Data & methods: sources, formulas and level thresholds

Updated 2026-09-28

Every number on SHIZUAN is calculated from public data with the same fixed procedure. This page lists the data, the formulas, the actual thresholds between levels, and the limits you should keep in mind.

Sources

“Retrieved” is the date this site downloaded the file. All of it was processed and compiled by this site; none of the agencies listed produced this map.

Coverage and period

• Area: 5,309 neighborhoods (chome and similar small areas) of mainland Tokyo, excluding the islands. Boundaries are the 2020 census small areas. For 51 of them (Hinode-machi (28 areas), Hinohara-mura (13 areas), Shinjuku-ku (2 areas), Machida-shi (2 areas), Akiruno-shi (2 areas), Hino-shi (1 areas), Akishima-shi (1 areas), Ota-ku (1 areas), Kunitachi-shi (1 areas)) the police figures cannot be matched reliably to the neighborhood — they are published for a whole village or large district, or under a different or older town name — so both incident layers show “No data” there rather than zero. Rows that could not be matched are not counted anywhere.
• Incidents: the Tokyo Metropolitan Police Department's reported Penal Code offenses by town for 2023 to 2025 (final calendar-year figures), added together. The police grand totals were 89,098 in 2023 and 99,349 in 2025 (including rows for other prefectures, overseas and unknown locations).
• Matching: police town names are joined to census boundaries by name. In every year at least 99.2% of rows and 99.4% of incidents were matched (99.6% if counts published only for whole villages or large districts are set aside). Rows that could not be matched are not counted in any neighborhood.
• The police files add the county to some town names (for example 西多摩郡瑞穂町); the county is removed before matching with the census.

Offense categories

Japan's National Police Agency groups Penal Code offenses into six categories. “Violent crimes” (凶悪犯) are murder, robbery, arson and non-consensual sexual intercourse; “assaults and similar” (粗暴犯) are unlawful assembly with weapons, assault, bodily injury, intimidation and extortion (National Police Agency, statistics on Penal Code offenses, 2023, glossary, in Japanese). The Tokyo police town-level files split violent crimes into “robbery” and “other”; murder, arson and non-consensual sexual intercourse all sit inside “other”. Sexual offenses outside the violent and assault categories, such as non-consensual indecent assault, are not counted in either layer.

Street incidents (per km²)

Meant for travelers and anyone walking around: what gets reported on and around the streets of an area.

“Weighted incidents” means counts multiplied by the offense weights and added up. Because this is per area, small, crowded nightlife districts come out high. The card always shows the area and the three-year total next to the rate.

Residential incidents (per 1,000 residents)

Meant for people choosing somewhere to live: incidents that tend to affect homes and residents.

Neighborhoods with fewer than 100 residents get no residential figure, because dividing by a tiny population produces meaningless swings (229 areas).

Smoothing small numbers

Many neighborhoods see zero to a few reported incidents over three years. Divided raw, a single case would flip them between levels. So before dividing we add the average rate of the surrounding city or ward, with the weight of two incidents (the “+ 2” above). Busy areas barely move; areas with very few cases are pulled slightly toward their local average.

Level thresholds (actual values)

Both incident layers are split into five levels by their ratio to the median smoothed rate of neighborhoods in Tokyo's 23 wards. The Tama area to the west uses the same baseline. The medians are 33.33 weighted incidents per km² per year (street) and 4.32 per 1,000 residents per year (residential).

Street incidents

Residential incidents

Much of the Tama area falls into “Very low” on the street layer, largely because people are spread over far more land than in the 23 wards. The baseline is printed in the legend and on every card so this stays visible.

Estimated quietness

There is no public noise measurement that can be compared block by block, so SHIZUAN estimates a score from 0 to 100 using conditions that usually go with noise. It is not measured noise.

Late-night venues are bars, pubs, nightclubs, karaoke boxes and similar places in OpenStreetMap, with izakaya counted as half. The download covered mainland Tokyo and its edges (10,641 entries); only those inside each neighborhood are counted. The reference values (600 venues per km² and 1,400,000 riders per day) are fixed, so the scale will not shift as more regions are added. Levels use fixed score bands:

The reasoning and the blind spots are explained in How the quietness estimate works, and its limits.

“Where to stay” default filters

A neighborhood matches when all four hold: street incidents at “Low” or lower, estimated quietness at “Quiet” or higher, at least three places to stay within 400 m of its center (counted from OpenStreetMap; 1,716 entries in the download area), and a station within 800 m. With the defaults, 17 neighborhoods match. The sliders on the map let you loosen each condition.

Known limits

SHIZUAN's figures and levels compare statistics. They do not guarantee anyone's safety. Reported incidents are cases known to the police. They are not a measure of how often things actually happen or of how safe a place feels. Quietness is estimated from public data, not measured noise.

• Only what is reported. Cases never reported to the police are missing, and the share that gets reported differs by offense.
• No-data areas. Some neighborhoods in Hinode-machi (28 areas), Hinohara-mura (13 areas), Shinjuku-ku (2 areas), Machida-shi (2 areas), Akiruno-shi (2 areas), Hino-shi (1 areas), Akishima-shi (1 areas), Ota-ku (1 areas), Kunitachi-shi (1 areas) show no figure (gray on the map) because police town names could not be matched to them. Where town names changed, some matched neighborhoods may still be undercounted.
• Crowds raise counts. Station fronts, nightlife districts and big shopping areas receive far more visitors than residents, so their counts are higher. Per-resident figures in those places are especially high.
• Small areas move easily. Even with smoothing, a few cases can shift a tiny neighborhood.
• Boundaries and addresses differ. Census boundaries do not always match street addresses.
• Sexual offenses. Non-consensual sexual intercourse is grouped with murder and arson and cannot be separated; other sexual offenses are not in either layer.
• Places to stay are an OpenStreetMap count. Anything not listed there is not counted.

English names

English neighborhood names are romanized from the readings in Japan Post's postal code data, using Hepburn without long-vowel marks (so 丸の内 becomes Marunouchi and 大手町 becomes Otemachi). 5,272 of 5,308 named neighborhoods (99.3%) have one; the rest are shown in Japanese rather than guessed. The syllabic “n” is always written n with no separator (新大橋 becomes Shinohashi). Station names in English come from OpenStreetMap and exist for 698 of 709 stations.

Download

The components of the quietness estimate (Z, P, T, S and the score) for every neighborhood are available as quiet-derived.csv under the Open Database License (ODbL) 1.0, because they include OpenStreetMap data. Please credit “© OpenStreetMap contributors” along with the other sources.

Update log

• September 2026: built with 2023–2025 incident data; English version added; fixed matching of Nishitama county towns and marked areas published only as larger totals as “No data”.

Check it on the map