The calculation of the PollyVote’s electoral vote forecast follows the same principle idea as the popular vote prediction. That is, we combine forecasts within and across different component methods to determine the winner of each state.
As shown in Table 1, the component methods are the same as in the popular vote forecast (except for index models, for which no forecasts are available).
TABLE 1: FORECAST DATA PER COMPONENT | |||
DATA AVAILABLE ON | |||
METHOD | WINNER | VOTE SHARE |
CHANCE OF WINNING |
I. Polls | Y | Y | Y |
+ 538 (polls-only) | Y | ||
+ NYT Upshot | Y | ||
+ PEC | Y | Y | |
+ Pollster | Y | Y | |
+ YouGov | Y | ||
+ RCP | Y | ||
+ Votamatic | Y | ||
+ Electoral-vote | |||
II. Models | Y | Y | Y |
+ DailyKos | Y | ||
+ DeSart | Y | ||
+ DeSart & Holbrook | Y | ||
+ Jérôme & Jérôme | Y | ||
+ Crosstab | Y | Y | |
III. Expert judgment | Y | Y | |
+ Cook Report | Y | ||
+ PollyVote expert survey | Y | Y | |
+ Rothenberg | Y | ||
+ Sabato’s Crystal Ball | Y | ||
IV. Prediction markets | Y | Y | |
+ Predictit | Y | ||
+ Predictwise | Y | ||
V. Citizen forecasts | Y | ||
+ US Election Atlas | Y |
Note, however, that the types of predictions that we were able to collect differ across the individual component forecasts. For some components, we only have predictions of who will win a particular state (e.g. 3 of the 4 components in the expert judgment category). For others, we have either a vote share forecast (e.g., Pollster) or a winning probability (e.g., NYT Upshot). Finally, there are some components that provide both a vote share forecast and winning probabilities (e.g., PEC).
Here is the procedure we use to combine the different types of forecasts within and across component methods in order to determine an election winner per state:
- For each individual component forecast, we derive a prediction of who will win a particular state.
- Where available, we collect state-level predictions of the candidates’ vote shares (e.g., Pollster) and winning probabilities. The candidate with a value above 50% is predicted to win that state.
- Some components neither provide vote shares nor winning probabilities. They only provide a prediction of who will win.
- We combine forecasts within each component method.
- We count the predicted wins for each candidate from forecasts within that component method.
- We average vote share forecasts (where available) from forecasts within that component method.
- We average winning probabilities (where available) from forecasts within that component method.
- For each component method, we determine the predicted winner in each state using the following heuristic
- The combined component forecast is that the candidate with more predicted wins (II.1) will win that state.
- If both candidates accumulate the same number of wins, the candidate with the higher average vote share prediction (II.2) is predicted to win.
- If the average forecast is that both candidates will achieve 50% of the vote, the candidate with the higher chance of winning (II.3) is predicted to win.
- If both candidates’ average winning probability is 50%, the component method predicts a tie.
- We combine forecasts across the five component methods.
- We count the predicted wins for each candidate from each component method.
- We average vote share forecasts (where available) from forecasts across the five component methods.
- We average winning probabilities (where available) from forecasts across the five component methods.
- We determine the predicted winner in each from the PollyVote
- The combined PollyVote forecast is that the candidate with more predicted wins (IV.1) will win that state.
- If both candidates accumulate the same number of wins, the candidate with the higher average vote share prediction (IV.2) is predicted to win.
- If the average forecast is that both candidates will achieve 50% of the vote, the candidate with the higher chance of winning (IV.3) is predicted to win.
- If both candidates’ average winning probability is 50%, the PollyVote predicts a tossup.
- The color coding for each state in the electoral map is based on the following rule:
- If the combined forecasts for each component predict the same candidate to win, then this state is coded as “safe.”
- If 3 of 4 methods predict the same candidate to win, then code this state as “leaning.”
- If the ratio of predicted wins is 50-50 (Ratio = 0.5), then
- Code as state as leaning towards the candidate with the higher average vote share (IV.2)
- If the average forecast is that both candidates will achieve 50% of the vote, code as the state as leaning towards the candidate with the higher average winning probability (IV.3).
- If both candidates’ average winning probability is 50%, code the state as a tossup.