About this Blog
Email Me

RealClearPolitics HorseRaceBlog

By Jay Cost

« The GOP's 2012 Problem? | HorseRaceBlog Home Page | Charlie Crist's Foolish Move »

Predicting the 2010 Midterm Election Results

Everybody wants to know what will happen in November. Here's my take:

I don't really know.

Predicting congressional midterms is a severely problematic task. It's beset with many difficulties, few of which are regularly discussed in popular commentary. Yet I've grown increasingly aware of them in recent years, and I find them to be utterly stultifying.

This article will lay out what I see as the biggest problems in trying to predict what will happen.

First, and most important, there are few observations. Since the end of World War II, there have been only 16 midterm elections. This is a major empirical problem. Expectations for future events must be built on past experiences. That's just common sense. The fewer past experiences you have, the less precise your prediction can be.

Second, congressional midterm elections have a very irregular pattern. The following chart tracks the share of House seats the party of the President has lost since 1946.

Midterm Losses.jpg

More than half of the elections are pretty uneventful. The party of the President has lost 10% or less of its House seats in 9 of the 16 midterms. That would be akin to Democratic losses of 26 or fewer seats this November, which I think most people would identify as a Republican underperformance.

So, if we think that the Republicans are going to make big gains, and we want to know how big - we only have about seven helpful elections.

But take a closer look at those seven elections. They are 1946, 1958, 1966, 1974, 1982, 1994, and 2006. This is a problematic list, and it points to the third problem. Election results necessarily depend on the social, economic, and political context of the day. As the context changes, so also will the results. Many contextual changes in American politics might have had a systematic influence on House elections. Here are eight notable possibilities:

-The party coalitions have flipped. In 1946, the Republicans were a Northern-based party and the Democrats a Southern-based one. Today, it is exactly the opposite.

-In 1946, about 12% of voting-age blacks were registered to vote. By 1970, that number had climbed to 67%. In response, politicians have put black voters in minority-majority congressional districts, which did not exist for the early elections.

- Salient issues have changed completely. Dominant issues in 1946 included labor union strikes and the emerging Soviet threat. How relevant can that be for understanding 2010?

-The contemporary congressional campaign is totally different. Nowadays, we have candidate-centered contests dominated by consultants, television advertisements, and the need to raise at least $1 million. This is a relatively recent phenomenon.

-There has been increasing party polarization. In the 1950s through 1970s, scholars thought the electorate was in the midst of a "dealignment." For example, the electorate of the past was much more likely to split its vote between the parties than the electorate of today. Also, congressional parties of the past were substantially more moderate than the contemporary ones.

-Today, gerrymandering is the one true political science. Politicians use computer-generated models to go precinct-by-precinct to find the right mix of voters to help their party win as many districts as possible. Such precision did not exist in the 1940s.

-For decades after the Great Depression, the Roosevelt coalition was broad enough that the Republicans generally could not win a House majority, even in good years. In particular, it was the "Solid South" that kept the Republicans at a ceiling of about 190 seats. This phenomenon lasted until the 1990s when the Republicans finally achieved parity south of the Mason-Dixon Line. For the first time since Reconstruction, both parties are now truly able to compete in every region of the country.

-It was not until the Warren Court decisions of the 1960s that states had to draw congressional districts according to the maxim of "one man, one vote."

Congressional elections depend not just on the national vote, but how that vote is distributed across the 435 congressional districts. Each of these eight factors could have influenced the vote or its distribution, and several could have influenced both. So, color me skeptical of any model (statistical or otherwise) that uses elections like 1946, 1958 or 1966 to predict the outcome in 2010. There are just too many changes in context for those elections to be terribly useful.

As if all this is not enough, there's a fourth problem. Three of the four recent big midterm elections - 1974, 1982, and 1994 - are extraordinary:

-The midterm of 1974 is far too unique to be of much use. Richard Nixon had resigned in August of that year and Gerald Ford had pardoned him in September. Any attempts to compare 1974 to 2010 are thus highly problematic.

-The midterm of 1982 saw the Republicans lose about 13% of their House caucus. At the time analysts were surprised; they expected the GOP to lose much more. Ronald Reagan's net job approval rating was at -6% by Election Day. The unemployment rate was at 10.8%. The Republicans had as many seats in 1982 as they did in 1974, yet they lost half as many in the midterm.

-The midterm of 1994 was surprising for opposite reasons. Bill Clinton's net job approval was at +1% on Election Day. By November, the unemployment rate had fallen to 5.6%, it's lowest point in more than four years. Virtually nobody expected a Republican takeover of Congress, let alone a takeover with so many seats to spare.

The 1994 midterm ended up scrambling most of the models at the time. Predictions were actually quite wide of the mark. A few years later, John Coleman of the University of Wisconsin would take to the pages of the Journal of Politics to ask if Republicans enjoyed some kind of systematic advantage in midterm elections. He wrote:

Few observers expected the massive Democratic defeat in the 1994 House election. In 1982 observers were surprised by how few seats the Republicans lost. These two examples suggest the possibility of a wider phenomenon: Republicans are relatively advantaged in midterm elections... These contrasting party fates may be related to different expectations voters bring to Republican and Democratic presidencies. Bringing party into midterm forecasting shows that the 54 seats lost in 1994 were not surprising for the Democrats, but under similar conditions the Republicans would lose only about 20 seats.

Coleman's argument has theoretical merit, but it runs into the same data problem as we've been discussing: too few observations for really solid testing. What I find most interesting about this assertion is that, after the waves of 1974, 1982, and 1994, electoral prognosticators were not sharpening their models, but revisiting core assumptions!

To summarize, here is where we are. In the last sixty years there have been only a handful of midterm elections, and even fewer can be considered waves. The midterms before the 1970s were so different in terms of context it's hard to derive much useful information from them. The midterm of 1974 was the product of a singular event in American history. The midterms of 1982 and 1994 caught everybody off guard. And the midterm of 2006 might not be relevant because the Republicans are different than the Democrats.

So, I don't really know what will happen. Are the Democrats going to lose seats in November? Almost certainly. Are they going to lose "a lot" of seats? Probably - the macro trends generally point in that direction, and we do know enough to get an approximate sense of how things will play out. Is it going to be seat losses in the 20-40 seat range, 40-60 range, or 60+ range? Nobody knows yet.

I understand that people want to see into the future. Lord knows I do. Analysts are already working hard to satisfy this demand by proffering all sorts of arguments. Their assertions are useful - but only to a point. I've come to believe that congressional elections are a poorly understood phenomenon, and I would encourage you to be a cautious consumer of predictive punditry.

-Jay Cost