Usage Notes

GERDA collects election results that were scattered across federal and state sources and reported in inconsistent formats, puts them in one standardized format, and maps historical results onto consistent municipal boundaries despite administrative changes.

How to Use the Data

Data sources and processing steps are described in the accompanying paper.

Denominators, missingness and result versions

Bayern municipality data uses Gesamtstimmen (first plus second votes from 1950); Hamburg and Bremen use multi-vote totals under their five-vote systems. These counts measure votes. Turnout measures people: number_voters / eligible_voters. Do not divide stored vote totals by two or five, or impose a universal valid_votes + invalid_votes == number_voters rule.

Imputed weights: state harmonization fills missing valid_votes with positive voters, then positive electorate, then a unit weight. For Bremen 1991/1995, share * valid_votes therefore gives proxy counts, not observed party votes. Harmonized counts are rounded to integers, which can slightly change source percentages. Completeness tables count nonmissing stored values, including these proxies.

Result versions also differ. For federal 2021, federal_cty_unharm uses the original certified result. federal_wkr_unharm uses the result including the February 2024 Berlin repeat election. Changing a denominator does not align these versions.

Missing is not zero. NA in RDS and empty cells in CSV/Excel mark unavailable values. sum(x, na.rm = TRUE) gives zero when every value is missing. Preserve an all-missing sum as NA, and report how many observations are known. A partial voter sum divided by a complete electorate is not statewide turnout.

Select party columns explicitly. Participation, turnout, identifiers, covariates and flag_* diagnostics are not parties. cdu_csu, far_right, far_left and far_left_w_linke overlap individual parties and must be excluded from individual-party sums. Use the state column schema and example for the selected file. Party aliases can differ across files; similar names do not justify merging distinct regional lists.

Opening in Excel / In Excel öffnen

The Excel (.xlsx) files on the download page open directly, including in Excel 2019, with the source rows, column names, and values unchanged. Geographic identifiers are text, vote shares are proportions displayed as percentages (the display does not round the stored value), and missing values are empty cells. Excel keeps about 15 significant digits, so use CSV or RDS when you need the original precision. Large tables take a while to open; tables longer than Excel’s row limit continue on further sheets with the same headers.

To import a CSV in Excel 2019:

1. Select Data → From Text/CSV and choose the file. Select UTF-8 (65001) and Comma as the delimiter.
2. Choose Transform Data. In Power Query, remove the automatically added Changed Type step, if present, before assigning column types. This prevents Excel from removing decimal points or leading zeros before you can choose the correct settings.
3. Set identifiers such as ags, ags_21, county, county_code, state and wkr_nr to Text. For numeric columns, choose Change Type → Using Locale… → Decimal Number → English (United States). Set date columns to Date where appropriate.
4. Select Close & Load. Format vote-share columns as Percentage if desired. A value of 0.283533 means 28.3533%; federal_muni_raw instead carries party vote counts.

For further details, see Microsoft’s guide to data types and locale settings.

If a CSV is saved with a .txt extension, rename it to .csv or select it through From Text/CSV. Renaming alone does not correct values that Excel has already misinterpreted; re-import the original download.

Federal Elections

State Elections

Constituency (Wahlkreis) Elections

Municipal Elections

European Elections

Mayoral Elections

Landrat Elections

County Elections

Harmonization

Harmonized files map every election onto fixed municipal and county boundaries, using official crosswalks to track mergers, splits, and boundary shifts.

• Population-weighted aggregation: Where multiple municipalities merged, votes are aggregated to the new municipality’s boundaries using population-based weighting.
• Mail-in vote allocation: For mail-in voting districts shared by multiple municipalities, mail-in votes are allocated proportionally based on the number of polling-card voters in each municipality.

Meinungsbild

Meinungsbild gives public opinion estimates for 43 policy issues at three levels (federal states, electoral districts, counties), estimated with Multilevel Regression and Poststratification (MRP); methods and the interactive map are on the Meinungsbild page. The estimates are for exploration and description only: MRP can be unreliable for small units or issues with little survey data, and the estimates are not ground truth. The Meinungsbild data is not available for download.

Code Availability

The code that generates the datasets is in the code folder of our GitHub repository; the scripts carry further instructions.

Authors

Andreas Wiedemann, Hanno Hilbig, Vincent Heddesheimer, and Florian Sichart.

Acknowledgements

We thank Cornelius Erfort, Sascha Riaz and Moritz Marbach for helpful comments. We also thank the anonymous reviewers at Scientific Data for their constructive feedback. Thanks to Daniela Gaus, Maurice Baudet von Gersdorff, and Luca Schenk for excellent research assistance and Victor Kreitman for providing code and data on election dates.

@article{Heddesheimer2025GERDA,
   author = {Vincent Heddesheimer and Hanno Hilbig and Florian Sichart and Andreas Wiedemann},
   doi = {10.1038/s41597-025-04811-5},
   issn = {2052-4463},
   issue = {1},
   journal = {Scientific Data},
   month = {4},
   pages = {618},
   title = {GERDA: The German Election Database},
   volume = {12},
   url = {https://www.nature.com/articles/s41597-025-04811-5},
   year = {2025}
}