G.O.P. Has 2-in-3 Chance of Taking House, Model Forecasts

In one sense, a strong performance by the Republicans on Nov. 2 is to be expected. The opposition party typically gains seats – on average, about 20 in midterm elections since 1994 – after the other party wins the White House, as the Democrats did in 2008. Nevertheless, both the magnitude of the Republicans’ potential gains, and the rapidity with which the political balance is poised to shift back to them after two cycles in which Democrats won nearly every competitive election, is unusual by recent standards. According to the model, Republicans have about a one-in-three chance of winning at least 54 seats, their total in 1994, and nearly a one-in-four chance of gaining at least 60.

Were the Republicans to achieve an outcome like that, one might need to look to the first half of the previous century for precedent; in 1948, for example, the Democrats added a net of 75 seats in the House, just two years after losing 54.

However, the headlines for this election have not yet been written, and there is considerable uncertainty in the outcome.

Several important national indicators, like the generic Congressional ballot, favor the Republicans. According to our analysis, Republicans have a lead of about 8 points on the generic ballot on the basis of likely voter models, which are generally more reliable in midterm elections than polls of registered voters (or of all adults). Meanwhile, President Obama’s approval ratings have been on a slow but steady decline for most of the 20 months of his tenure, particularly on the critical issue of the economy. On the basis of indicators like these, it is easy to envision Republican gains of 50 to 60 seats, or perhaps more, as some academic models that rely on these indicators do.

But FiveThirtyEight’s prediction model does not look solely at these national indicators. Instead, it evaluates the outcome in each of the 435 Congressional districts in which voters will cast ballots this fall. Although the generic ballot has some predictive power even at the local level, our analysis has found that it is not any more important – and in many circumstances, less important – than a series of metrics that can be applied to individual districts. For example, race ratings published by expert forecasters – and particularly those produced by Cook Political and CQ Politics – have an outstanding track record and are used extensively in our model. These forecasts generally point to a somewhat milder outcome for Democrats – perhaps a loss of about 40 seats.

The model also makes use of polls of individual House districts, where they are available. Because nonpartisan polls of individual races are relatively rare, this model includes polls commissioned by campaigns, or by other explicitly partisan organizations (although such polls are subject to a significant adjustment that anticipates their potential bias). However, in contrast with our other forecasting products, our House model gives no particular deference to the polls, and instead they are but one of the series of indicators that we evaluate. Polling in individual House races can often be unreliable, as many voters don’t familiarize themselves with the candidates until late in the election cycle, and as the demographics of individual Congressional districts can be idiosyncratic as compared with larger regions like states.

And this year, there are often significant disagreements – sometimes, even double-digit gaps – among pollsters who are surveying the same House races. The polls can nevertheless have some value when analyzed carefully. But, except in the rare case where there are several trustworthy polls available in the same House district, it is best to combine them with other indicators, our analysis finds.

The model also looks at the amount of fund-raising by each candidate – in particular, at the amount raised from individual contributors, which can be a leading indicator of both the enthusiasm for a particular candidate and the strength of his or her organizational infrastructure. This fund-raising data can have quite a lot of predictive power, although it diminishes somewhat as the campaign evolves (if the candidate is not successful at using their monetary resources to persuade individual voters). Democratic candidates for the House had a very slight edge in individual contributions through the Federal Election Commission’s June reporting period, although it may reverse itself when third-quarter fund-raising reports are released this month.

Finally, the model evaluates past voting in the Congressional district, both in presidential campaigns and prior House elections. But such data needs to be interpreted carefully: past voting is much less meaningful once an incumbent retires, for instance, or if the candidate had not faced serious opposition from the other party in the last election. And voting follows slightly different patterns for the presidency and for Congress: poorer districts sometimes split their ticket between a Congressional Democrat and a Republican president, while the reverse is sometimes true for wealthier ones.

In essence, then, the model takes a “kitchen sink” approach, applying as much data to the problem as possible. Its reason for doing so is utilitarian: such a blend would have performed more reliably in forecasting the outcome of past House elections than any one or two indicators would have on their own. The model, therefore, represents a compromise of sorts between the district-by-district analysis provided by experts like Cook, and the macro-level models sometimes published by political scientists.

But sometimes, the tension between the two approaches shows through. For example, most of the districts that Cook regards as “tossups,” the FiveThirtyEight model in fact regards as leaning toward the Republicans. On the other hand, it is also the case that even if the Republicans are likely to enjoy substantial gains over all, the outcome is hardly so certain in any individual Congressional district. Although the model regards almost three dozen Democratic incumbents as being underdogs to retain their seats, for instance, it is by a slim margin in almost all cases.

On average, the model predicts a net gain of 45 to 50 seats for Republicans, which if achieved would put the G.O.P in the majority. But there are 90 districts, the model finds, in which either party still has at least a 10 percent chance of prevailing, and it is on a district-by-district basis that each party will learn of their fate on Nov. 2.

Click here to read the full story.