⚙️ How-To Restaurant inspections Maps & visualisation Notebook By · Updated

How to Map Restaurant Inspection Scores by Neighbourhood

On this page

Problem statement

Which New York neighbourhoods have the cleanest restaurant kitchens, and which the worst? Mapping it means choosing a measure (score, grade or share of bad inspections), choosing which inspections count, and placing restaurants in neighbourhoods. The inspection file carries 2010 neighbourhood codes that match no current boundary, so the reliable route is a spatial join. This page maps the median score of initial cycle inspections across New York's 2020 Neighborhood Tabulation Areas and ranks the neighbourhoods with enough inspections to compare.

Quick answer

Keep initial cycle inspections with a score and a non-zero position, join their points to 2020 NTA polygons in a projected CRS, and take the median score per neighbourhood. Only compare neighbourhoods with enough inspections; 30 is used here.

ins = insp[insp.date.dt.year > 1900].drop_duplicates(["CAMIS", "INSPECTION DATE", "INSPECTION TYPE"])
r = ins[ins["INSPECTION TYPE"].str.contains("Cycle Inspection / Initial") & ins.SCORE.notna() & (ins.lat > 0)]
pts = gpd.GeoDataFrame(r, geometry=gpd.points_from_xy(r.lon, r.lat), crs=4326).to_crs(2263)
j = gpd.sjoin(pts, nta[["NTA2020", "NTAName", "BoroName", "geometry"]], predicate="within")
per = j.groupby(["NTA2020", "NTAName", "BoroName"]).agg(inspections=("SCORE", "size"), median_score=("SCORE", "median")).reset_index()
print(f"{len(r):,} initial cycle inspections; placed in an NTA: {len(j) / len(r):.1%}; NTAs with 30 or more: {(per.inspections >= 30).sum()}")
45,816 initial cycle inspections; placed in an NTA: 100.0%; NTAs with 30 or more: 198

45,816 initial cycle inspections have a score and a position, and all of them fall inside a 2020 neighbourhood. 198 neighbourhoods have 30 or more inspections to compare.

Map of New York neighbourhoods shaded by the median score of initial restaurant inspections, darkest red in parts of Queens such as Flushing and on Staten Island's north shore, palest around JFK airport and Yankee Stadium. Map of New York neighbourhoods shaded by the median score of initial restaurant inspections, darkest red in parts of Queens such as Flushing and on Staten Island's north shore, palest around JFK airport and Yankee Stadium.
Median initial inspection score by 2020 neighbourhood, for neighbourhoods with 30 or more inspections.

Choosing inspections and joining to neighbourhoods

1. Use initial inspections only

print(ins.groupby("INSPECTION TYPE").SCORE.agg(["median", "mean"]).loc[["Cycle Inspection / Initial Inspection", "Cycle Inspection / Re-inspection"]].round(1).to_string())
                                       median  mean
INSPECTION TYPE                                    
Cycle Inspection / Initial Inspection    13.0  17.6
Cycle Inspection / Re-inspection         13.0  17.0

Initial inspections and re-inspections have the same median (13) and similar means (17.6 and 17.0). Mixing them still double-counts restaurants that failed a first visit, so keep to the initial ones.

2. Join points to 2020 neighbourhoods

print(f"restaurants carrying a 2010 NTA code: {insp.NTA.notna().mean():.1%}; codes that are 2020 codes: {insp.NTA.isin(nta.NTA2020).mean():.1%}")
restaurants carrying a 2010 NTA code: 98.4%; codes that are 2020 codes: 0.0%

98.4% of rows have an NTA code, but none of them is a 2020 code: the file still uses the 2010 neighbourhood scheme. Joining on that column to 2020 boundaries matches nothing, so the points are joined spatially instead.

Horizontal bars of the eight lowest and eight highest neighbourhood median scores, from Yankee Stadium at 4 and JFK Airport at 7 up to Flushing-Willets Point at 20. Horizontal bars of the eight lowest and eight highest neighbourhood median scores, from Yankee Stadium at 4 and JFK Airport at 7 up to Flushing-Willets Point at 20.
Lowest and highest median scores among neighbourhoods with 30 or more inspections.

3. Rank neighbourhoods with enough inspections

ok = per[per.inspections >= 30].sort_values("median_score")
print("lowest median score:", ok.head(3)[["NTAName", "BoroName", "inspections", "median_score"]].to_dict("records"))
print("highest median score:", ok.tail(3)[["NTAName", "BoroName", "inspections", "median_score"]].to_dict("records"))
lowest median score: [{'NTAName': 'Yankee Stadium-Macombs Dam Park', 'BoroName': 'Bronx', 'inspections': 45, 'median_score': 4.0}, {'NTAName': 'John F. Kennedy International Airport', 'BoroName': 'Queens', 'inspections': 82, 'median_score': 7.0}, {'NTAName': 'Rockaway Beach-Arverne-Edgemere', 'BoroName': 'Queens', 'inspections': 85, 'median_score': 11.0}]
highest median score: [{'NTAName': 'East Flushing', 'BoroName': 'Queens', 'inspections': 79, 'median_score': 19.0}, {'NTAName': 'Tompkinsville-Stapleton-Clifton-Fox Hills', 'BoroName': 'Staten Island', 'inspections': 74, 'median_score': 19.0}, {'NTAName': 'Flushing-Willets Point', 'BoroName': 'Queens', 'inspections': 742, 'median_score': 20.0}]

The lowest medians are places dominated by stadium and airport concessions: Yankee Stadium–Macombs Dam Park (4) and JFK Airport (7), then Rockaway Beach (11). The highest are East Flushing and Flushing–Willets Point in Queens (19 and 20) and Tompkinsville–Stapleton on Staten Island (19). Flushing's 742 inspections make it one of the busiest restaurant districts in the city as well as one of the worst-scoring.

