1
Why We Need a Better Approach 2
Why We Need a Better Approach will decide this weekend McCain by half a point Too close to call Obama- Biden Too close to call 3
Why We Need a Better Approach 4
Why We Need a Better Approach 5
Why We Need a Better Approach 6
Why We Need a Better Approach
Why We Need a Better Approach
Why We Need a Better Approach
Why We Need a Better Approach Cherry-Picked Obama Polls Cherry-Picked Romney Polls Obama +5.0 Colorado Romney +3.0 Obama +2.3 Florida Romney +5.3 Obama +6.3 Iowa Romney +1.0 Obama +8.0 Michigan Obama +0.8 Obama +7.0 Nevada Obama +1.7 Obama +6.4 New Hampshire Romney +1.7 Obama +2.0 North Carolina Romney +6.7 Obama +5.3 Ohio Romney +1.0 Obama +7.7 Pennsylvania Obama +2.0 Obama +5.3 Virginia Romney +3.3 Obama +8.0 Wisconsin Obama +0.7
Why We Need a Better Approach Cherry-Picked Obama Polls Actual Results Cherry-Picked Romney Polls Obama +5.0 Colorado Obama +5.4 Colorado Romney +3.0 Obama +2.3 Florida Obama +0.9 Florida Romney +5.3 Obama +6.3 Iowa Obama +5.8 Iowa Romney +1.0 Obama +8.0 Michigan Obama +9.5 Michigan Obama +0.8 Obama +7.0 Nevada Obama +6.7 Nevada Obama +1.7 Obama +6.4 New Hampshire Obama +5.6 New Hampshire Romney +1.7 Obama +2.0 North Carolina Romney +2.0 North Carolina Romney +6.7 Obama +5.3 Ohio Obama +3.0 Ohio Romney +1.0 Obama +7.7 Pennsylvania Obama +5.4 Pennsylvania Obama +2.0 Obama +5.3 Virginia Obama +3.9 Virginia Romney +3.3 Obama +8.0 Wisconsin Obama +6.9 Wisconsin Obama +0.7
The FiveThirtyEight Philosophy Basic Philosophical Principles: 1. Look for consensus in the data 2. Update frequently Today is the first forecast of the rest of your life 3. Quantify and measure uncertainty 4. Combine theory and empiricism. Data-driven, but not reality-blind 12
The FiveThirtyEight Philosophy Look for Consensus Many polls are better than one. Many economic indicators are better than one. 13
The FiveThirtyEight Philosophy Update Frequently Forecasts are updated each day as new data comes in. However, more often than not, it doesn t change much. Laypeople tend to overrate new + near data points (availability heuristic) Professionals sometimes get dug in and make the opposite mistake. 14
The FiveThirtyEight Philosophy Quantify the Uncertainty This seems to be key to people who have achieved gains in forecast accuracy: Weather forecasters Gamblers Incentives for most experts discourage this. 15
Quantify the Uncertainty Flood Prediction: 49 Levee: 51 16
Quantify the Uncertainty Margin of Error: ±9 Flood Prediction: 49 Levee: 51 17
Quantify the Uncertainty 18
The FiveThirtyEight Method Major Steps in Model 1. Calculate polling average in each state 2. Adjust polling average 3. Combine polling model with economic conditions 4. Simulate Electoral College outcomes 19
The Economy and Elections Weight assigned to economic component declines as Election Day approaches. 20
The Economy and Elections 21
The Economy and Elections 22
The Economy and Elections 23
The Economy and Elections 24
The Economy and Elections We use a composite economic index from seven variables: 1. Jobs (payrolls) 2. Inflation (CPI) 3. Stock prices (S&P 500) 4. Industrial production 5. Consumption (PCEs) 6. Personal income 7. GDP (forecasted) 25
The Economy and Elections We use a composite economic index from seven variables: 1. Jobs (payrolls) 2. Inflation (CPI) 3. Stock prices (S&P 500) 4. Industrial production 5. Consumption (PCEs) 6. Personal income 7. GDP (forecasted) 26
The FiveThirtyEight Method Wide variance in pollster quality Only 10% of people respond to polls! Many pollsters miss cellphone voters Pollsters cheat/herd off one another Certain polling firms have persistent and predictable partisan biases. Polling is like golf: there s more you can do wrong than right. 27
The FiveThirtyEight Method When is there a real trend in the race? Textbook example of distinguishing signal from noise. Best way is to look at intra-poll movement, e.g. the Gallup poll moved 2 points toward Romney. 28
The FiveThirtyEight Method Lowess smoother Fitted values 5 10 15 20 25-400 -300-200 -100 0 polluntil bandwidth =.9 29
The FiveThirtyEight Method Lowess smoother Fitted values 5 10 15 20 25-400 -300-200 -100 0 polluntil bandwidth =.05 30
The FiveThirtyEight Method 31
The FiveThirtyEight Method 32
The FiveThirtyEight Method 33
The FiveThirtyEight Method 34