How this works

Same formula every day, and every past miss shown

Two forecasts of the 2028 presidential popular vote, built and judged the same way. One uses only economic data, one uses only presidential approval.

The two lines

• Economy: Fair’s model and Hibbs’s Bread and Peace model, combined with fixed weights set by how accurate each was in the backtest (1/error²). A toggle swaps in Abramowitz’s Time for Change as a third model.
• Mood: one model using the sitting president’s net approval and whether the party is in its first term.
• Every model predicts the party in power’s share of the two-party popular vote; for 2028 that is the Republicans, with no incumbent running. A share becomes a chance of winning using the model’s own backtest error (a t-distribution).

How the models are judged

• All four are re-fitted by ordinary least squares on the same 20 elections, 1948–2024. Published coefficients are not copied.
• Accuracy is leave-one-out only: each election is predicted by a version fitted without it. In-sample fit flatters a model and is never published as accuracy.
• Economic inputs use the figures as first published before each election day where they exist: output from 1960, inflation from 1968, income and CPI from 1976–80. Earlier years use the earliest published version or today’s data, tagged on the backtest page.
• 2020 stays in for every model, though its pandemic quarter strains the income and GDP models.

How the home page puts it in words

The headline only restates the two forecasts; it adds no model of its own. Each forecast is described by the favored party’s chance of winning the popular vote: below 60 in 100 is “slightly favored”, 60 to 74 is “favored”, 75 or more is “strongly favored”, and exactly 50 is “a toss-up”. If the two forecasts favor different parties, or either is a toss-up, the headline is “Too close to call”. If they favor the same party, the weaker of the two sets the wording: below 60 “Slight edge to”, 60 to 74 “Leans”, 75 or more “favored”.

Before 2028’s data exists

• If the election were today (default): the latest published quarters and today’s approval stand in for the missing 2027–28 figures.
• If the economy follows the Fed’s projections: the FOMC’s median projections fill the missing quarters; PCE inflation stands in for the GDP deflator, and income per person grows with GDP per person.

The approval number

The backtest uses Gallup’s last June poll of each election year. Gallup stopped polling presidential approval after December 2025, so for 2028 the input is the median net approval of the published polling averages listed on Wikipedia, using averages updated in the last 14 days (at least 3). The rule is fixed; changing it would be a new model version. Mixing sources is a known weakness, which Abramowitz’s own model also faces.

Model versions

Data and credits

• Economic data: FRED and ALFRED, Federal Reserve Bank of St. Louis (BEA, BLS and Census series, public domain).
• Projections: FOMC Summary of Economic Projections, Board of Governors of the Federal Reserve System.
• Source: Survey of Consumer Expectations, © 2013-26 Federal Reserve Bank of New York (FRBNY). The SCE data are available without charge at www.newyorkfed.org and may be used subject to license terms posted there. FRBNY disclaims any responsibility or legal liability for this analysis and interpretation of Survey of Consumer Expectations data.
• Current approval: averages compiled by Wikipedia contributors, CC BY-SA 4.0. Historical June approval: Gallup.
• Vote shares 1948–2020: Ray Fair, Yale. 2024: Federal Election Commission. War fatalities: Douglas Hibbs (2007).
• Not used: prediction markets and some well-known indexes whose terms don’t allow republication.

Known weaknesses

• Twenty elections is a small sample. Typical errors of 2.4–4.7 points mean most elections fall inside the margin of error.
• Approval and the economy overlap, so the two lines aren’t independent.
• These are popular-vote forecasts. The Electoral College can differ, as in 2000 and 2016.
• This site is not betting or investment advice.