Methodology

Estimates are produced using Multilevel Regression and Poststratification (MRP), a statistical technique that combines survey data with census demographics to produce reliable subnational opinion estimates even for small geographic areas.

Survey Data

~118,000 survey responses aggregated from five German survey programs: GLES Tracking, GLES Cross-Section 2025, GLES RCS 2025, GLES Cumulation, and ALLBUS (2009–2025).

Geographic Coverage

Estimates at three levels: 16 federal states (Bundesländer), 299 electoral districts (Wahlkreise), and 401 counties (Kreise).

Validation

MRP estimates validated against direct state-level survey estimates with median correlation r = 0.899 and median RMSE = 5.5 percentage points.

Read more about the methodology

MRP Approach

Multilevel Regression and Poststratification (MRP) is a two-step process for estimating subnational public opinion from national surveys:

1. Multilevel Regression: A multilevel logistic regression model (fitted with lme4::glmer() in R) predicts individual-level survey responses based on demographic characteristics (age, gender, education), geographic random effects (state, electoral district, county), and contextual covariates (AfD vote share, CDU vote share, turnout, population density). The model includes deep two-way demographic interactions following Ghitza and Gelman (2013).
2. Poststratification: The model predictions are weighted by the actual demographic composition of each geographic area using Census 2022 data, producing population-representative estimates for each region.

Survey Sources

Survey	Respondents	Period
GLES Tracking	~52,336	2009–2023
GLES Cross-Section 2025	~7,337	2025
GLES RCS 2025	~8,561	2025
GLES Cumulation	~21,040	2009–2021
ALLBUS	~29,112	2023–2024

Variable Harmonization

Each of the 43 policy issues is harmonized across survey programs. Variable-specific concordance tables map different question wordings and response scales onto a common binary scale. The issue_concordance.csv file documents the exact mapping for each issue across all five surveys.

Validation

MRP estimates are validated by comparing state-level MRP predictions against direct survey estimates (disaggregated means from surveys with sufficient state-level sample sizes). Validation shows a median correlation of r = 0.899 and median RMSE of 5.5 percentage points across all 43 issues. Top-performing issues reach correlations above 0.98 (e.g., ukraine_arms: r = 0.993, rent_control: r = 0.992).

References

• Ghitza, Y., & Gelman, A. (2013). Deep interactions with MRP: Election turnout and voting patterns among small electoral subgroups. American Journal of Political Science, 57(3), 762–776.
• Selb, P., & Munzert, S. (2011). Estimating constituency preferences from sparse survey data using auxiliary geographic information. Political Analysis, 19(4), 455–470.
• Warshaw, C., & Rodden, J. (2012). How should we measure district-level public opinion on individual issues? The Journal of Politics, 74(1), 203–219.
• Gao, Y., Kennedy, L., Simpson, D., & Gelman, A. (2021). Improving multilevel regression and poststratification with structured priors. Bayesian Analysis, 16(3), 719–744.
• Goplerud, M. (2024). Re-evaluating machine learning for MRP given the comparable performance of (deep) hierarchical models. American Political Science Review, 118(1), 529–536.
• Heddesheimer, V., Hilbig, H., Sichart, F., & Wiedemann, A. (2025). GERDA: The German Election Database. Scientific Data, 12, 618.

Raw survey data must be obtained separately from GESIS due to licensing restrictions. The MRP code and issue definitions are available in the Meinungsbild folder of the GitHub repository.