In June 2026, NorthPoint Development offered 4,500 households in Hazle Township, Pennsylvania $10,000 each – in cash – if they would support approval of a 1,300-acre data center in their community. Locals pushed back, calling it a bribe. The township had already rejected the project on zoning grounds and enacted a moratorium on data center construction. The median household income in the area is around $60,000.
That offer crystallizes a question that is now live in communities across the United States: when a large industrial facility is proposed near a residential area, how much should affected residents actually be compensated – and how would you even calculate the right number?
The AI data center buildout has made this question urgent. In Q2 2026 alone, community resistance blocked or delayed 45 data center projects worth $68 billion across the U.S., according to Data Center Watch via parameter.io. The Brookings Institution framed the backlash as a proxy for a broader question: who decides where AI infrastructure goes, and who bears its costs?
Residents object for concrete reasons – noise, water use, power consumption, and the suspicion that nearby homes will lose value. That last concern is empirically grounded. A 2011 hedonic pricing study from the Netherlands (de Vor & de Groot, using data on 70,684 home sales) found that homes within 250 meters of industrial sites sold for roughly 15% less than comparable homes beyond 2,250 meters – a discount that fades to near zero by about 1.4 miles out. Research specifically on data centers is more mixed. A November 2025 George Mason University study on Northern Virginia home sales found prices higher near data centers – but Northern Virginia is an atypical case: a high-income tech corridor where the data center industry is already integrated into the local economy and a major employer. Resident complaints near operating facilities (noise from cooling equipment, power infrastructure, and truck deliveries) are documented in local press from Georgia, Ohio, and Virginia, but rigorous before-and-after property value studies at specific data center sites are not yet publicly available.
None of this research answers the compensation question directly, because it describes outcomes – what actually happened to property values in places that already have industrial land – not what should happen in places considering it. To get at that, you need to know what living near industrial land correlates with in terms of income and revealed preference, across many communities and many distance bands.
That is what this analysis provides, for New York City.
This report began as a logistics gap analysis. The original question was narrow: which NYC census tracts fall outside a 1-mile radius of any industrial lot, and how many residents live in those gaps? The spatial analysis answered that – 31 tracts, 61,426 residents, concentrated in outer Queens and Brooklyn.
When I added income data, the logistics framing became secondary. Median household income rises monotonically with distance from industrial land across every distance band. The income gap between tracts within 0.5 miles of industrial land and those beyond 0.5 miles is roughly $18,000 at the median – a pattern that is statistically robust but reflects sorting, not a directly measurable cost.
That finding connects to the NorthPoint / Hazle Township situation via the Netherlands property value research. If a home near the proposed facility is worth $200,000 and industrial proximity discounts it by 15%, that is a $30,000 loss in asset value – three times the NorthPoint offer. The NYC income gradient points in the same direction but is a sorting signal, not a cost figure. The Dutch property value evidence is the more direct benchmark, and against it $10,000 is hard to justify for the closest neighbors.
This report presents both the original spatial analysis and the income gradient finding, and frames both in the context of the current debate over industrial facility siting and community compensation.
The spatial analysis uses NYC’s PLUTO property database. Land use code 06 maps to building classes E* (warehouses), F* (factories), and L* (loft buildings) per the NYC DCP PLUTO Data Dictionary. This is a broader category than distribution-oriented warehouses alone. Data centers are not typically classified under code 06 in PLUTO; depending on their building classification they may appear in commercial or utility categories, though the exact classification varies by facility. The income gradient finding applies to existing NYC industrial land as classified by PLUTO, not to data centers specifically.
The analogy to data centers is directional, not direct. Industrial lots (warehouses, factories) and data centers share characteristics relevant to neighboring residents – noise, truck traffic, land area, visual impact – but differ in others (data centers have less truck traffic, more power infrastructure, and different noise profiles). The Netherlands property value research covers industrial sites broadly; whether its 15% discount applies to data centers specifically is unknown.
What this analysis does establish: across NYC’s 2,327 census tracts, higher income is consistently associated with greater distance from industrial land. That pattern is robust, city-wide, and cannot be explained away as a quirk of the 31 gap tracts alone.
The 1-mile proximity threshold used for the binary gap classification is intentionally tight. A 5-mile radius eliminates virtually all gaps in NYC. The tighter radius surfaces meaningful geographic variation across the boroughs and aligns with walkable proximity (roughly 20 minutes on foot).
Data note (updated October 7, 2026): An earlier version of this report (published October 4) showed 38 gap tracts and 169,223 residents, with 14 Manhattan gap tracts. Those numbers were wrong. The Socrata API returns results in pages, and the initial pull hit the default 1,000-row limit, returning an incomplete PLUTO industrial lot dataset. A corrected pull (no row limit) retrieved the full set. With complete data, Manhattan has near-complete industrial land coverage (only 1 gap tract), and the city-wide gap falls to 31 tracts and 61,426 residents. All numbers in this version reflect the corrected pull.
