+13 is not a fluke: what special election overperformance really tells us

Special elections are the most underused forecasting tool in American politics. They're off-cycle, low-salience, and they strip out the nationalizing effect of a presidential ballot — so the result is closer to a pure read on local fundamentals. The 2025–2026 cycle, Democrats have overperformed the equivalent prior presidential result by a weighted average of +13.1 points across 11 contested specials. The historical conversion ratio from special-election overperformance to November House popular vote sits at roughly 0.55 to 0.65 in modern cycles. That gives you a +7 to +8.5 swing baked in before any national-environment move. This piece is about why that relationship holds, and the three places it breaks.

Why specials are predictive

The model is simple in concept: turnout is the variable, and turnout in a special is structurally different from turnout in a presidential year. A presidential election pulls in the demobilized base. A special election pulls in the motivated base. The size of the motivated-base advantage, in either direction, is the special-election signal — and it's been the most consistent mid-cycle indicator we have since the 1990s.

Caveat one: small-N races

Specials with under ~30,000 total votes have a wider confidence interval. The +13 average is heavily weighted by the 5 larger races. Trim the dataset to the 6 smaller races and the average still lands at +9.4, but the standard error doubles. Use the small-N as a direction, not a magnitude.

Caveat two: candidate quality

Specials can be won or lost on the individual candidate. A retiring incumbent's chosen successor is not the same as a scandal-driven open seat, and the conversion ratio is different. Adjust for candidate quality before treating the special number as a referendum read.

Caveat three: turnout composition shift

If the special-election turnout composition shifts (e.g., a heavily-Republican rural district with low-propensity Republican surge), the overperformance number can be misleading. Cross-check with the absentee and early-vote split when available.

What the 2025–2026 number actually says

Even at the conservative end of the conversion range, with a -1.5 discount for the small-N caveat and a -1.0 discount for the candidate-quality adjustment, the 2025–2026 specials imply a Democratic House popular-vote lead of +5.5 to +6.0. That is consistent with the polling aggregate. The specials and the polls are saying the same thing, which is the rarest and most useful state of affairs in mid-cycle forecasting.

Receipts

• Special-election dataset, 11 contested races, 2025-01 through 2026-07, weighted by vote share
• Historical conversion ratios, 1994 through 2024, with adjustments for cycle type
• Turnout composition data where reported, RNC + DNC filings