Our objective is to assess the accuracy of polling, not to predict the outcome of the election. The values depicted above intend to represent an aggregate view of state-level public polling. Our values can be thought of as the polling implied outcome; if polling is correct, the election's outcome should match our values.

Each state's outcome is treated as an independent event as is each candidate's percentage. In reality this is not the case. Outcomes are most certainly correlated based upon empirical information like geography and demographics. Our view, which is consistent with letting polling tell the story, is that any such correlation should be observable and expressed through polling.

Candidate totals are calculated using a quadratic local regression with an alpha of 80%; if too few polls are available a least-squares regression is used. The calculated value for each candidate’s regression, along with the variance, is used as input to calculate the probability that a given candidate's total will be greatest. A standard distribution is used to determine the likelihood of victory in a given state. Once each state's probability is known, a 51-dimension cartesian product is generated to assess the likelihood of receiving 270 or more electoral votes.

More thorough information containing technical details and our data collection criteria is available in our methodology series.