All data is pulled programmatically via public APIs. No manual file downloads are required to reproduce this analysis.
| Dataset | Source | Year | Access Method |
|---|---|---|---|
| Census tract boundaries + population | U.S. Census Bureau (Decennial Census) | 2020 | tidycensus R package |
| Borough boundaries | U.S. Census Bureau (TIGER/Line) | 2024 | tigris R package |
| Industrial/warehouse tax lots | NYC Dept. of City Planning — PLUTO | Current | NYC Open Data Socrata API |
| Median household income by tract | U.S. Census Bureau (ACS 5-year) | 2020 | tidycensus R package |
A note on the industrial lot data: PLUTO classifies every NYC tax lot by land use code. This analysis filters for code 06 (Industrial & Manufacturing), which maps to building classes E* (warehouses), F (factories), and L (loft buildings). The category is broader than distribution-oriented warehouses alone: it includes factories and loft buildings, and gives no indication of whether a lot hosts an active logistics operation. These lots are used here as a publicly available proxy for industrial land; they are not a direct measure of delivery or logistics infrastructure.
The analysis follows four steps:
1. Data acquisition. Census tract population and geometry are pulled from the 2020 Decennial Census via the tidycensus package. Borough boundaries come from the Census TIGER/Line files via tigris. Industrial lot locations are queried from the NYC PLUTO dataset via the Socrata JSON API with no row limit, then converted to point geometry using the lot centroid coordinates. Median household income is pulled from the 2020 ACS 5-year estimates.
2. Coordinate reference system. All layers are projected to EPSG:2263 (New York State Plane, Long Island), which uses feet as its unit of measurement. This CRS is optimized for accuracy in the New York metro area and is the standard used by NYC government GIS data.
3. Buffer analysis. Each industrial lot centroid is buffered by 5,280 feet (1 mile). The resulting circles are dissolved into a single merged coverage polygon using st_union(). This merged polygon represents the total area of NYC within 1 mile of at least one industrial lot.
4. Gap identification. Each census tract is tested for intersection with the coverage polygon using st_intersects(). Tracts with no intersection are classified as “far from industrial land.” In addition, the straight-line distance from each tract centroid to its nearest industrial lot is computed using st_distance(), enabling continuous distance-band analysis beyond the binary classification.
library(sf)
library(dplyr)
library(ggplot2)
library(scales)
library(here)
library(knitr)
library(tidyr)
nyc_tracts <- readRDS(here("data/processed/nyc_tracts_analyzed.rds"))
borough_summary <- readRDS(here("data/processed/borough_summary.rds"))
warehouses <- readRDS(here("data/processed/warehouses.rds"))
warehouse_buffer <- readRDS(here("data/processed/warehouse_buffer.rds"))total_tracts <- nrow(nyc_tracts)
underserved_tracts <- sum(!nyc_tracts$served)
underserved_pop <- sum(nyc_tracts$population[!nyc_tracts$served], na.rm = TRUE)
total_pop <- sum(nyc_tracts$population, na.rm = TRUE)
pct_underserved <- underserved_pop / total_pop
cat(
"Total census tracts: ", comma(total_tracts), "\n",
"Tracts far from industrial land: ", comma(underserved_tracts), "\n",
"Population in those tracts: ", comma(underserved_pop), "\n",
"Share of total population: ", percent(pct_underserved, accuracy = 0.1), "\n"
)Total census tracts: 2,327
Tracts far from industrial land: 31
Population in those tracts: 61,426
Share of total population: 0.7% Across NYC’s 2,327 census tracts, 31 tracts containing 61,426 residents fall outside a 1-mile radius of any industrial lot. That represents 0.7% of the city’s total population by the buffer method (a tract counts as covered if any part of it intersects the 1-mile zone). The stricter centroid method – a tract is covered only if its center falls within a mile – puts the city-wide gap at roughly 3.4% of residents. The distance-band table below shows the full breakdown by borough.
The binary gap classification (covered vs. not) masks variation in how far residents actually are from industrial land. This section reports the share of population within each distance band, by borough, and the median straight-line distance from tract centroid to the nearest industrial lot.
Two distance measures are in use here and they give different numbers intentionally. The borough table (below) classifies a tract as “covered” if any part of it intersects a 1-mile buffer – the most generous possible reading. The distance table here uses the tract centroid, which is stricter: a large tract whose center sits 1.1 miles away counts as a gap by centroid even if its edge is technically within range. The 0.7% city-wide figure comes from the buffer method; the centroid method shows about 6% of Queens residents and 12% of Staten Island residents more than a mile from industrial land.
