About estimador.pt
Independent, data-driven election forecasting for Portugal
Our Mission
estimador.pt provides transparent, probabilistic forecasts for Portuguese elections. We believe democratic societies benefit when citizens have access to rigorous, unbiased analysis of electoral data—not just punditry and speculation.
Our goal is to bring the standards of modern election forecasting—pioneered by projects like FiveThirtyEight and The Economist—to the Portuguese context, while adapting our methods to Portugal's multi-party proportional representation system.
What We Do
Election Forecasting Bayesian models that combine polling data with historical patterns, accounting for uncertainty at every step.
Seat Projections District-level analysis using the D'Hondt method to project how national vote shares translate to parliamentary seats.
Trend Analysis Gaussian Process models that track the evolution of party support over time, separating signal from noise.
Polling Analysis Systematic estimation of pollster house effects—the consistent biases each polling firm shows toward certain parties.
Our Approach
We use Bayesian statistical methods that explicitly quantify uncertainty. Unlike point predictions, our forecasts produce probability distributions that reflect what we actually know—and don't know—about election outcomes.
Key principles:
- Transparency: Open methodology with detailed documentation of our models and assumptions
- Independence: No political affiliations, no partisan funding, no editorial pressure
- Accuracy: Continuous validation against actual election results, with honest assessment of forecast errors
- Accessibility: Presenting probabilistic forecasts in ways that general audiences can understand and use
About the Project
estimador.pt was created by Bernardo Caldas as an independent project combining Bayesian statistics, computational methods, and analysis of Portuguese electoral data.
For questions, suggestions, or media inquiries, please contact us at info@estimador.pt.
Limitations & Disclaimers
Election forecasting is inherently uncertain—that's why we present probabilities, not predictions. Our models account for several sources of uncertainty:
- Polling error: Systematic and random polling error, including potential bias not captured by past elections
- Campaign dynamics: Late-breaking news and campaign events that can shift voter preferences
- Turnout: Differential turnout patterns that may not match historical patterns
- Regional variations: District-level variations that deviate from national trends
A forecast showing a party with a 70% chance of winning most seats also means there's a 30% chance they won't. Our forecasts should be interpreted as informed probabilistic assessments, not certainties.