Get FutureBallot by email. Emails when FutureBallot publishes a new explainer or video, when each state's results are certified, when it reports how its forecasts did, and when the 2028 forecast begins.

Why election forecasts can disagree

November 8, 2026 · By FutureBallot

The same expected lead can produce different chances of winning when forecasts allow different amounts of uncertainty. A simple worked example shows why.

A forecast’s win probability answers how likely a candidate is to finish ahead. Vote share answers how much of the vote a candidate receives. An 84% chance of winning does not mean an expected 84% of the vote.

Same lead, different uncertainty

Suppose two hypothetical models both put Candidate A’s expected margin over Candidate B at two percentage points. Both represent that margin with a normal distribution. One uses a standard deviation of two points; the other uses four. Standard deviation describes the spread around the expected margin.

image/svg+xml Matplotlib v3.10.8, https://matplotlib.org/ −10 −5 0 5 10 Candidate A margin over Candidate B (percentage points) 0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200 Probability density Same expected lead, different chances of winning Spread 2 points: Candidate A 84%, B 16% Spread 4 points: Candidate A 69%, B 31% Tie Same expected margin: A +2 Hypothetical normal distributions; not a forecast of an actual race. Spread = standard deviation.

In this illustration, A wins whenever the margin is above zero. The narrower distribution puts about 84% of its probability above zero, leaving B about 16%. The wider distribution gives A about 69% and B about 31%. Both models favor A, but the wider one allows more outcomes in which B finishes ahead. These are illustrative calculations, not forecasts for an actual race or a description of FutureBallot’s full model.

Try different leads

The table changes the expected lead while keeping those two spreads. Entries are A’s chance of winning; B’s chance is 100% minus the entry.

Try different leads
Expected A margin (points) Spread of 2 points Spread of 4 points
+0 50.0% 50.0%
+1 69.1% 59.9%
+2 84.1% 69.1%
+3 93.3% 77.3%
+4 97.7% 84.1%

The calculation is P(A wins) = Φ(expected margin ÷ standard deviation), where Φ is the standard normal cumulative distribution function. For the two-point lead, Φ(1) = 0.8413 and Φ(0.5) = 0.6915. The example assumes a continuous margin, so an exact tie has probability zero. The example omits turnout, individual polls and discrete ballots.

What to compare in actual forecasts

Start with the question and timestamp: are both forecasts describing the same contest at the same point in time? Then compare the information included, the weight placed on it and the uncertainty allowed around the estimate. A difference in the published number need not come from a disagreement about who currently leads.

For chamber control, also examine how each model treats errors shared across races. Several close contests moving in the same direction can change control together. FutureBallot’s methodology page describes its inputs, historical-error assumptions and shared-error simulations, along with limitations.

Those comparisons can explain a difference; they cannot establish that two forecasts are equally good. That requires evidence about performance. Understanding the assumptions is the beginning of evaluating a forecast, rather than a substitute for it.