The Primary Model gives President Donald Trump a 91% chance of winning reelection in a matchup against Democrat Joe Biden (http://primarymodel.com/). Trump is predicted to get 362 electoral votes, Biden 176 (Norpoth Reference Norpoth2020).Footnote ¹ To be sure, the forecast of a near-certain Trump victory strains credulity. Polls have shown Trump trailing Biden in the RealClearPolitics average every day of this election year thus far. This, of course, feels eerily familiar. In 2016, polls and poll aggregation forecasts projected a nearly insurmountable lead for Hillary Clinton: Trump simply had no path to the presidency. In contrast, the Primary Model gave Trump an 87% chance of winning, although that was based on a popular vote forecast that turned out to be wrong (Norpoth Reference Norpoth2016).

To avoid such confusion, the Primary Model simply predicts the electoral vote this time. It does so not by extrapolating it from a popular vote forecast but by making it straight from the model predictors. In the event that one candidate wins the popular vote and the other the electoral vote, as happened in 2016, all that matters is the electoral vote. For the record, the Primary Model gets the electoral vote winner right in 25 of the 27 elections from 1912 to 2016.

In addition to primaries, the Primary Model takes advantage of a dynamic in presidential elections that can be tracked for almost 200 years, ever since the formation of a two-party system (Norpoth Reference Norpoth2014). As an illustration, consider the pattern displayed in table 1 for elections since 1960. Whenever a party had held the White House for one term, it went on to win (W) six of seven elections; whenever it held it for two or more terms, it ended up losing (L) seven of eight. All by itself, this pattern makes a first-term president like Trump the odds-on favorite to win in 2020 while also making Hillary Clinton’s loss in 2016, after two Democratic terms, look like no aberration. This conveys a strong sign that American presidential elections obey the swing of an electoral pendulum, at least since 1960.

Table 1

Record of the White House Party in Presidential Elections since 1960 (Electoral Vote)

Figure 1 tracks presidential elections for a longer time frame, from 1828 to 2016. Note that it uses the electoral vote, specifically the Democratic number of electors, normalized on 538, the current total. One can spot roughly 10 cycles with the naked eye, which translates into a cycle length of about 20 years. In other words, from 1828 to 2016, a party held the White House for about 10 years, on the average; that is, between two and three terms. To capitalize on this swing of the electoral pendulum for forecasting, I borrowed an algorithm invented by the British statistician Yule in the 1920s to gauge cycles in sunspots (Yule Reference Yule1927). It is worth noting that the sunspot series over two centuries bears an eerie resemblance to the electoral series; electoral analysis, after all, has more in common with astronomy than astrology.

Figure 1

Democratic Electoral Votes, 1828–2016, Normalized on 538

Yule’s algorithm, which boils down to a second-order autoregressive process, yields the following coefficients for the electoral vote series (1828–2016):

$$ {\displaystyle \begin{array}{l}{\mathrm{DEMEV}}_{\mathrm{t}}=253+0.464\ {\mathrm{DEMEV}}_{\mathrm{t}-1}-0.283\ {\mathrm{DEMEV}}_{\mathrm{t}-2}\\ {} (22) (0.150) (0.129)\end{array}} $$

It is a simple step from this equation to a 2020 forecast, because we know DEMEV_t-1 and DEMEV_t-2, the Democratic electoral votes in 2016 and 2012. It yields 221 electoral votes for the Democratic candidate (Biden) and 317 for the Republican (Trump). It may seem far-fetched that the results of past presidential elections can predict the future without any other information, be it the identity of the candidates, the record of incumbent administration, or some other factor. But that is what 200 years of electoral swings can deliver. To be sure, the chances are modest that this kind of prediction will come true. For 2020, it gives Trump a 66% chance of winning: better than a flip of the coin, but not quite good enough to bet the house. Which predictor should be added to warrant such confidence?

Any review of election forecasting (Norpoth Reference Norpoth, Mintz and Terris2018) shows that the economy is by far the favorite predictor, followed by presidential approval and polls of vote intentions. Primaries are not, if they make the cut at all. Yet primaries expose candidates to a real-life election, with all the bells and whistles of a campaign. Winning or losing primaries tells us something about how voters size up candidates. Those evaluations have long been proven pivotal for voters’ choices in November (Lewis-Beck et al. Reference Lewis-Beck, Jacoby, Norpoth and Weisberg2008). Furthermore, primary performance of incumbents turns out to correlate very strongly with their mid-year approval rating (.76).

