Blending Polls and Fundamentals Improves Multi-Party Election Forecasts

What the Study Asks

Marcel Neunhoeffer and Sebastian Sternberg develop a forecasting approach for multi-party elections that asks whether combining published pre-election polls with fundamentals-based models yields better predictions and better-calibrated uncertainty than fundamentals-only forecasts.

How the Model Works

The authors propose a dynamic Bayesian forecasting model that explicitly accommodates multiple parties and the uncertainty that comes from poll sampling and temporal dynamics. The model fuses two information sources: (1) published pre-election public opinion polls and (2) signals from fundamentals-based forecasting models. By working in a Bayesian framework the model produces full probability distributions for election outcomes—not just point estimates—allowing direct statements about quantities such as the probability a party wins a plurality or that a particular coalition secures a parliamentary majority.

Testing the Model

Neunhoeffer and Sternberg present two ex ante forecasts for elections held in 2017 to evaluate performance. The evaluation compares the combined Bayesian model to forecasts based only on fundamentals, focusing on two aspects:

• Predictive accuracy (how close point forecasts are to realized vote shares), and
• Calibration of uncertainty (whether the model's probability intervals appropriately reflect realized outcomes).

Key Findings

• The combined dynamic Bayesian model outperforms fundamentals-only forecasting in terms of accuracy.
• It also delivers better-calibrated measures of uncertainty, meaning its probability statements and credible intervals better match what actually happened.
• The approach is designed for general use in multi-party settings wherever historical and current polling data are available.

Practical Takeaway

The paper offers a practical forecasting tool for analysts and scholars studying multi-party elections: blending current polling with fundamentals in a dynamic Bayesian framework improves both point forecasts and the reliability of probabilistic claims about pluralities and coalition majorities. The model is applicable across multi-party systems given sufficient polling history and up-to-date poll releases.