North Carolina is a swing state that has relatively few swing voters. Rhode Island is not a swing state, but it has quite a lot of swing voters.
Are you confused yet?
The concept I’m getting at here is a pretty basic one, although it is sometimes hard to explain because the term “swing” is applied in such a wide variety of different ways to describe voter behavior. So it may be time to introduce a new term into the political lexicon, which I’ll call an elastic state in reference to a similar concept in economics.
Let’s define an elastic state as one that is relatively sensitive or responsive to changes in political conditions, such as a change in the national economic mood. (This is in the same way that, in economics, an elastic good is one for which demand is highly sensitive to changes in prices.)
For instance, if there are a series of strong jobs reports this summer, and President Obama’s standing improves by five percentage points nationwide, we’d expect his standing to improve by more than 5 points in an elastic state. This works both ways: if we went into another recession and Mr. Obama suffered a five-point decline in his popularity, he’d experience a larger decline in an elastic state.
An inelastic state, by contrast, is one which is relatively insensitive to these changes. In an inelastic state, a five-percentage-point change in the national environment might only affect Mr. Obama’s numbers by three percentage points instead.
Elastic states are those which have a lot of swing voters — that is, voters who could plausibly vote for either party’s candidate. A swing voter is very likely to be an independent voter, since registered Republicans and registered Democrats vote with their party at least 90 percent of the time in most presidential elections. The swing voter is also likely to be devoid of other characteristics that are very strong predictors of voting behavior. For instance, he is unlikely to be African-American, which very strongly predicts Democratic voting. And she is unlikely to be a Southern evangelical, which very strongly predicts Republican voting, at least recently.
The classic example of an elastic state is New Hampshire. It has a very high percentage of independents, and those voters are also independent-minded in practice. Almost all of New Hampshire’s voters are white, but very few of them are evangelicals, characteristics that roughly balance out (Mr. Obama won about 55 percent of the nonevangelical white vote in 2008).
A good example of an inelastic state is North Carolina. It has quite a few African-American voters, who are almost sure to vote for Mr. Obama. But it also has plenty of rural white Southerners, many of them evangelical conservatives, who almost certainly won’t. To a lesser extent, it also has some highly educated and very liberal white voters in the Research Triangle, who are also quite likely to be Obama voters. That doesn’t leave very many voters left over. North Carolina is a swing state (or at least it was in 2008), because the coalition of Democratic base voters was quite close in size to the coalition of Republican base voters. But it wasn’t a state with a lot of persuadable voters: it’s the kind of place where elections mostly boil down to turnout, and Mr. Obama — with his considerably stronger ground game — was able to edge out a win there in 2008.
We can be a bit more rigorous about this and quantify each state’s degree of elasticity. In 2008, I modeled this by comparing the changes in state polls to the national trendline at different points over time. This calculation figured into the way that our presidential forecasting model simulated the different outcomes that might occur in case the national environment changed again. (We will be introducing this year’s version of the model shortly.)
That isn’t likely to work so well this year, however, because the state polling data has been quite sparse so far and because there hasn’t been all that much variance in the national trend. (You can’t model the way a dependent variable responds to an independent variable if the independent variable stays the same.) So, I’m opting instead for what I think is a more robust approach.
This method gets back to the idea of swing voters. We can directly model, from the ground level up, how many swing voters there are in each state and how they might respond to changes in the electoral environment.
In particular, I’m using individual-level voter data from the 2008 exit polls. Between the national exit poll and the state-by-state surveys, there were about 100,000 voters interviewed.
Each of these voters was a real (albeit anonymous) human being with a set of real demographic characteristics. Because the exit polls asked quite a few demographic questions and because the very large sample size permits us to be quite inclusive in selecting variables, we thus know something about each voter’s race, religion, gender, age, party identification and so forth, as well as which state the voter lived in.
Through logistic regression, we can estimate the chance that each of these voters would vote for Mr. Obama based on their demographic characteristics. For example, a 66-year-old Latina Catholic living on a fixed income and who describes herself as a conservative Democrat might have about a 75 percent chance of voting for Mr. Obama, according to the model. Or a high-income white independent moderate voter who has an advanced academic degree and lives in a suburb in the South might have about a 45 percent chance of doing so.
The regression model also contains a constant term common to all voters which can be changed in order to reflect a shift in Mr. Obama’s overall standing among all voters. If there is a positive development in the economy, for instance, and Mr. Obama goes from having 50 percent of the vote nationwide to 55 percent, we can shift the constant term in order to reflect the overall change in Mr. Obama’s standing, and use this to back into an estimate of how that change plays out at the level of individual voters and individual states.
The key concept is that a voter who was close to 50 percent likely to vote for Mr. Obama to begin with is more likely to switch his or her vote than one whose vote could be determined with a high degree of certainty. It helps to imagine each voter as having a scale. For a swing voter, the scale is close to evenly balanced, so it won’t take much new information to tip the voter in a new direction. Conversely, those voters who already have a disproportionate amount of weight on their scale either for or against Mr. Obama aren’t likely to have their decision changed by adding a little bit more weight on either side. (This can also be seen logically: someone who was already almost 100 percent certain to vote for Mr. Obama can’t really become any more likely to do so if something favorable happens to his campaign.)
