However unpredictable the ascent of Donald Trump onto the stage of presidential politics may have been, one forecast model has been highly confident for months that he would win the election on November 8, 2016. The Primary Model predicted on March 7, 2016 that Trump would defeat Hillary Clinton with 87 percent certainty. 1
The forecast of a near-certain Trump victory at a moment when he was trailing both Clinton and Bernie Sanders in every poll, some by double-digits, was greeted with a heavy mix of shock, cheers, amazement, and derision, much of it on social media but also regular media outlets. Many offered bets against the forecast, gleeful that it would turn out wrong. There is nothing to add to or subtract from the March forecast here. It was unconditional, final, and not subject to updating. Just in case Hillary Clinton would not be the Democratic nominee, the Primary Model gave the nod to Trump over Bernie Sanders with 99% certainty; forecasts for Republican nominees other than Trump were also issued. 2 What are the ingredients of this forecast model?
As the name indicates, the Primary Model relies on presidential primaries as a predictor of the vote in the general election; it also makes use of a swing of the electoral pendulum that is useful for forecasting (http://primarymodel.com/). For the record, the Primary Model, with slight modifications, has correctly predicted the winner of the popular vote in all five presidential elections since it was introduced in 1996 (Norpoth 1996, 2001, 2004, 2008, Norpoth and Bednarczuk 2012). 3 In recent elections the forecast has been issued as early as January of the election year. Also note that for all elections from 1912 to 2012, the Primary Model picks the winner, albeit retroactively, every time except in 1960.
Turning to the primary predictor of the model, it never ceases to amaze its author how many students of elections are surprised to learn that presidential primaries predict anything beyond perhaps who wins the nomination. Yet the outcomes of these primaries prove to be uncanny leading indicators of wins the general election for president in November. It also comes as a surprise that presidential primaries have been around as far back as the days before World War I. They were introduced into the presidential nomination process in 1912. William Howard Taft was President. A Republican, he had succeeded Theodore Roosevelt with the best wishes of his friend. But now Roosevelt was itching to return to the White House. He seized on the new invention and challenged Taft for the Republican nomination in the primaries. The sitting President lost badly in those contests, yet managed to secure the nomination at the GOP national convention nonetheless. Meanwhile a former professor of politics won the primary battle in the Democratic Party. Woodrow Wilson then went on to secure the presidential nomination at the Democratic national convention. And in November of 1912 Wilson was elected President while Taft lost. Hence the candidate who won his party’s primary vote, Woodrow Wilson, went on to defeat the candidate who lost his party’s primary vote, William Howard Taft, in the general election (see table 1).
This was not the big story of the 1912 election, if it made any news. Was it a fluke? Or was it a precedent that was unappreciated at the time but that would establish a rule connecting primary and general election outcomes? A look at elections over a hundred years since then makes it clear that 1912 was no fluke. To be sure, things are not always as simple as in 1912, when a primary winner faced a primary loser in November. Often both candidates are primary winners. As a general rule, we would predict that the candidate with the stronger showing in his or her party’s primaries wins out in November. That prediction holds up pretty well as can be seen in table 1 with a representative sample of elections from 1912 to 2012. Note that for elections prior to 1952, all primaries were included while for elections since 1952 only the New Hampshire Primary has been used to measure primary performance.
Besides primaries, the forecast model relies on a swing of the electoral pendulum, which generates cycles in the vote for President (Norpoth 2014). Since 1960, as illustrated in table 2, the party controlling the White House has won six of the seven elections after one term while losing five of six after two terms. During that span of time, the presidential party succeeded but once to win a third term—with George H. W. Bush in 1988, following two Reagan terms. After two terms of Democrat Barack Obama in the White the electoral pendulum is poised to swing back to the Republicans in 2016.
How typical is the pattern revealed in table 2? To check for cycles in presidential elections we can go back as far as 1828; popular voting became widespread then and the two-party system took shape. The Democratic percentage of the two-party popular vote is charted in figure 1. One can spot about ten cycles in figure 1 over nearly two centuries. So a cycle lasts about 20 years or 5.2 terms to be precise, using estimates from a second-order autoregressive model (Norpoth 2014, 333). Hence it is a safe bet that Democrats and Republicans can count on a second term in the White House but the odds to win a third term are less favorable. How favorable or unfavorable?
According to the statistics of the cyclical model, that depends on how well the party does in its reelection compared to its vote when it first captures the White House (Norpoth 2014). A better showing in the reelection bodes well for a third term, while a poorer showing bodes ill for it. Consider the Reagan elections: Ronald Reagan defeated Walter Mondale in 1984 by a margin almost twice as large as the margin by which he beat Carter in 1980, a good omen for the Republican Party to extend its White House lease in 1988 for a third term, as George H. W. Bush managed to do handily. The country appeared to want more of the Reagan experience and was not itching for change. Compare this to the Obama elections. The record shows that Barack Obama won reelection in 2012 with a smaller margin than he got when he was elected in 2008. This is bad news for a Democrat to extend the party’s White House lease this year for a third term. “Time for a change” sounds more appealing in 2016 than “more of the same.” The electoral pendulum is poised to swing back this year. The statistics of the cyclical model favors the Republicans to recapture the White House with 51.4% of the two-party vote (Norpoth 2014, 334). The question is whether the Republican nominee can take advantage of this opportunity, which comes with only a modest certainty (61%).
