This is a guest post by Michael S. Lewis-Beck, Wendell Miller Distinguished Professor, Department of Political Science, University of Iowa
Recently, different teams of political scientists who specialize in US election forecasting released their fall forecasts at the annual professional meeting. These forecasts have been discussed in the media, and have now appeared in print, including the forecast that Charles Tien and I made (see, respectively, Balz, 2016; Lewis-Beck and Tien, 2016). We forecast a win for Hillary Clinton, with 51.0% of the popular vote share (the Democratic and Republican total), which converts to an estimated Electoral College win of 274 votes. What forms the basis of this close, but winning, margin? Our Political Economy model points to two factors, presidential popularity and economic growth. The first variable, the popularity of president Obama, exercises a positive pull; the second variable, economic growth, exercises a positive pull, but much less so. In this essay, I will lay out the workings of the model, focusing on the power of the sitting president in this equation.
Our Political Economy model relies on enduring theory, as manifest in numerous scholarly investigations. This model was our first, appearing over thirty years ago and we have returned to it now, on evidence that it is, after all, our best effort ever (Lewis-Beck and Rice, 1984). The theory contends that the presidential election constitutes a referendum on how the party in the White House handles the big political and economic questions of the day. The better he (or she) does on the critical issues, the more votes go to the party (Lewis-Beck et al. 2008). We measure these two variables simply. For the president’s treatment of leading political issues, we utilize the Gallup pool question on job approval. For performance on economic issues, we utilize economic growth. Further, since we wish to make a forecast, from an optimal lead time, we measure these things in the mid-summer of the election year.
Put another way, the underlying explanation finds this expression:
Incumbent Vote = Presidential Approval + Economic Growth.
Of course, more complicated models are possible. (See, for example, Dassonnville and Lewis-Beck, 2015). However, this uncomplicated one appears to work rather well, drawing on estimates from the multiple regression equation below, applied to the 17 presidential elections, 1948-2012.
Vote = 37.50 + .26 Popularity + 1.17 Growth
Predicting the 2016 presidential election, we merely plug in the final numbers of these two variables, as of August 26, 2016. That is, July Popularity = 51% and Gross National Product Growth (2016, first two quarters, non-annualized) = .20, yielding this forecast:
Vote = 37.50 + .26 (51) + 1.17 (.20)
= 51.0 % of the popular two-party vote to Clinton, the Democratic candidate.
A popular vote share of this magnitude yields an Electoral College forecast of 274 (derived from the following formula:
Electoral Vote % = -198 + 4.88 Popular Vote).
How much error can we expect from such a prediction? Perhaps surprisingly, not that much. Looking at Table 1, we see the actual forecasting error from the prediction of each election in the series. It is important to note that these jackknife forecasts are out-of-sample, i.e., the election called is forecast from data gathered only on the other elections. Observe that the median error is just 2.0 percentage points. More generally, column 3 of the Table below reports that the forecasts correctly pick the winner in 13 out of the 17 contests, meaning it has been right 82 percent of the time.
What lies behind this level of accuracy? Mostly, it comes from the power of the presidential variable. That variable – public approval of how the president handles the job – correlates with vote at about .83 across the series, and accounts for the lion’s share of the variance in the multiple regression equation (where the R-squared = .76). Indeed, if we look at the presidential approval variable by itself, we can formulate an interesting rule: when the president’s mid-summer approval exceeds 50%, the party in the White House will win the national popular vote. (Curiously, this rule had an early, but rather neglected, emergence in the literature; Lewis-Beck and Rice, 1982). This rule has held for all 7 of the elections across this period where the 50% standard is met or exceeded, namely 1956, 1964, 1972, 1984, 1988, 1996, 2000.
A straightforward, theoretically sound Political Economy model suggests that Clinton will be the next president. The promise of that suggestion comes mostly from the relatively high popularity that President Obama, as a Democrat, has recently enjoyed. His comparative success on big issues appears to have allowed a majority of voters to transfer their support to the new candidate of his party, Hillary Clinton. When presidential elections are viewed as referenda on the performance of the party in the White House, that gives the man (or woman) occupying that office enormous de facto power to determine the next winner.
Balz, Dan. September 3, 2016. “Election Forecasters Try to Bring Some Order to a Chaotic Political Year,” Washington Post.
Lewis-Beck, Michael S. and Ruth Dassonneville. 2015.“Comparative Election Forecasting: Further Insights from Synthetic Models,” Electoral Studies, 39,2015, 275-283.
Lewis-Beck, Michael S., William Jacoby, Helmut Norpoth and Herbert Weisberg. 2008. The American Voter Revisited. Ann Arbor: University of Michigan Press.
Lewis-Beck, Michael S., and Tom Rice. 1982. “Presidential Popularity and Presidential Vote.” Public Opinion Quarterly, 46 (Winter), pp.534 – 537.
Lewis-Beck, Michael S., and Tom W. Rice. 1984. “Forecasting Presidential Elections: A Comparison of Naïve Models.” Political Behavior 6, 9-21.
Lewis-Beck, Michael S. and Charles Tien. 2016. “The Political Economy Model: 2016 US Election Forecasts,” PS: Political Science and Politics, October, 2016.