[Updated as of 3/22/2024]

There is great interest in foreseeing the outcome of the 2024 election. In light of this, the American Presidency Project presents instructive graphs comparing incumbent President Biden with prior incumbents both who won and who lost their bid for reelection.

Of course, there are no recent—or even distant—close analogies to the 2024 election. The last president to run for reelection as an incumbent, suffer a defeat, and then run for election again and win, was Grover Cleveland in 1892.[1] No president, having been impeached, was again a candidate for the presidency. No president, having left office, has been formally accused of crimes. No U.S. election has featured two such aged candidates. In fact, only fifteen of 46 presidents have outlived Donald Trump, the younger of the two major 2024 candidates.[2]

So, as the investment companies warn, “past performance does not guarantee future results.”

In the past, I have used variables like those in the graphs below to generate election forecasts primarily to stimulate classroom discussion. The models used are a variation on the “time for change” model developed by Alan Abramowitz.[3] More on this model below.

First Graph Pair: Gallup Approval

The first graphs compare the trends of presidential job approval in the two years prior to the election. The data plotted are the proportion “approving” of the incumbent on the well-known Gallup question: “Do you approve or disapprove of the way [name] is handling his job as president?” No other presidential evaluation question has been in use as long as Gallup’s.

In the winners graph we see that except for Truman, every winning president had greater than 50 percent job approval. In the 1948 data for Truman, note that Gallup’s last poll was in late June of that year. So, post-June 1948 movement of public opinion—if any—was not measured.[4]

Among the incumbent losers, all had public approval ratings below 50 percent. The highest approval for a loser was 46 percent (Trump).

For monitoring purposes, then, one will watch to see if Biden’s approval rate begins to trend upward just as Obama’s did in 2012.

CLICK ON IMAGE TO SEE LARGER GRAPH

Second Graph Pair: Employment Growth

The second graphs display the annual growth rate of seasonally adjusted non-farm employment.[5] The change shown in these graphs ranges from strongly negative (-0.13) to strongly positive (+0.11). The statistical forecasting model, uses data from about 6 months prior to the election--or the 180 days point on the graph X-axis.

In the winners graph, the final election-day employment growth rates are all positive and above 1.5 percent. The range for winners is from around +0.0145 (or 1.45 percent) for George W. Bush to +0.043 (or 4.3 percent) for Reagan. Good economic numbers matter. Current numbers put Biden at about 0.018 (or 1.8 percent). That is, Biden's current numbers, if sustained, would leave him higher than both Obama and George W. Bush at election time.

In the losers graph the election-day range is from -0.063 (or -6.3 percent) for Trump to +0.032 (or +3.2 percent) for Ford. If Biden's current numbers are sustained, he will be well above Carter, George H. W. Bush, and Trump. 

Ford, of course, is a special case. He is the only President not elected either to the Presidency or the Vice-Presidency. He succeeded the disgraced President Nixon in office, and his decision to pardon Nixon was very unpopular.

CLICK IMAGES FOR LARGER GRAPH IMAGE

Time for Change Model

The “time for change” idea is that a small, but significant, proportion of the electorate prefers to see a partisan change in the presidency after either major party holds the presidency for two consecutive terms. The implication of this (backed by strong statistical results) is that candidates like Hillary Clinton, Al Gore, and John McCain faced an additional election hurdle. To win, they would have to overcome this aversion to extended partisan domination. The 1988 election of George H. W. Bush shows that it is possible for the incumbent party to win a “time for change” election. In such a case, it may be helpful to face an inept opponent.

All that is Ignored

In the classroom I use the statistical model to address this question: How can a model work if it ignores almost everything about the election contest that dominates the news? What about the issues? What about the qualities of the candidates and their campaigning? What about lies, smears, and dirty tricks?

Every candidate and campaign has weaknesses and strengths. There are issues that work both for and against them. Smart, well-run campaigns emphasize their strengths and the opposition’s weaknesses. They figure out what appeals work for the marginal voters in swing states—which may be different from effective partisan appeals nationally.

Vigorous, smart campaigns tend to offset each other’s strengths and weaknesses. It’s the nature of acompetitive process. So, what matters when there is vigorous competition are the basic “fundamental” features: Do enough voters see the incumbent as doing a good job? How good is the economy? Is it a time for change?

The model definitely does not mean that campaigns do not matter, or that candidate performance is irrelevant. Quite the contrary. Does a candidate have vulnerabilities that cannot be overcome or diminished?

Remarkably enough, the statistical model works well for past elections!

[In coming weeks, I will update this analysis to illustrate in more detail the past performance of the model, and further discuss the implications for 2024.]

 

[Revision Note:  The version of this analysis originally published on March 2, 2024 erroneously included data for Biden that were from earlier in his term.  Those data have been corrected; graphs updated; related text corrected.]

---

[5] I am using a two-month moving average monthly annual change, where the compound growth rate = ln(currentE/prior12moE), where E= total employment. The series is calculated by the Bureau of Labor Statistics on a monthly basis. The series identifier is CES0000000001 and can be downloaded from https://data.bls.gov/timeseries/CES0000000001