Comparing 2008 to Other Bad Years

According to our model indicator of economic growth (GNP percentage change over the first six months of the election year), the worst years in the series are 1952, 1956, and 1980. Suppose we conduct out-of-sample forecasts for each of these contests (estimating the incumbent vote share on the basis of the model observations minus the year forecast). The errors are small (1952 = −0.52; 1956 = 2.67; 1980 = −0.86) when compared to the 2008 error of 3.1 (i.e., forecast = 43.4%; actual = 46.5%). Consider 1980, which may be the election season most comparable to 2008, in terms of political and economic woe under the ruling party. Going into that election, the country faced stagflation, high oil prices, and a dramatic hostage situation abroad. The 1980 incumbent party candidate, Jimmy Carter, got 44.7% of the vote; the 2008 incumbent party candidate, John McCain, got more, 46.5%

Looking at the within-sample residuals, one sees a similar pattern: 1952, 1956, and 1980 are closely predicted by the model (i.e., −0.38, 1.22, and −0.17, respectively). Also, the model generally does well for the worst years according to our other economic variable, job creation. With respect to job creation, the worst years for growth are 1992, 2004, and 2008. The former two yield modest errors, respectively, of −0.94 and −0.67. However, for 2008 the error is 2.33, the biggest residual in the series. In 2008, the economic variables—GNP and job creation—did not deliver with their usual force.

Bailout Effects, Blunted by Race?

Clearly, Obama benefited from the economic voting the bailout engendered. We would like to formally incorporate this effect into the model, suggesting at the same time that economic voting was diminished by the role of race. The model, like most other statistical forecasting models, relies on measures taken in the summer. As long as the subsequent campaign operates within usual bounds, this measurement assumption poses little difficulty. However, when a major shock occurs post-measurement, as happened with the autumn economic and leadership crisis, it behooves the sensible forecaster to consider its impact. Therefore, as an exercise, we use more current estimates for the model's independent variables, namely Jobs (from January 2005 through October 2008) = 4.95; GNP × E (from second quarter data updated October 30) = 0.37; Incumbency Advantage = 0 (as before); presidential approval (for October) = 25. Plugging these independent variable values into the Jobs Model (Lewis-Beck and Tien Reference Lewis-Beck and Tien2008, 688) yields a McCain forecast of 41.3% and an Obama forecast of 58.7%. This forecast, even more so than our original, bespeaks an Obama landslide. While his victory was comfortable, it was not a landslide. We argue that the expected landslide did not materialize, because a portion of the electorate could not bring itself to vote for a black candidate. What portion? In our paper, we estimated that number, on net, at 11.5%. If we apply that correction to this current Obama forecast, we get 58.7 × 0.885 = 51.9 %. This estimate is very close to the two-party popular vote share that candidate Obama won.