For the Primary Model, the question is answered by performance in primaries. Forecasts of previous elections (2000–2012) relied exclusively on the vote in the New Hampshire Primary. Being the first on the presidential primary calendar, it garners a disproportionate share of coverage; it is shockingly easy for anyone to enter (just a fee of $1,000, no petitions, etc.); the state is small enough to cover from one end to the other; its primary is open to Independents, who generally make up close to half of a party’s primary electorate in New Hampshire; and it draws a voter turnout on par with a general election. Is there another state that offers a better test of the electoral strength of a would-be nominee for President? Since 1952, when New Hampshire put the names of presidential candidates on the primary ballot, almost every winner in November won his party’s primary in New Hampshire.
In recent years, however, some presidential candidates have built winning campaigns for the nomination with strong support from a group that is nearly invisible in New Hampshire: African-Americans. Obama did so in 2008, overcoming a defeat in the New Hampshire Primary with a victory shortly after in the South Carolina Primary. African-Americans make up close to half of the Democratic electorate in the “First in the South” Primary. For the favorite of this large and loyal Democratic constituency, the showing in New Hampshire may not be such a telling indicator of future electoral strength. Like Obama in 2008, Hillary Clinton is the favorite of African-Americans this year, ironically the group that helped her opponent beat her eight years ago. To gauge primary performance in 2016 I decided to include the South Carolina Primary along with New Hampshire. As it happened, Hillary Clinton beat Bernie Sanders handily in South Carolina after losing to him badly in New Hampshire. On the Republican side Donald Trump won both contests, with John Kasich coming in second in New Hampshire and Ted Cruz in South Carolina. Using the standard metric, which gauges the primary performance of a nominee relative to the strongest rival, gives Trump a higher score than Clinton for the combined showing in New Hampshire in South Carolina (see table 3).
The prediction formula of the Primary Model, as shown in table 4, leads to this forecast: in the match-up between the Republican and Democratic primary winners, Donald Trump will defeat Hillary Clinton with 52.5% of the two-party popular vote, with her getting 47.5%. 4 It is 87% certain that Trump, not Clinton, will be the next President. Trump benefits from a swing of the electoral pendulum to the Republican side in 2016 and his superior performance in early primaries. While Trump won the Republican primaries in both New Hampshire and South Carolina, Hillary Clinton split the Democratic primaries in those states with Bernie Sanders.
Note: The formula predicts the Democratic share of the two-party vote. Hence the Republican share is (100 – Dem. %). Clinton gets the weight for an incumbent-party candidate (0.429), Trump the weight for an opposition-party candidate (0.170). Clinton’s primary score equals her vote (56) minus the mean (54) for incumbent-party candidates. Trump’s primary score equals his vote (64) minus the mean (49) for non-incumbent party candidates, inverted (-1) for Republicans because the formula predicts the Democratic vote.
It is 87% certain that Trump, not Clinton, will be the next President.
For the record, the Primary Model, as of March 7 this year, also made forecasts for other likely match-ups in November. It predicted that Trump would defeat Bernie Sanders with 99% certainty; after all, Bernie was the primary loser on the Democratic side. Hillary Clinton would defeat Ted Cruz, who wound up the strongest among the primary losers on the Republican side, with 86% certainty. And Ted Cruz would defeat Bernie Sanders with 89% certainty. All these match-ups are academic now. 5 Nonetheless the capability to make electoral forecasts for candidates by name rather than just for parties is a strong selling point of the Primary Model, a is the ability to make a November forecast early in the election year, when the battle for the nomination is just getting started. Early prophecy is good for publicity but full of risks, or so said Mark Twain.
1. An earlier forecast, which predicted a Trump victory over Clinton with 97% certainty and received much notoriety, was made before the Democratic primary in South Carolina was held and relied on polling reports for that state. Clinton wound up beating Sanders by a much bigger margin than was indicated by pre-primary polls. That closed the margin and lowered the certainty of the forecast. The earlier forecast is available here: http://www.huffingtonpost.com/helmut-norpoth/trump-nearcertain-to-defe_b_9403762.html?1457390306J
2. Any resemblance between predicted and preferred winners is purely coincidental.
3. For an overview of election forecasting, the variety of approaches and models, and applications in the United States as well as abroad, see Stegmaier and Norpoth (2013).
4. The parameters of the prediction formula were estimated with election results (primary and general) spanning a century (1912–2012). The dependent variable is the Democratic percentage of the two-party vote in presidential elections; for the 1912 election, however, the two-party vote division was approximated by the House vote. The primary support variables are capped in the 35–65 range and mean-inverted for years of Republican control of the presidency. For the pre-New Deal period, a partisan adjustment was required, which was handled by a variable coded 1 for elections up to 1932, and 0 for elections since. All coefficients of the prediction formula are significant beyond the 0.001 level. The R-squared of the formula is 0.93 and its standard error is 2.18. The Ljung-Box test indicates that there is no significant error autocorrelation (Q = 2.2 for the first four lags, p > .60).
5. For a contest between Trump and Sanders, the formula would predict a victory for Trump with 57.7% to 42.3% for Sanders. In the event that the Republican Party were to nominate Ted Cruz instead of Trump as its presidential candidate, Clinton would defeat Ted Cruz by a margin of 52.4% to 47.6% of the two-party vote; any other GOP candidate (except Trump) would fare the same way against Clinton. In the event that the Democrats were to nominate Bernie Sanders instead of Hillary Clinton, Cruz would beat Sanders by a margin of 52.8% to 47.2% of the two-party vote; any other GOP candidate (except Trump) would fare the same way against Sanders.