---
license: other
license_name: research-use-only
license_link: LICENSE
language:
  - de
pretty_name: German Name Frequencies by State & District (D-Info 2005)
size_categories:
  - 100K<n<1M
tags:
  - onomastics
  - surnames
  - forenames
  - germany
  - demographics
  - historical
configs:
  - config_name: surnames_district
    data_files: surnames_district.csv
  - config_name: forenames_district
    data_files: forenames_district.csv
  - config_name: surnames_state
    data_files: surnames_state.csv
  - config_name: forenames_state
    data_files: forenames_state.csv
  - config_name: distinctive_surnames_state
    data_files: distinctive_surnames_state.csv
  - config_name: distinctive_forenames_state
    data_files: distinctive_forenames_state.csv
---

# German Name Frequencies by State & District (D-Info 2005)

Regional frequency of **surnames and forenames** in Germany, from the *D-Info
2005* telephone-directory CD-ROM (klickTel, data status 02.06.2005), at two
administrative levels aligned with census-2022 geography:

- **State** = *Bundesland* — the 16 federal states.
- **District** = *Landkreis / kreisfreie Stadt* — the 377 districts, keyed by
  their 5-digit *Kreisschlüssel* (AGS).

For every name each table gives its number of 2005 telephone listings in a
region and the conditional probability `P(name | region)`. **Aggregate
statistics only** — no individuals, addresses or phone numbers.

## Subsets (configs)

There is **no default/combined config** — load a subset by name:

```python
from datasets import load_dataset
ds = load_dataset("stefan-it/d-info-2005-names", "surnames_district", split="train")
```

| Config                        | Level    | Threshold | Rows    | Content                           |
|-------------------------------|----------|-----------|---------|-----------------------------------|
| `surnames_district`           | District | N ≥ 10    | 170,642 | P(surname \| Landkreis)           |
| `forenames_district`          | District | N ≥ 10    |  89,497 | P(forename \| Landkreis)          |
| `surnames_state`              | State    | N ≥ 20    | 124,109 | P(surname \| Bundesland)          |
| `forenames_state`             | State    | N ≥ 20    |  19,453 | P(forename \| Bundesland)         |
| `distinctive_surnames_state`  | State    | N ≥ 20    | 124,109 | surnames ranked by regional lift  |
| `distinctive_forenames_state` | State    | N ≥ 20    |  19,453 | forenames ranked by regional lift |

## Schema

**District** subsets (`*_district`):

| Column          | Type   | Meaning                                                 |
|-----------------|--------|---------------------------------------------------------|
| `district`      | string | 5-digit AGS *Kreisschlüssel* (e.g. `01001` = Flensburg) |
| `district_name` | string | district name                                           |
| `state`         | string | ISO 3166-2:DE code of the district's Bundesland         |
| `name`          | string | surname / forename (UTF-8; umlauts & ß preserved)       |
| `count`         | int    | listings carrying the name in that district             |
| `prob`          | float  | `count / observations-of-that-kind in that district`    |

**State** subsets (`*_state`): columns `state, name, count, prob` (same
meaning, region = Bundesland).

**Distinctive** subsets add two columns and are sorted by `lift` descending
within each state:

| Column       | Type  | Meaning                                            |
|--------------|-------|----------------------------------------------------|
| `p_national` | float | `P(name)` — the name's share nationwide            |
| `lift`       | float | `prob / p_national` — regional over-representation |

State ISO codes: `SH` Schleswig-Holstein, `HH` Hamburg, `NI` Niedersachsen,
`HB` Bremen, `NW` Nordrhein-Westfalen, `HE` Hessen, `RP` Rheinland-Pfalz,
`BW` Baden-Württemberg, `BY` Bayern, `SL` Saarland, `BE` Berlin,
`BB` Brandenburg, `MV` Mecklenburg-Vorpommern, `SN` Sachsen, `ST` Sachsen-Anhalt,
`TH` Thüringen.

## The numbers

$$
P(\text{name} \mid \text{region}) = \frac{\text{count}(\text{name},\ \text{region})}{\text{observations of that kind in region}}
$$

- **Surnames** are conditioned on all listings in the region (≈ every listing
  has a surname).
- **Forenames** are conditioned on listings whose forename is known (2005 lists
  many people by initial only), and a forename is its **first given name** —
  compound names are aggregated to word 1, so **no row contains a space** and
  there is no separate compound table.

**Distinctive names** measure regional concentration rather than raw frequency
(Müller is #1 in every state and so tells you nothing regional):

$$
\text{lift}(\text{name}) = \frac{P(\text{name} \mid \text{state})}{P(\text{name})} = \frac{\text{share of the name in the state}}{\text{share of the name nationwide}}
$$

`lift = 1` → as common regionally as nationally; `lift ≫ 1` → concentrated in
the state → *typical* of it. Rows are ranked by `lift`. Because the only floor
is the k-threshold (N ≥ 20), the very top of each state list is **rare-but-
highly-localised** names; for *common-and-typical* names (e.g. München-region
Huber, Bavarian Aigner) filter on a higher `count` (say ≥ 200) before reading
off the ranking. Example top-lift forenames: **BY** → Emmeran, Kunigunda,
Kreszenz (Bavarian-Catholic); **SN** → Rico, Anett, Sylke (East-German).

## Privacy

**Per-cell k-anonymity.** A `(name, region)` row is published **only if its own
count meets the threshold** — N ≥ 10 for districts, N ≥ 20 for states — so every
published cell represents at least that many people sharing the name in that
region and no individual can be singled out (GDPR Recital 26). A national-total
floor alone is *not* sufficient at district level: a name common nationwide can
still be count = 1 in one district. Only aggregate counts are released.

## Provenance & licence

Source: *D-Info 2005* CD-ROM, klickTel (data status 02.06.2005). Only aggregate,
k-anonymized frequency statistics are published here. Released under a custom
**research-use-only** term (see `LICENSE`): use for non-commercial research,
with attribution, and no attempt at re-identification.