The Crystal Ball Forecasting Elections in the United States
I. Long-Term Forecasts: Can we do better than flipping a coin? A. Elections have patterns: Winning streaks B. Streaks tend to be 2-4 elections long
C. The weighted coin flip model 1. Best guess for Presidential elections years in advance
2. Performance: Better than flipping a coin…
D. Congress: Bet on the incumbents
II. Short-term forecasts A. Opinion polls: How well do they predict elections? 1. Continuous polling = recent development
2. How did tracking polls perform?
1992
Generic Congressional Ballot
B. The Wisdom of Crowds 1. How often is the public right? (early Oct)
2. Electoral Stock Markets You can “buy” stock in a candidate (real money futures contracts) Theory: people who invest money have a huge stake in the outcome, so have incentives to weigh information carefully (invisible hand)
Market predictions for 2006:
IV. Presidential Elections
Accounting for Fatigue…
Performance: 1992 and 1996
Performance: 2000
2000: Were the polls any better?
Labor Day Polls: Good predictors of winner, poor predictors of vote share
The economy: Also a good predictor of winner
Comparing Political Science Models
V. Congressional Elections A. Timing: President’s party tends to lose seats in midterms (worse for Democrats) B. Exposure: How many seats are exposed? 1. House: Party has higher % of seats than historical average 2. Senate: Number of each party’s seats up for grabs C. The referendum model: Presidential approval helps/harms incumbent party
1. House seats: approval, growth, and timing
House seat model performance
2. Senate Seats: Exposure, Referendum, and Partisan Advantage
D. Midterm polls can be misleading 1. General format = “Generic ballot” 2. Generic ballot overestimates Democrats’ chances
3. Using the generic ballot to predict House seats
VI. Do Campaigns Matter? Yes…but we still don’t know exactly how