Tomorrow, America goes to the polls. Certainly, different voters will be motivated by different factors when choosing the candidates for whom they will vote. However, are there any single factors that dictate the overall direction of voter sentiment in an election?
In this article, I will make the argument that, during midterm elections such as this one, the one overriding factor in voter choice is the performance of the President. That is, midterm elections are essentially referenda on the President and his party—in this case, President Barack Obama and the Democratic Party.
The graph below plots the number of seats gained or lost in the Senate and in the House of Representatives by each President’s party in a midterm election against that President’s approval rating, as measured by Gallup, for all midterm elections from 1946 to 2006. Positive numbers represent a net gain of seats, while negative numbers indicate a net loss. Using standard least-squares linear regression, I have produced a best-fit line for each data set.
The least squares regression generates the equation S = 0.2009A – 14.195 for the Senate, where S is the gain or loss in Senate seats and A is the President’s approval rating at the time of the midterm election; and the equation H = 1.403A – 97.455 for the House, where H is the gain or loss in House seats and A is, again, the President’s approval rating on election day. The goodness of fit R^2 for these models are 0.2080 and 0.4639, respectively.
Although the fit is not especially good—actual midterm election results have deviated from the Senate model by as many as 9 seats, and from the House model by as many as 31 seats—one aspect of the two models suggests that the model is, in fact, reliable in principle, though highly imprecise in practice. Specifically, both regression lines cross the A-axis at nearly the same point. This improbable coincidence suggests that both trends, though weak, are indeed causally related to the same phenomenon; and that the threshold of Presidential approval that determines whether or not the voters choose to reward or check a President by increasing or decreasing his party’s strength in Congress, is practically the same for both the Senate and the House. These facts point to the conclusion that these models are highly accurate but also highly imprecise.
With the model explained, all that is left is to make predictions based on the models and test those predictions against reality when the actual election results are reported tomorrow night.
Obama is certainly below the threshold at which his party could have avoided a loss of seats in Congress. Specifically, the models predict that the Democratic Party will lose five seats in the Senate and thirty-four seats in the House of Representatives in tomorrow’s election. Given the historical margins of error for these models, the Democratic Party could actually lose as many as fourteen Senate seats and as many as sixty-five House seats. The real election results will almost certainly be located between the extremes. Most conventional analysts have predicted that the Democratic Party will lose control of the House of Representatives but retain control of the Senate; this prediction is within the margin of error of these models. Indeed, given the margin of error, it is more likely than not that the Democrats will in fact lose control of the House of Representatives tomorrow.
Another set of races that may be affected by the President’s approval rating are the gubernatorial elections being held in several states tomorrow. Below is a plot of the number of governorships lost by each President’s party versus that President’s approval rating, for every midterm election from 1978 to 2006. A linear model has been plotted, and is described by the equation G = 0.3924A – 23.943, where G is the net number of governorships gained or lost by each President’s party, and A is that President’s approval rating.
This model has a better fit that the previous two—the goodness of fit R^2 is 0.5266. The gubernatorial model predicts a net loss of six governorships by the Democratic Party in tomorrow’s election.
In science, hypotheses are constantly modified or rejected outright in response to new data. This study of poll data and election results is no different. Tomorrow’s election results will be added to the data sets, which will help produce more accurate models or, alternately, disprove the models and their underlying principles entirely. In the days following the election, I will post a complete analysis of tomorrow’s election results and the impact they have on these models.