German Municipal Broadband Availability, 2005-2021

Current version: v3.0.0 (July 2026). If you downloaded an earlier version, please update: v3.0.0 removes trailing non-reporting zeros in 2019-2021 (mostly gemeindefreie Gebiete dropping from full coverage to 0% at 2021) and recovers one previously dropped unit; v2.1.0 added tier_* columns recording which reported speed tier feeds each share variable (the named tiers are not reported in all years; see the caveats) and corrected codebook errors; v2.0.0 removed a large block of false zeros in 2010-2014, added Berlin and Hamburg, and recovered previously dropped municipalities. See CHANGELOG.md for what changed between versions and who is affected.

This repository provides a harmonized panel of broadband internet availability for German municipalities. The public file is output/panel_data_public.csv.

The data should be used with care. The source files come from three historical Breitbandatlas data packages, reporting methods changed over time, and some historical AGS codes cannot be mapped to official 2021 municipal boundaries. This release is a best-effort harmonization of imperfect source data.

Versioning

Releases follow a semantic scheme: major versions change existing values or observations, minor versions add data or variables, patch versions change documentation only. Each version is a git tag and a GitHub Release with the panel CSV attached, and the current version is recorded in the VERSION file. All changes are documented in CHANGELOG.md.

v3.0.0 (July 2026)

This update removes trailing non-reporting zeros and recovers one previously dropped unit. The change is small and well-contained (141 municipality-years removed, 2 added, 1 value revised); the panel has 140,093 municipality-year rows.

• Trailing non-reporting zeros removed from 2015-2021. A group of units, predominantly gemeindefreie Gebiete and forest districts, report positive coverage through an earlier year and then exactly 0 across all speed tiers in a later year, overwhelmingly a 100% to 0% drop at 2021. Coverage does not physically vanish, so an all-tier-zero municipality-year for a unit that had positive coverage earlier is treated as non-reporting and removed (141 municipality-years, 115 of them in 2021). Persistent and leading zeros are retained. Full list in output/nonreporting_zero_blocks_2015_2021.csv.
• 03159501 ("Harz, gemeindefreies Gebiet") recovered for 2019-2020. It is reported under a pre-2016 Kreis code that a relaxed reform-chain fallback now maps to its 2021 boundary. The residual unmapped filter is now 9,726 rows (0.32%) over 121 codes.

v2.1.0 (July 2026)

This update makes the measurement tiers behind the share variables observable and corrects the codebook:

• The named speed tiers are not reported in all years. The share variables take the lowest reported tier at or above their named threshold, so before 2018 they measure stricter tiers than their names suggest: share_gte6mbps is >=50 Mbps coverage in 2010-2012 and >=16 Mbps in 2013-2017, share_gte30mbps is >=50 Mbps through 2017, and share_gte1mbps (and the 2010+ baseline) is >=2 Mbps through 2017. Earlier codebook versions incorrectly stated that these columns were missing before their named tier was reported.
• New tier_baseline, tier_gte1, tier_gte6, tier_gte30 columns record, per municipality-year, the tier (in Mbps) that produced each share value. All existing share values are unchanged.

v2.0.0 (July 2026)

This update recovers most of the data previously dropped during AGS standardization and removes a large block of false zeros:

• Non-reporting zeros removed from 2010-2014. About 6,600 municipalities per year carried exactly 0 across all speed tiers throughout 2010-2014 while averaging 87% baseline coverage in 2008 and 98% in 2015. These were non-reporting municipalities coded as zero in the historical source, not true zeros. The affected 33,138 municipality-years are removed; 2010-2014 now covers roughly 4,400-4,500 reporting municipalities per year. This resolves most of the previously documented 2015 break: the unweighted mean jump from 2014 to 2015 was 57 percentage points before the fix, while the within-municipality jump among continuous reporters is 1.2 percentage points.
• A nearest-year crosswalk fallback maps historical AGS codes delivered under a different boundary vintage than their year label (for example, Sachsen-Anhalt 2005-2006 files carry post-2007-Kreisreform codes). This recovers 1,084 of the previously unmapped 1,206 codes; the residual filter now removes 9,948 long rows (0.33% of the standardization input) covering 122 codes.
• Berlin and Hamburg are now included. Their absence was a bug in the construction of the Destatis 2021 reference (their AGS codes were mangled by scientific-notation formatting), not a property of the source data. Hamburg has city-level data for 2005-2021; Berlin for 2005-2008 and 2015-2021 (its 2010-2014 source rows were non-reporting zeros). Both cities' 2020-2021 values are aggregated from Bezirk-level data (see caveats).
• Municipality counts rise to 10,994 Destatis 2021 AGS codes; the panel has 140,232 municipality-year rows.
• Population weights are now applied consistently when historical constituents are aggregated to 2021 boundaries, which revises some published 2005-2007 values in Sachsen-Anhalt, Sachsen, and Thüringen (recovered constituent municipalities are now included in the population-weighted means).

v1.0.0 (May 2026)

Relative to the initial June 2025 upload, this release fixed several issues:

• 2005-2008 baseline coverage now uses the historical >=0.128 Mbps DSL measure instead of returning zeros.
• AGS codes are standardized to Destatis 2021 municipality boundaries.
• Invalid percentages outside [0, 100] are filtered rather than capped.
• Missing speed tiers are preserved as missing values instead of being filled with zero.
• Duplicate AGS-year-technology-speed cells are collapsed before border harmonization.

