Senate forecast · House forecast
Historical candidate performance
Included provisionally in both forecasts as of October 20, 2026.
I compare prior election performance with the expected partisan and national environment. Recent comparable results receive more influence. Senate candidates can bring records from the House, a governorship or the Senate; the transfer from each office is fitted separately. Alaska’s House comparison uses certified final ranked-choice results. Repeated wins are averaged, not added together as separate bonuses.
For House incumbents, the revised measure accounts for the prior district map, national environment and whether the earlier race was won as an incumbent, challenger or open-seat candidate. Existing available candidate history is retained when no comparable replacement exists.
The change replaces overlapping prior-performance terms through paired refits. It does not add a second fixed incumbency advantage. Polling weight and total uncertainty are preserved, and each race’s incremental change is capped at three margin points. The same formula is applied across the matched candidate records; no Alaska-specific bonus is imposed.
This is a provisional modeling choice. The Senate pooled-history version slightly worsened historical error (4.60 to 4.70 points); the House context version modestly improved it (4.77 to 4.67). Neither established a reliable improvement across independent elections. The ranges remain conditional on the selected history formula.
Seven unpolled Senate estimates retain their existing structural history treatment because the matched polling comparison does not establish a replacement for them. Missing comparable history is unknown, not zero candidate ability. Fixed-party House contests remain fixed.
Every race and its adjustment · Saved assumptions · Validation · Earlier comprehensive tests
Source records and formulas
Senate candidate comparisons and source links · House prior-election evidence
Senate performance is capped at15 points before an eight-year recency decay and is estimated separately by previous office. Earlier elections fit the transfer coefficients. The published effect is the difference between the richer and original structural predictions, applied once to the existing current estimate. House comparisons similarly replace the prior-performance feature in the same earlier-election regression. No current election outcome enters either fit.
Races with a changed estimate