No sooner had primaries been introduced than a former president took advantage of this invention to challenge a sitting president of his own party, which happened in 1912. Theodore Roosevelt wound up soundly defeating incumbent William Howard Taft in the Republican primaries. Even so, the GOP chose to stick with Taft and handed him the nomination for another term at its national convention. Meanwhile, on the Democratic side, Woodrow Wilson, a former professor in our profession, won the primary battle in that party, which then nominated him for president at its national convention. We all know what happened in November 1912: Wilson was elected president. Putting it all together, one can spot a pattern: The party that nominates the primary winner goes on to defeat in the general election the party that nominates a primary loser.

It seems nobody noticed at the time or wondered whether primaries might make for a handy predictor of the vote in November. More than a hundred years on, we have a record that can tell whether 1912 was just an anecdote. Table 2 assembles a representative sample of presidential elections since primaries were introduced. The list includes 1948, which polls and pundits got wrong. The primaries that year, in contrast, firmly point to Truman as the winner, with Dewey joining Taft in the hall of primary–general election losers. In 1980, both candidates were primary winners, but Reagan by a far larger margin than Carter, who ended up losing in November. Turning to the most recent election, it is true that Hillary Clinton amassed a larger share of the Democratic primary vote in 2016 than did Trump on the Republican side. But given the varied number of viable contenders—just two versus a dozen —it is necessary to adapt the metric of primary performance. For contests without a sitting president, that metric uses the relative strength of the nominee compared to the strongest rival.

Table 2

Primary Vote for Presidential Candidates and their Strongest Rivals in Selected Elections

Notes: 1912–1948: all primaries; 1952–2004: New Hampshire only; 2008–2020: New Hampshire and South Carolina averaged.

Not all primaries are equal, of course, in their ability to influence presidential elections. Privilege goes with being the first to let voters choose presidential candidates in a primary election. New Hampshire began offering that opportunity in 1952, while also making it extremely easy for candidates to get on the primary ballot, requiring no more than a $1,000 filing fee. No serious contender can afford to skip this state. For voters, the New Hampshire primary is not closed to partisans but allows Independents to participate; they generally make up close to half of the electorate on Primary Day. Candidates hoping to win the primary here must appeal to a broader segment than the base. The New Hampshire primary puts would-be nominees for president to a test like no other state. For all these reasons, the measure of primary performance switched from all primaries to just New Hampshire in 1952.

Still, this state’s primary it is not a perfect test. A frequently heard complaint is that New Hampshire is not representative of the nation. In particular, it lacks a demographic critical for success of some Democratic candidates in recent years: African Americans. South Carolina, a state with an early primary right on the heels of the New Hampshire event in recent years, fills the need for balance. Half of the Democratic voters in the “First in the South” primary are Black. For the 2016 forecast South Carolina was added to New Hampshire to gauge primary performance; the same was done for the 2020 forecast, at least for the Democratic candidate.Footnote ² The addition of further states does not seem warranted at this time to cover constituencies that might be missing from the mix of New Hampshire and South Carolina.

Joe Biden scored a solid win in South Carolina to rebound from a poor finish in New Hampshire; his average score in those two primaries relative to the strongest rival came to 48%. By comparison, Trump soundly defeated a prominent Republican, former Massachusetts governor William Weld, in New Hampshire (with 86% of the vote); it is all but certain Trump would have repeated that showing in South Carolina had the Republican primary not been canceled. Superior primary performance added to the first-term benefit makes Trump the heavy favorite to win in November. Using the estimates presented in table 3, the Primary Model predicts 362 electoral votes for Trump and 176 for Biden, giving the president a 91% chance of winning reelection.Footnote ³ Table 4 compares the in-sample predictions with the actual electoral votes for all elections from 1912 to 2016. The offshoot is that the model gets it right in 25 of those 27 elections. One of the two misses is the notorious 2000 election in which, as is widely believed, “the wrong man [was] elected president” (Mebane Reference Mebane2004). Overall, the model’s historical record adds to the confidence in the forecast of a Trump victory in 2020.

Table 3

Predicting the Democratic Electoral Vote, 1912–2016

Notes: *p <.1; ** p <.01; *** p <.001.

Table 4

Model Predictions of Democratic Electoral Votes