Bad inspections and boroughs

The data used on this page

New York City Department of Health and Mental Hygiene restaurant inspection results from NYC Open Data, saved on this site on 16 October 2026, and the city's 2020 Neighborhood Tabulation Areas.

from pathlib import Path

import geopandas as gpd
import numpy as np
import pandas as pd
import requests
from shapely import wkt

Path("data").mkdir(exist_ok=True)
# NYC Department of Health and Mental Hygiene (DOHMH) restaurant inspection results (NYC Open Data, dataset 43nn-pn8j),
# downloaded on 16 October 2026 and saved on this site (xz-compressed, unchanged), because the city replaces it daily.
# Live: https://data.cityofnewyork.us/api/views/43nn-pn8j/rows.csv?accessType=DOWNLOAD
# Also NYC 2020 Neighborhood Tabulation Areas (dataset 9nt8-h7nd), saved on 15 October 2026.
FILES = ["nyc-restaurant-inspections-2026-10-16-v2.csv.xz", "nyc-nta-2020-2026-10-15.zip"]
for name in FILES:
    if not Path(f"data/{name}").exists():
        Path(f"data/{name}").write_bytes(requests.get(f"https://data.spatialworkflow.io/{name}", timeout=600).content)
insp = pd.read_csv("data/nyc-restaurant-inspections-2026-10-16-v2.csv.xz", dtype=str)
insp["date"] = pd.to_datetime(insp["INSPECTION DATE"], format="%m/%d/%Y")
insp["SCORE"] = pd.to_numeric(insp.SCORE)
insp["lat"], insp["lon"] = pd.to_numeric(insp.Latitude), pd.to_numeric(insp.Longitude)
n = pd.read_csv("data/nyc-nta-2020-2026-10-15.zip")
nta = gpd.GeoDataFrame(n.drop(columns="the_geom"), geometry=n.the_geom.map(wkt.loads), crs=4326).to_crs(2263)
print(f"DOHMH inspections file: {len(insp):,} rows for {insp.CAMIS.nunique():,} restaurants")
DOHMH inspections file: 296,514 rows for 31,531 restaurants

Example 1: share scoring 28 or more

j["c_range"] = j.SCORE >= 28
c = j.groupby("NTAName").agg(n=("SCORE", "size"), share=("c_range", "mean"))
print(c[c.n >= 30].sort_values("share", ascending=False).head(5).round(3).to_dict("index"))
{'Gravesend (South)': {'n': 33, 'share': 0.364}, 'Queensboro Hill': {'n': 46, 'share': 0.326}, 'Flushing-Willets Point': {'n': 742, 'share': 0.321}, 'South Richmond Hill': {'n': 57, 'share': 0.316}, 'Cambria Heights': {'n': 40, 'share': 0.3}}

By the share of inspections scoring 28 or more (the C range), Gravesend (South) leads at 36.4%, followed by Queensboro Hill and Flushing–Willets Point (32.1%). Shares from 30–50 inspections are noisy; give them less weight than Flushing's.

Map of New York neighbourhoods shaded by the share of initial inspections scoring 28 or more, darkest in parts of southern Brooklyn and Queens. Map of New York neighbourhoods shaded by the share of initial inspections scoring 28 or more, darkest in parts of southern Brooklyn and Queens.
Share of initial inspections in the C range by neighbourhood: Gravesend (South) leads at 36.4%.

Example 2: by borough

print(j.groupby("BoroName").SCORE.agg(["size", "median", "mean"]).round(1).to_string())
                size  median  mean
BoroName                          
Bronx           4389    13.0  17.5
Brooklyn       11992    12.0  17.5
Manhattan      17330    12.0  17.0
Queens         10396    13.0  19.2
Staten Island   1688    13.0  16.9

Borough medians are 12 or 13 everywhere, and Queens has the highest mean (19.2). The variation is between neighbourhoods, not boroughs, which is why a borough map hides it.

What a neighbourhood score measures

A neighbourhood's median score summarises the restaurants that happen to be there, so it mixes place with type of restaurant. Areas dominated by chain coffee shops and airport concessions score low, and areas with many small independent kitchens score higher, partly because older buildings make pests and maintenance harder. The map is a description of inspection outcomes, not a ranking of how safe it is to eat in each place.

Map choices

  • Use the median, not the mean: a few inspections above 100 points pull means up.
  • Set a minimum count. Parks and airports have few restaurants and extreme medians.
  • Do not use the file's NTA column for 2020 boundaries; it holds 2010 codes.
  • Points with zero coordinates must be dropped before the join or they vanish silently.

FAQ

Which NYC neighbourhood has the worst restaurant inspection scores?

Among neighbourhoods with at least 30 initial inspections, Flushing–Willets Point in Queens has the highest median score, 20, from 742 inspections.

Why not map restaurant letter grades?

Grades are biased toward A by the re-inspection rule. The score of the initial inspection is the comparable measure.