Webinar
Election polling: why is it so difficult?
- Forecasting relies on the premise that polling accuracy improves as elections near, driven by voters crystallizing preferences and reducing uncertainty, though 100% accuracy is theoretically impossible.
- Historical analysis indicates that averaging polls with margin-of-error adjustments provides superior understanding of electoral dynamics compared to individual polls, a method refined since George Gallup's 1936 success.
- For the 2022 French presidential election, the approach involves a chained, step-by-step model that separately averages polls for the two rounds and runs 10 million simulations to quantify uncertainty.
- Simulations conducted approximately 10 weeks before the election (February 5, 2022) indicated Emmanuel Macron reaching the second round in 92% of 100 trials and winning 88% of direct matchups, with his overall victory probability estimated at 80% two months prior.
- Reliability is quantified by historical data showing polls can deviate by roughly four points 100 days before an election, necessitating the adjustment of models based on the specific quality of available polling or historical fundamentals.
- Models are inherently limited by the quality of input polls and cannot account for extreme unforeseen events such as pandemics or wars, with outcomes varying across simulation runs to reflect the range of potential realities.