TLDR:

• [Motivation] Despite research that living near gas stations exposes people to harmful pollutants, hundreds of thousands of New Yorkers live near — or adjacent to — gas stations. This post describes an analysis of residential proximity to gas stations in NYC using MapPluto and Census/ACS data (2002-2023).

• [Topline] In 2002, about 275,000 residential units in NYC (8.5% of total) were within 300 feet of a gas station. By 2023, this number fell to around 210,000 (5.7% of total). The Bronx, with the highest share, was also the only borough where the share of apartments near a gas station increased (9.2% in 2002 to 9.3% in 2023), while Manhattan saw the largest decline (6.7% to 3.5%).

• [Spatial Inequity] In both 2002 and 2023, housing units within 300 feet of a gas station were located in neighborhoods that, on average, had higher percent Black and Hispanic, lower percent White, and lower median household income.

• [Redevelopment economics] The number of gas stations declined 37% from 2023 to 2002 (from ~1100 to 700), driven mostly by redevelopment of gas station sites on valuable land in Lower Manhattan and nearby parts of Brooklyn/Queens.

How many New Yorkers live near gas stations?

In 2002, about 275,000 residential units in NYC (8.5% of all units) were within 300 feet of a gas station (these units were spread across 43,000 parcels). By 2023, the number of residential units near gas stations fell to 210,000 — about 5.7% of all units — and spread across about 29,000 parcels. In both years, this share was highest in the Bronx and lowest in Staten Island (see Figure below).

Assuming 2.5 people per household would imply that about 700k residents in 2002 and about 500k residents in 2023 lived within 300 feet of a gas station. This is a a super rough approximation. I spent some time exploring using the average household size in NYC, by borough, over time — combined with estimated vacancy rates — to improve this approximation, but it didn’t change the overall numbers much (actual household sizes are a bit smaller than 2.5 per household and vacancy rates are nonzero, so they basically offset). Perhaps one could use average dwelling size from the parcel data to further refine this.

Below is a panel of maps illustrating residential exposure to gas stations in 2002 (left) and 2023 (right). The top plots shade Census Block Groups by whether there is any residential unit within 300 feet of a gas station in that Block Group, whereas the two bottom plots show the share of residential units within a Census Block Group that are within 300 feet of a gas station. Perhaps the biggest takeaway is that the share of units within 300 feet of a gas station declined, particularly in lower Manhattan, during this period.

The Environmental Justice angle

Across both years, the share of residential units within 300 feet of at least one gas station decreases in neighborhood income. Below are some Figures illustrating this negative relationship between neighborhood income and the share of units near gas stations both citywide and by borough in 2002 and 2023 (the units of analysis here are Census Block Groups, and what is shown is a GAM, a smoothed line of fit.)

There are also racial and ethnic divides. I took the average (weighted by the number of residential units) of each residential parcel’s Census Block Group median income, percent Black, percent non-Hispanic White, percent Hispanic, and percent Asian for residential parcels within 300 feet of a gas station and all other residential parcels.

• In 2002, parcels within 300 feet of a gas station had (according to the 2000 Census) a higher CBG percent Black (29% vs 23%), lower percent White (73% vs 78%), higher percent Hispanic (27% vs 22%), and lower median household income (~34k vs 44k). Percent Asian wasn’t much different (7% vs 8%).

• In 2023, parcels within 300 feet of a gas station had a higher CBG percent Black (24% vs 18%), lower percent White (24% vs 37%), higher percent Hispanic (35% vs 26%), and lower median household income (~74k vs 97k). Percent Asian wasn’t much different (13% vs 14%).

There are pretty big gaps — by race/ethnicity and income, with lower income and Black/Latino people more likely to live unhealthily close to gas stations. This isn’t exactly surprising, given what we know about environmental racism and the politics of locally unwanted land uses.

Decline in the number of gas stations over time

The number of gas stations in New York City declined ~37% between 2002 and 2023.

• In 2002, there were 1,118 gas stations (land uses “G3”, “G4”, “G5” in MapPLUTO). In 2023, there were 707 gas stations, indicating a decline of 411 gas stations.

• 477 gas stations that existed in 2002 but not 2023 — these parcels redeveloped into other uses. Meanwhile, 66 gas stations existed in 2023 that did not exist in 2002. This varied by borough (see Figure below).

• As the Figures show, all the boroughs saw declines, though the degree varied across boroughs. In percentage terms, the decline in gas stations was steepest in Manhattan (~66%), followed by Brooklyn (~48%), Queens (~29%), the Bronx (~24%), and Staten Island (20%).

What is driving change over time?

Probably simple economics — there’s a handful of press articles about the strong economic pressures to redevelop gas stations in Lower Manhattan, where landowners of gas stations could maximize profit by selling to a developer. One clear pattern in the map below is that gas stations were basically wiped out in Lower Manhattan during this period (cluster of green points) — otherwise, it mostly looks to be a core-periphery dynamic, though not perfectly, with new gas stations (purple points) mostly far away from Manhattan, and redeveloped parcels (green) more often closer to the city (the purple dot in Lower Manhattan is a private diesel refueling station for UPS trucks on a UPS-owned parcel). It’d be cool to explore zoning and the extent to which 2002 zoning is associated with land use change.

Concluding Thoughts

Gas stations are just an example — there are many land uses with noxious or salutary effects on nearby residents, and the analyses described can be extended to those, too. Ideally one would replicate this analysis on the universe of noxious land uses and then generate an index of exposure to pollutants. One could also (and perhaps I will!) replicate this analysis on some enriching land uses (e.g., parks) and explore inequities there — I am sure a version of this has been done many times over but, as with gas stations, not with the granularity that the spatial data allows.

More on Methodology

Conceptualizing proximity to noxious land use

• I use a 300 foot threshold to categorize residential units as proximate to a gas stations.

  • Distance-based thresholds are a common approach in academic research exploring effects of living near noxious land uses (Brender et al 2011).

  • The 300 feet threshold stems from a standard adopted in 2005 by the California Environmental Protection Agency’s California Air Resources Board; specifically, the board recommended that local authorities “avoid siting new sensitive land uses within 300 feet of a large gas station (defined as a facility with a throughput of 3.6 million gallons per year or greater). A 50 foot separation is recommended for typical gas dispensing facilities.”

  • Hilpert et al 2019 measure and describe exposure to benzene and other pollutants around gas stations and find substantially higher emissions (they say 10x) than previous research. Hence they conclude that the California 300 feet setback rule may not be strict enough.

Data & Workflow

• The underpinning of this project is a free, public, comprehensive spatial database of NYC parcels — with property assessment information — called MapPLUTO. I used the 2023 Release 23v3.1 MapPLUTO – Shoreline Clipped (FGDB) file. For 2002 data, I used the “02b” MapPLUTO Shapefile from the “Bytes of the Big Apple” Archive.

• I used the R Package tidycensus to retrieve 2002 5-year ACS (for 2023 gas stations) or the 2000 Decennial Census (for 2002 gas station) Census Block Group level demographic data.

• The basic data wrangling workflow is as follows:

  • Create a 300 ft buffer around all gas stations. Perform a spatial join of residential parcels with the buffer. Sf package in R.

  • Spatially join each residential parcel with its Census Block Group. To do this, I’m using a tigris shapefile with 2022 Census Block Groups for 2023 gas station and parcel data and a tigris shapefile with 2000 Census Block Groups for 2002 gas station and parcel data.

  • Retrieve Census Data (neighborhood demographics); join Census data with parcels by Block Group Census identifier.

  • From there, make plots, analyses, and maps!