Polling: Voting Intention
GB Polling Average
Methodology
1. Accuracy Weighting 🎯
Every pollster is assigned a score derived from their historic accuracy. This is displayed as a 0–100 scale in the coloured circles beside each pollster. Grey circles (50) indicate a new pollster without a track record, assigned the score equivalent to an average error of 3%.
2. Frequency Balancing ⚖️
To prevent high-volume pollsters from skewing the average through sheer quantity, weights are adjusted for frequency. If a firm publishes multiple times within a 10-day window, their total weight is distributed across those polls.
3. Time-Based Kernel Smoothing 📅
The trendline uses a Time-Based Kernel Average (a 35-day window).
4. Outlier Rejection (Bisquare) 📉
A robust two-pass system measures how far a poll deviates from the initial trend. Extreme outliers have their influence drastically reduced or zeroed out.
Trendline
Data Filters
Model Parameters
35 d