A group of political scientists published this piece back in October, before the election. They predicted
every state correctly, a popular vote victory for Trump, and 3/4 chance of Trump victory in Electoral College.
The same people, using the same model, predicted a Biden victory in 2020 and 49/50 states correctly (they missed Georgia, which they predicted for Trump back then).
Their model uses the following fundamentals, for each state:
1. State deviation from national vote in t-1 election
2. Presidential approval rating in each state (estimated with Multilevel Regression and Post-stratification)
3. Economic conditions in each state (personal income deflated by CPI, unemployment rate, non-farm payroll employment, manufacturing hours by production workers)
4. Presidential and VP home-field advantage adjustments if they apply in the state
5. If VP runs, an adjustment to account for difference in approval rating between P and VP in question
