America First Insight Pollster Scorecard
The America First Insight (AFI) Pollster Scorecard is designed to offer a transparent evaluation of pollsters based on their accuracy and impartiality. This scorecard utilizes three key metrics to grade each pollster:
- Average Miss: This metric calculates how far off, on average, a pollster's predictions were from actual election outcomes. It employs a weighted average derived from polls conducted within the 90 days leading up to the elections in 2018, 2020, 2022, and 2024. The weighting considers both the proximity to election day and the competitiveness of the polled race. For instance, a late October poll in Wisconsin holds more weight than an early September poll in Wyoming.
- Average Bias: Using the same methodology as the Average Miss, this metric assesses whether a pollster consistently overstates support for one political party over another. Identifying bias is crucial as it indicates if a pollster systematically favors one party, thereby skewing public perception.
- Percentage of Polls Within Three Points of the Final Result: This measures the consistency of a pollster's predictions. A poll within three percentage points of the actual result is considered accurate, reflecting a standard margin of error for polls with a sample size of approximately 1,000 voters.
To compute the overall score, these metrics are combined into a weighted average. While all metrics contribute, the Average Miss and Average Bias are given greater weight due to their direct impact on poll reliability and integrity.By providing these detailed evaluations, the AFI Pollster Scorecard aims to increase transparency in polling data, helping the public better understand and trust the polls they encounter.
Using a near identical method to normal pollster scoring, the main difference in aggregate scoring is two fold. We less the values of Average Misses, and Average biases, and we made the margins for a "good" poll more strigent. Decreasing a good poll from within 3 points to within 1.5 points. We also care about calling the right person. Polling Aggregates are supposed to be all about getting races correct, unlike normal pollsters, by having dozens maybe hundreds of polls, they should be nearly uneffected by random variance to such a high degree as a single poll. Thus most polling aggregates are similarly graded, to whichever pollsters are dominating their charts.