A new online aggregator has been launched, combining polls from various companies to provide more accurate results. According to Henrique Oliveira, the responsible professor for this initiative, this approach reduces errors by enlarging the sample size through the combination of multiple polls.
“The aggregator creates a kind of a film, showing the evolution over time, allowing us to discern trends,” stated Henrique Oliveira in an interview.
The tool uses statistical smoothing (Laplace) to adjust for small sample sizes, delivering daily estimates with confidence intervals reaching 95%.
Henrique Oliveira addressed some issues with polling, notably sample bias, citing fixed-line phone surveys that tend to reach older populations, thereby favoring parties with older electorates.
By merging various polls, according to the mathematician, “this bias disappears” because it’s “compensated by each other,” resulting in an aggregator with “less bias than individual polls.”
This tool also forecasts the number of mandates each party may obtain by considering the historical vote utilization of each party, acknowledging the lack of individual polls across the 22 electoral districts.
