A new website, “Poll Aggregator” has been launched, featuring all voting intention polls for parties with parliamentary representation collected from credible and verified public sources. Unique to this aggregator are new features such as forecasts of seat distribution based on the electoral efficiency of each party, based on the aggregated voting intention estimates. Additionally, the site includes a matrix detailing vote transfers between political parties.

“With this initiative, we aim to enhance democratic participation and clarify public opinion. It’s another academic project designed to make a real impact”, explains Henrique Oliveira, a professor in the Department of Mathematics at Técnico and one of the project’s leaders alongside Alexandre Francisco, a professor in the Department of Computer Science and Engineering at Técnico. “This initiative offers a rigorous, transparent and accessible tool for monitoring public opinion, combining solid foundations with clear presentation”, adds Henrique Oliveira.

The aggregator will be updated with the polls for the legislative elections on 18 May 2025. The project aims to extend the model to all elections that may arise, with methodological adjustments appropriate to each context.

The project is based on research conducted at Instituto Superior Técnico, namely in Applied Mathematics (dynamic models, statistical inference) and Computer Engineering (interactive visualisation, data automation and structuring). It is aligned with themes of time series analysis and the fusion of imperfect information.

Key features of the aggregator include:

• Each survey is weighted according to its date (with more recent polls carrying greater weight) and sampling precision;
• The significance of older polls decreases over time, reflecting process noise and decreasing relevance. The aggregator will update daily until the elections, even if no new polls are available, to account for increased error over time;
• New data is automatically integrated into the model as it becomes available;
• Statistical smoothing (using Laplace) is applied to compensate for small sample sizes or extreme results;
• Daily estimates and their 95% confidence intervals are calculated, with the error increasing as time passes without new polls;
• The expected number of parliamentary seats will be estimated, adjusted for the composition of the Portuguese Parliament, using an algorithm that assesses the efficiency of each party’s votes.

Only verified data is included, and statistical inference techniques ensure an accurate reading of electoral trends. The quality of the aggregator depends on the reliability of the various estimates and methodologies employed by the organisations conducting the polls.