Pennsylvania, and Michigan. Prediction markets are less bullish about the Democrats’ chances in Florida.”

An insight: “For all the attention paid by pundits to opinion polls, our analysis suggests that, at this point in the election cycle, they simply aren’t very informative. As such, we give them very little weight. But as the election gets closer, polls … will warrant greater weight.”

What we ignore: Strömberg’s model “excludes those themes that have dominated recent media coverage: the personalities, age, race, and gender of the candidates, as well as their former preachers, and possible VP candidates. Vague concepts like ‘electability’ don’t figure into the equation.” Even so, the model matches prediction market prices pretty well, “suggesting that markets are more informed by economic and political fundamentals than broader campaign narratives.”

A nailbiter: “The Strömberg model also suggests that there is a one-in-four chance that the 2008 race will come down to the decisions of less than one-in-100 voters in just one state, a situation that would be similar to Florida in 2000, or Ohio in 2004.”

Where to look
? “Which states are likely to be decisive? Our list of battleground states are headed by the usual suspects [Florida, Ohio, Pennsylvania, and Michigan], but campaign strategists take note: Beyond the big four, a few surprises emerge. Colorado, Virginia, and California are next-most likely to determine the election.”

You can read the full article here.

Roland Kappe

I have to be the department's standard bearer here: While I like Strömberg's model for getting closer at the data-generating process, Helmut Norpoth's very parsimonious model so far did a good job at predicting the popular vote in every election since 1952 with the exception of JFK's narrow victory over Nixon. All this based mainly on the New Hampshire primary, and therefore allowing for a prediction months before election day. An excellent description can be found
Washington Post
">in his guest post at pollster.com and this Washington Post article.

The only problem this time around being that the prediction is virtually a tie, giving only a 50.1 percent lead to Senator Obama.


#8 charlie:
Actually, Florida doesn't look to be close at all this year; I have yet to see McCain's lead over Obama below double digits there. More likely are Ohio, Virginia, and Michigan.

Also, it's a little nitpicky, but a 1% swing in a state did NOT decide the 2004 election. 1% of Ohio voters picking Kerry instead of Bush would still have given Bush a win, though only by about a tenth of a percent.


A good question #5. I remember the Minnesota gubernatorial election of 1998. It was a 3 way race between two very well known MN political figures, Norm Coleman and Skip Humphrey, and the independent (Reform Party) former pro wrestler Jesse Ventura. Ventura was consistently polling in the teens in polls based on "likely voters". In the end, in large part thanks to a creative ad campaign, the 18-29 demo was brought out in massive numbers, and heavily leaned Ventura's way. He ended up narrowly winning with 37.0% of the vote, shocking all of the politicos who were laughing at his candidacy even the night of the vote.

Given that Obama's greatest appeal is to those populations who tend to be underrepresented among the voting population, high turnout could turn what looks like a close race in to an easy victory.

R. Vangala

This is interesting stuff, and seems to confirm what a number of political scientists have also concluded. See, for example, the "Time for a Change" model of Alan Abramowitz, which predicts that, "Regardless of who wins the Democratic and Republican nominations, 2008 will be a time-for-change presidential election. Based on President Bush's approval rating in June of 2007, the recent growth rate of the economy, and the fact that the Republican Party will have controlled the White House for eight years, the Democratic nominee would be predicted to win the national popular vote by a comfortable margin. For the Republican nominee to have a reasonable chance of winning the 2008 presidential election, there would have to be a dramatic improvement in President Bush's approval rating during the next [one month]."

Check it out here:


Kyle S

Justin, check out Nate Silver blogging at fivethirtyeight.com. He's developed a similar model to the Strömberg model and updates it often based on the latest poll data. Each poll is given a reliability score based on that poll's historical accuracy. Plus, Nate is pretty sharp at other forms of analysis - he works at Baseball Prospectus and developed their player projection model, PECOTA, which is the best in the business as far as those things go.

Brooks Gracie III

I wonder if the model takes into account the surge in registrations for new voters (prior to the primary, and after)? Obama has inspired a lot of young people to vote, when they otherwise wouldn't have bothered.


The Strömberg model was built to use only in the US election? Can I use the Strömberg model in other countries?


Well we have seen how useful the polls have been in the last election in 2006 and in many of the primaries. Using models like this is how you win football and basketball pools too. Forget the talking heads, polls, etc. Yes, sometimes they are right, but overtime, this is the way to go. After all, the polls told me Howard Dean in 2004 and Hillary versus Gulliani this time around.


lets not forget that 18 million Democrats did not vote for Barack Obama.

That's meaningless, millions of people don't vote for the winner in every party primary. John McCain had millions of Republican voters who didn't vote for him in the primary, but it's a far stretch to then assume they'll vote for the Democrat.

The is the only campaign season i can think of in my lifetime where people have seriously made the suggestion that because someone didn't win 100% of the primaries, they are somehow crippled.

Heck, if Romney hadn't dropped out so suddenly, we might well have seen him and McCain go on for months in the primaries, as neither one had particularly broad support among the party base.

Shawn K.

I am interested in these types of data-analysis, but it doesn't seem to take in account any variations of what people really think. I mean, for example, when polled 80% of Americans think the country is headed in the wrong direction, but when asked about their own personal well-being and economic well-being, the results are quite different than what the "80% wrong direction" poll would suggest.

Also, lets not forget that 18 million Democrats did not vote for Barack Obama. Likeability of candidates seems to have no effect on the data?


Note to Barack: Concentrate on Ohio and Florida. Send Check for advice.

Note to John: Concentrate on Ohio and Florida. Send Check for advice.

M. Carter

A good new model, but it's difficult to assess the
voting public's, er, emotional mood. One of our nomination races has been a celebrity smackdown, and the other a wierd process of elimination based on relative offense. That sort of thing doesn't always yield the most qualified candidates, eh?
Also, we need to recognize the probability of dirty work at the polls. I'm an election worker, and am grateful that New York wasn't hasty enough to buy those godforsaken touch-screen machines. Watch out for results where these are in use, as well as the aforementioned Ohio, Virginia, Michigan, and some others.
Have to say, when I look at what's coming up it gives me the cauld grues.

Keith M

For a pretty comprehensive breakdown, check out this Economist summary:


In short, whoever spends the most money will win (isn't that how it always works in America?). Since Obama has the most of any nominee in HISTORY, looks like the next prez will be a donkey, if not an ass.