Some saw the 2016 election as an upset but many indicators hinted at a Republican win.
The 2016 presidential election was not a good year for pollsters. Most anticipated a strong Clinton win in both the popular and the Electoral College vote, precipitating a lot of subsequent discussion about why polls performed so poorly in this election. There are a number of possible reasons. First, polling needs random samples of people who will vote, and this may be getting harder to do. Second, in this election people may have been reluctant to give their true views. Third, people’s views may simply have been unstable over time. Whatever the case, polls did not do that well.
The 2016 presidential election was not a good year for pollsters.
An alternative approach to predicting presidential elections is to look at the fundamental forces driving people’s voting behavior. This is the approach I take in my election work. The approach does not take polling data into account. It does not ask people how they are going to vote, but tries to explain what motivates them to vote the way they do.
Using econometric methods and data extending back to 1916, I have found four conditions that affect votes for president. The first is whether the president is running again. If so, this has a positive effect on votes for the president. The second is how long a party has controlled the White House. Voters like change; when a party has been in power for two or more consecutive terms, this has a negative effect on votes for that party’s candidate. The third is the slight but persistent bias in favor of the Republican Party.
Finally, the state of the economy is a significant factor: A good economy at the time of the election has a positive effect on votes for the incumbent party candidate. The economic variables that matter most are the rate of growth of output (GDP) and the inflation rate. Of particular importance is GDP growth in the first three quarters of the election year, although there is evidence that people take into account the entire four years of an administration.
Of these conditions, the first three were working against the Democrats in 2016. The president was not running; the Democrats had been in power for two terms; and there is the lingering Republican bias. According to the equation behind my work, then, the economy had to be fairly strong to give the Democrats a good chance of winning. Inflation was low, which was good for the Democrats, but output growth was not strong by historical standards, which hurt them.
An advantage of this approach is that one can make a prediction many months ahead. The candidates do not have to be known because the prediction is just of the share of the two-party vote, Democrat and Republican. I made my first prediction two years ahead, November 2014. At that time the first three conditions were known, and forecasts of the economy for 2015 and 2016 were available. The Democrats were predicted to get 48.7 percent of the presidential two-party vote. So even as far back as the end of 2014 the information in the equation was that the Democrats had an uphill struggle. This prediction was based on fairly optimistic forecasts of the economy, and as time passed, the economic forecasts became less optimistic. The economy did not grow as fast in 2015 and 2016 as it was predicted to at the end of 2014. The last prediction made right before the election, when the economic variables were known, was 44.0 percent for the Democrats—in other words, a large Republican victory.
In the 2016 election Hillary Clinton got 51.1 percent of the two-party vote. Compared to the equation’s last prediction of 44.0 percent, this yields an error of 7.1 percentage points, about twice the size of the average error of the equation, which in the past has been about 3.5 percentage points. While this year’s error is probably not large enough to warrant any specification changes when the equation is updated through the current election, it does raise the question, Why such a large error?
While this is not possible to test, most people would probably say that it is due to Trump’s personality. Had the Republicans nominated a more mainstream candidate, they may have done much better—much closer to what the equation was predicting. The prediction from the equation all along was that the Republicans were heavily favored. The election was theirs to lose because the first three conditions and the economy were slanted in their favor—and they almost lost it!
To conclude, polls, while interesting in their own right, are ultimately limited in helping us understand what motivates people to vote the way they do. Most polls simply ask people their voting plans, not how or why they arrived at those plans. The bottom line in this election is that the pundits and betting markets put too much emphasis on the polling data and not enough on the fundamentals. Trump had many negatives, but they were not quite large enough to offset the fundamentals.
Start reading Predicting Presidential Elections
and Other Things »