Hillary Clinton and Donald Trump make for one of the most — if not the most — controversial electoral matchups in American history. As a result, polls will be making headlines on a daily basis, often with conflicting results. National surveys, polls for all 50 states, and estimations of who has the edge in the decisive electoral college will come fast and furious.
With the Presidential Poll Tracker, the CBC will cut through the noise to make some sense of one of the biggest news stories of the year.
As with the CBC Poll Tracker during the 2015 federal election, the Presidential Poll Tracker is based on weighted averages of polls to determine the overall picture of the presidential election as well as how that is likely to break down at the state-level, where the election will be decided.
Below you'll find a full explanation of the methodology used to make these estimations.
The Presidential Poll Tracker will be updated daily throughout the campaign. And we'll add new features in the coming weeks.
The national popular vote
Though the popular vote does not determine the winner of the election, it is an important measure of where the race stands. It is also a useful barometer in determining which candidate has momentum and which campaign is floundering.
All mainstream national polls conducted and published in the United States are used to calculate the national polling average. The polls are sourced from the Huffington Post's Pollster, which uses a set of criteria to determine which polls are legitimate and which ones are not. The list of polls compiled by Pollster is exhaustive, and less restrictive than the polls included in other U.S. aggregators, such as RealClearPolitics.
These polls are entered into the model, and weighted according to three factors:
- Date: New polls are awarded more weight than older polls, and the rate of decay increases as the election approaches.
- Sample size: Polls with larger sample sizes are weighted more heavily than those with smaller sample sizes. The weighting is determined by what would be the margin of error of the sample of decided voters, if that sample was probabilistic.
- Pollster rating: The reliability of the pollster is determined by FiveThirtyEight.com's authoritative Pollster Ratings.
Polls are also adjusted to exclude undecided voters. Unlike their Canadian counterparts, who typically report numbers "among decided voters," most American pollsters include undecided voters in the top line numbers. But the methodology is inconsistent: some choose to include undecideds only, while others break out those who say they aren't voting or those who refuse to answer. And some exclude undecideds entirely.
As a result of differing methodologies, the proportion of voters who say they do not intend to vote for Hillary Clinton, Donald Trump, or another candidate can vary dramatically, from a few points to upwards of two dozen.
In order to better compare apples to apples, the weighted polling average removes these undecideds from reported polling results and distributes undecideds according to the voting intentions of decided voters in the survey. In other words, if a candidate has 55 per cent support among decided voters, the model assumes that candidate has 55 per cent support among undecided voters too.
This makes for a closer approximation of election night results.
Poll results that include the Libertarian candidate Gary Johnson are given priority over poll results that do not include Johnson. (The Poll Tracker, however, displays support for all third party candidates combined — which can include Green Party candidate Jill Stein and undefined "Others.")
The winner of the presidential election is determined by the electoral college.
Each state is awarded a number of electoral college votes equal to the state's number of members of Congress. Washington, D.C. also receives a number of electoral college votes. With the exception of Maine and Nebraska, the winner of the popular vote in each state is awarded the entirety of the state's electoral college votes — in other words, they are winner-take-all.
This makes the outcome in each of the 50 states (and Washington, D.C.) more important than the popular vote. For example, George W. Bush won the 2000 presidential election despite losing the popular vote because he won the electoral college.
The projection of the results in each state is based on two elements: polls and a uniform swing model. State-level polls are handled in the same way as national polls.
In many states, however, only a few polls will be conducted, if any. In order to project the outcome in these states, a uniform swing model is employed as well. This shifts the results in each state from the 2012 presidential election according to how the national polling average differs from the national results of the 2012 vote.
For example, if Clinton is down two points nationwide from Barack Obama's performance in 2012, she will be estimated to do two points worse than Obama did in each state.
The weight of the uniform swing projection is based on the number of national polls included in the model. When polls for a given state are available, they are given at least as much weight as the uniform swing model no matter how the polls would otherwise be weighted.
In addition, a small bonus worth two percentage points is given to the Republicans in Indiana and the Democrats in Virginia to reflect the vice-presidential candidacies of Mike Pence and Tim Kaine, respectively.
Electoral college estimations
The state-level projections are used to determine how likely a candidate is to win that state. This is determined by the projected margin between the two candidates. Based on experiments with the model for the last three presidential elections, it is estimated that the model will project the margin in any state to within 3.7 points a little over two-thirds of the time.
States projected to be won within 3.7 points are considered "Lean" states, and are used to determine the high and low ranges for the electoral college projection. States projected to be won by between 3.7 and 8.2 points are considered "Likely" wins for either the Democrats or Republicans.
In only 5 per cent of cases is the projection model expected to miss the margin of victory by over 8.2 points. "Safe" states are those projected to be won by 8.2 points or more, with 95 per cent confidence.
Using polls conducted during the last three presidential elections, this methodology would have projected the correct winner in each case, with the state-level margin of victory in each state being projected to within an average of three points, missing only one state in the 2004 and 2012 elections and three states in 2008. In those five misses, the average error on the projected margin of victory was 1.1 points.