Main Caveats

• There is no 2009 municipality panel in the current source files.

• The share variables track the lowest reported tier at or above their named threshold, and the reported tiers change over time. The source files do not report the 1, 6, and 30 Mbps tiers before 2018, so each share is measured at a stricter tier than its name suggests in earlier years. The typical tier behind each column (recorded per observation in the tier_* columns) is:

  Column	2005-2008	2010-2012	2013-2017	2018-2021
  share_broadband_baseline	0.128	2	2	1
  share_gte1mbps	missing	2	2	1
  share_gte6mbps	missing	50	16	6
  share_gte30mbps	missing	50	50	30

  This creates definitional breaks at 2013 (share_gte6mbps switches from >=50 to >=16) and 2018 (all columns switch to their named tiers, coinciding with a source change): part of the level jumps at these years is definitional, not rollout. In 2010-2012, share_gte6mbps and share_gte30mbps are identical by construction. Analyses of the >=6 or >=30 tiers that span these years should condition on the tier_* columns or restrict to years with a constant tier.

• The baseline measure changes over time: share_broadband_baseline uses >=0.128 Mbps in 2005-2008, >=2 Mbps in 2010-2017, and >=1 Mbps from 2018 onward.

• 2010-2014 covers only reporting municipalities (roughly 4,400-4,500 per year). Municipality-years with exactly 0 across all speed tiers in this window were non-reporting coded as zero in the source and have been removed (see output/nonreporting_zero_blocks.csv for the full list). Reporters skew toward larger municipalities, so unweighted means over 2010-2014 describe a selected sample.

• The 2015 provider change (method_change_2015) remains flagged, but after the removal of false zeros the residual break is small: the within-municipality baseline jump from 2014 to 2015 is 1.2 percentage points, and higher-tier jumps are in line with adjacent-year rollout growth. The main effect of 2015 is the return to near-complete municipality coverage, i.e., a composition change.

• Trailing non-reporting zeros are removed from 2015-2021 (141 municipality-years, mostly gemeindefreie Gebiete dropping from full coverage to 0% at 2021; see output/nonreporting_zero_blocks_2015_2021.csv). 422 all-tier-zero municipality-years remain: units that are zero in every observed year (persistently zero, where a genuine zero is plausible) or zero only before their first positive year. These are retained.

• Step 06 filters 9,948 post-deduplicated long rows, 0.33% of the standardization input, because 122 historical AGS codes cannot be mapped defensibly to Destatis 2021 boundaries (mostly Bavarian unincorporated areas and sub-municipal district codes).

• Berlin and Hamburg values for 2020-2021 are unweighted means across their 19 Bezirke, because the source reports only Bezirk-level coverage in those years and contains no Bezirk household weights. All underlying Bezirk values lie between 97 and 100, so the approximation error is below 1 percentage point.

Details on filtered AGS codes are in docs/unmapped_ags_documentation.md.

Output Files

• output/panel_data_public.csv: public municipality-year panel.
• output/panel_data_with_treatment.csv: internal version with treatment indicators and event-time variables.
• docs/data_processing_pipeline.md: processing details and validation notes.
• output/average_annual_coverage_plot.png and output/large_yoy_changes_plot.png: diagnostic plots.
• output/county_gte30_coverage_map_2021.png: county-level 2021 map of >=30 Mbps coverage, aggregating municipality values with population weights.

Public Data Codebook

Variable	Type	Description
AGS	character	8-digit official municipality key, standardized to 2021 boundaries.
year	integer	Observation year. The panel covers 2005-2021, excluding 2009.
share_broadband_baseline	double	Share of households with access to basic broadband, measured at the lowest reported tier: >=0.128 Mbps in 2005-2008, >=2 Mbps in 2010-2017, >=1 Mbps from 2018. See tier_baseline.
share_gte1mbps	double	Share of households with access at the lowest reported tier >=1 Mbps (>=2 Mbps in 2010-2017, >=1 Mbps from 2018); missing when no such tier is reported (2005-2008). See tier_gte1.
share_gte6mbps	double	Share of households with access at the lowest reported tier >=6 Mbps (>=50 Mbps in 2010-2012, >=16 Mbps in 2013-2017, >=6 Mbps from 2018); missing when no such tier is reported (2005-2008). See tier_gte6.
share_gte30mbps	double	Share of households with access at the lowest reported tier >=30 Mbps (>=50 Mbps in 2010-2017, >=30 Mbps from 2018); missing when no such tier is reported (2005-2008). See tier_gte30.
tier_baseline	double	Speed tier in Mbps at which share_broadband_baseline is measured for this observation.
tier_gte1	double	Speed tier in Mbps at which share_gte1mbps is measured; missing when the share is missing.
tier_gte6	double	Speed tier in Mbps at which share_gte6mbps is measured; missing when the share is missing.
tier_gte30	double	Speed tier in Mbps at which share_gte30mbps is measured; missing when the share is missing.
method_change_2015	integer	Indicator for the 2015 methodological break.

All share variables are percentages on the 0-100 scale.

Attribution

The original data come from the Breitbandatlas / Gigabit-Grundbuch. When using this harmonized dataset, cite the original source and make clear that the data have been modified and harmonized.

Original source: Breitbandatlas | Gigabit-Grundbuch (https://gigabitgrundbuch.bund.de)