Method.

Candidate Modifiers

• Home-state: direct home-state lift, with a separate swing-state fit table for candidates with obvious regional strengths.
• State fit: bounded adjustments for traits such as Appalachia, Sun Belt, Midwest, suburbs, rural states, Hispanic-heavy states, and independent-leaning states.
• Ideology: small penalties or bonuses in states where the candidate profile is a poor or strong fit.
• Favorability, finance, and electability: live national favorability averages when reachable, plus bounded donor-network and grassroots-capacity estimates.

State Baselines

Each state starts with a 2024 presidential baseline, then receives long-term trend, elasticity, and current-cycle adjustments.

• Baseline: 2024 Democratic margin, electoral votes, region, and population.
• Trend: state movement over the last decade-plus, including states such as Texas and Florida moving differently.
• Elasticity: close and fast-moving states react more to national conditions than deeply safe states.
• Current cycle: 2026 Senate state margins are lightly blended where they exist; generic ballot is shared with the Senate model.

Correlated Polling Error

The model generates correlated polling error across states:

• National error: Drawn from normal distribution (SD = 3.5 points)
• State-specific error: Drawn from normal distribution (SD = 4.0 points)
• Correlation: 0.6 (60% of state error comes from national component)
• This accounts for systematic polling bias that affects all states similarly

Monte Carlo Simulation

For each of 100,000 simulations:

1. Generate correlated polling error for all states
2. Calculate final margin for each state (baseline + modifiers + error)
3. Determine state winner based on margin
4. Sum electoral votes for each candidate
5. Track state wins, EV distribution, and popular vote

Final probabilities are calculated as the proportion of simulations each candidate wins.

Tipping Point State

The tipping point state is the state most likely to decide the election:

• Calculated as the state with the highest "tipping power"
• Tipping power = min(Dem wins, Rep wins) × state EV
• States that are close to 50% probability with high EV have highest tipping power

Historical Backtesting

The model is validated against historical presidential elections:

• 2020: Biden vs Trump (306-232 EV)
• 2016: Clinton vs Trump (227-304 EV)
• Additional cycles (2012, 2008, 2004) can be added

Backtest metrics include:

• Mean Brier score (lower is better)
• Overall accuracy
• Calibration by probability bucket (50-60%, 60-70%, etc.)

Fundamentals

Current national inputs:

• Generic ballot: read from the Senate forecast's stored polling blend.
• Presidential approval: blended from the Senate forecast cache and live approval pages when reachable.
• Consumer sentiment: FRED UMCSENT, weighted more heavily than raw GDP because voter mood often diverges from topline growth.
• Economic indicators: unemployment and GDP are secondary terms.
• Migration: Census state population growth and net migration are used as a small electorate-change term.
• State and matchup polling: included when reachable and parseable, with lower weight this early in the cycle.

National conditions pass through state elasticity, so a one-point national move does not affect Wyoming, Texas, Wisconsin, and Vermont the same way.

Data Sources

Primary source families:

• Shared 2026 polling blend: generic ballot from the Senate forecast's stored source blend.
• Approval: Senate Pollfinity cache plus live approval pages when reachable.
• Consumer sentiment: FRED UMCSENT.
• Migration: Census Vintage 2024 state population estimates and migration components.
• Presidential polling: RealClearPolling and Public Sentiment Institute pages when reachable and parseable.
• Candidate favorability: Ballotline and USPollingData pages where they publish a candidate average.
• Campaign finance: normalized donor-network, donor-constraint, and grassroots-capacity profiles until named 2028 presidential committees are consistently available.
• Current-cycle state signal: 2026 Senate state forecast margins, lightly weighted.

Limitations

• Some candidates still fall back to stored favorability estimates when no consistent public national average is reachable.
• Campaign finance is currently a profile-based estimate for most candidates because 2028 presidential committees are not consistently registered under these matchups.
• 2028 head-to-head and state polling is sparse this early.
• Candidate traits are intentionally bounded; they do not model campaign quality, running mates, scandals, or turnout operations.
• Historical presidential backtest currently covers 2020 and 2016.
• Some live polling and approval pages can block automated requests.