By summing up the behavior of the individual voters, we can then see how these numbers affect behavior at the state
The numbers you see below reflect my estimate of the elasticity of each state and the District of Columbia. If a state has an elasticity of (for example) 1.1 points, as Wisconsin does, that means a one-percentage-point change in the national numbers would be expected to change the Wisconsin numbers by 1.1 points. Or, likewise, a five-point change in the national numbers would change that state’s voting preferences by 5.5 points.
The more elastic states are those in the left-hand column of the chart. Some of them, like New Hampshire, aren’t so surprising. But others like Rhode Island, which actually tops the list, may be less expected.
What’s going on in Rhode Island? That state has a ton of independent voters, who are similar in many respects to the voters in New Hampshire although Rhode Island is more downscale economically.
However, among those Rhode Island voters who aren’t swing voters, many more are Democrats than Republicans. That is, Rhode Island has a lot of swing voters and a lot of Democratic base voters, but very few Republican base voters.
In recent presidential elections, these Rhode Island swing voters haven’t been persuaded by the conservative options offered by Republicans. In these cases, where many or most of the swing voters side with Democratic base voters, the Democrat will win by a wide margin.
However, if the swing voters have a Republican option that is more suitable for them, they may break from the Democrats and give the Republican a narrow victory. This can be seen in gubernatorial elections, for instance, where the Republican or the independent candidate is selected locally and may be a better fit for Rhode Island moderates. In fact, Rhode Island hasn’t elected a Democratic governor since 1990.
In fact, most of the states that rate as having high elasticity have this quality of sometimes delivering a surprising outcome. Alaska, for instance, is sort of the polar opposite of Rhode Island — it has quite a lot of independent voters, but the Republican candidate usually wins because the Republican base there is larger than the Democratic base. But Alaskan voters are flexible enough to keep Republicans on their toes. A relatively liberal Democrat, Mark Begich, was narrowly elected to the Senate in Alaska in 2008 because Republicans had a problematic incumbent in Ted Stevens and because the national environment was very favorable for Democrats that year.
Massachusetts rates as having high elasticity, which helps to explain how Scott Brown won a Senate seat there in 2010. So does West Virginia, which explains why Barack Obama is very unpopular there but why Joe Manchin III, also a Democrat but a conservative one, was fairly easily elected to the Senate in 2010.
The states that rate as having low elasticity produce more predictable results. Georgia is a low-elasticity state, for instance. Although its demographics aren’t significantly imbalanced, they are also relatively rigid and the Republican coalition is a little larger than the Democratic one. Thus, Democrats swung and missed when they tried to pick up a Senate seat in Georgia in 2008. New York, meanwhile, which has far more ethnic minorities than New England, rates as having low elasticity, and Republicans made very little headway there in 2010 despite having a strong year otherwise.
As should be clear from examples like Rhode Island and Alaska, a state can have a strong partisan lean and nevertheless be elastic, or vice versa. Both partisanship and elasticity are useful in considering how a state might react to changes in the electoral environment. I don’t want to overdo the elasticity concept too much — for example, the most elastic states, like New Hampshire and Rhode Island, might be 20 or 30 percent more responsive to changes in national trends than the average state, but not 200 percent or 300 percent more responsive. Elasticity, however, does help to explain why the results seem to be more variable in some states than others, both within a given election and between different types of elections.
The states that are traditionally thought of as swing states — meaning that they are close to the national average in their partisan orientation — might be broken down into different groups based on their elasticity.
Persuasion swing states are states that are close to the national average in terms of partisanship and are also highly elastic. New Hampshire, as I mentioned, is a prominent example. So is Iowa, another state with a lot of white moderates. Western states like Colorado and New Mexico often fit into this category. Arizona does as well, which is why (despite my earlier protestations) it might be a more plausible pickup for Mr. Obama than a less elastic state like Georgia. Persuasion swing states will be relatively more responsive to the effects of the campaign and changes in the national environment, including economic conditions.
Turnout-battle swing states, meanwhile, are those like North Carolina with relatively evenly-matched electoral coalitions but few persuadable voters in the middle. Pennsylvania is a classic example: can Democrats turn out enough voters in Philadelphia and Pittsburgh to overcome the conservative central portion of the state? It’s pretty much the same question in Pennsylvania every cycle and we often get about the same answer. Virginia, whose demographics are similar to North Carolina, is likewise more of a turnout-battle state. That is not to say there are no swing voters there — Virginia has some in its northern suburbs, for instance — but there just aren’t as many as a proportion of the electorate as somewhere like New Hampshire.
States like Ohio and Florida, finally, are swing states that have about average elasticity. In these cases, the Democratic base, the Republican base and the number of swing voters are all close to the national average.
We’ll be using this individual-level voter data in other ways as we roll out the general election model. For instance, it can be used to estimate the effects on the electoral math if there are “shocks” that affect different demographic groups individually rather than felt by all voters equally. Some electoral maps are intrinsically more plausible than others because similar states like Michigan and Ohio are more likely to move in the same direction than those like Virginia and New Mexico that have few demographic properties in common.