# Requires dist_to_nearest_miles column added by 02_spatial_analysis.R
bands <- c(0.5, 1, 2, 3)
tracts_with_borough <- nyc_tracts |>
st_drop_geometry() |>
mutate(
borough = case_when(
substr(GEOID, 3, 5) == "061" ~ "Manhattan",
substr(GEOID, 3, 5) == "047" ~ "Brooklyn",
substr(GEOID, 3, 5) == "081" ~ "Queens",
substr(GEOID, 3, 5) == "005" ~ "Bronx",
substr(GEOID, 3, 5) == "085" ~ "Staten Island",
TRUE ~ NA_character_
)
) |>
filter(!is.na(population), !is.na(dist_to_nearest_miles), !is.na(borough))
borough_bands <- tracts_with_borough |>
group_by(borough) |>
summarise(
within_0.5mi = percent(sum(population[dist_to_nearest_miles <= 0.5]) / sum(population), accuracy = 0.1),
within_1mi = percent(sum(population[dist_to_nearest_miles <= 1.0]) / sum(population), accuracy = 0.1),
within_2mi = percent(sum(population[dist_to_nearest_miles <= 2.0]) / sum(population), accuracy = 0.1),
within_3mi = percent(sum(population[dist_to_nearest_miles <= 3.0]) / sum(population), accuracy = 0.1),
median_dist_mi = round(median(dist_to_nearest_miles, na.rm = TRUE), 2),
.groups = "drop"
) |>
arrange(desc(median_dist_mi))
# City-wide row
citywide_band <- tracts_with_borough |>
summarise(
borough = "NYC (all)",
within_0.5mi = percent(sum(population[dist_to_nearest_miles <= 0.5]) / sum(population), accuracy = 0.1),
within_1mi = percent(sum(population[dist_to_nearest_miles <= 1.0]) / sum(population), accuracy = 0.1),
within_2mi = percent(sum(population[dist_to_nearest_miles <= 2.0]) / sum(population), accuracy = 0.1),
within_3mi = percent(sum(population[dist_to_nearest_miles <= 3.0]) / sum(population), accuracy = 0.1),
median_dist_mi = round(median(dist_to_nearest_miles, na.rm = TRUE), 2)
)
band_table <- bind_rows(borough_bands, citywide_band)
kable(
band_table |>
select(
Borough = borough,
`Within 0.5mi` = within_0.5mi,
`Within 1mi` = within_1mi,
`Within 2mi` = within_2mi,
`Within 3mi` = within_3mi,
`Median tract dist. (mi)` = median_dist_mi
),
caption = "Population share within each distance band, by borough (straight-line distance from tract centroid to nearest industrial lot)"
)| Borough | Within 0.5mi | Within 1mi | Within 2mi | Within 3mi | Median tract dist. (mi) |
|---|---|---|---|---|---|
| Staten Island | 47.4% | 87.7% | 100.0% | 100.0% | 0.50 |
| Manhattan | 80.3% | 99.8% | 100.0% | 100.0% | 0.31 |
| Bronx | 85.8% | 96.8% | 100.0% | 100.0% | 0.28 |
| Queens | 72.1% | 93.6% | 99.8% | 100.0% | 0.27 |
| Brooklyn | 91.0% | 98.7% | 100.0% | 100.0% | 0.21 |
| NYC (all) | 80.4% | 96.6% | 99.9% | 100.0% | 0.26 |
nyc_tracts |>
st_drop_geometry() |>
mutate(
borough = case_when(
substr(GEOID, 3, 5) == "061" ~ "Manhattan",
substr(GEOID, 3, 5) == "047" ~ "Brooklyn",
substr(GEOID, 3, 5) == "081" ~ "Queens",
substr(GEOID, 3, 5) == "005" ~ "Bronx",
substr(GEOID, 3, 5) == "085" ~ "Staten Island",
TRUE ~ NA_character_
)
) |>
filter(!is.na(dist_to_nearest_miles), !is.na(borough)) |>
ggplot(aes(x = dist_to_nearest_miles, fill = borough)) +
geom_histogram(bins = 40, alpha = 0.8, color = "white", linewidth = 0.2) +
facet_wrap(~borough, nrow = 2) +
scale_x_continuous(limits = c(0, 5), labels = function(x) paste0(x, "mi")) +
scale_fill_brewer(palette = "Set2", guide = "none") +
labs(
title = "Distance to Nearest Industrial Lot, by Borough",
subtitle = "Straight-line distance from census tract centroid; tracts weighted equally",
x = "Distance to nearest industrial lot (miles)",
y = "Number of tracts",
caption = "Source: NYC PLUTO via Socrata API | U.S. Census Bureau (2020)"
) +
theme_minimal() +
theme(
plot.title = element_text(face = "bold"),
strip.text = element_text(face = "bold")
)
borough_table <- borough_summary |>
select(borough, coverage, tracts, population) |>
filter(!is.na(borough)) |>
tidyr::pivot_wider(
names_from = coverage,
values_from = c(tracts, population),
values_fill = 0
) |>
mutate(
`% in Gap` = scales::percent(
population_Underserved / (population_Served + population_Underserved),
accuracy = 0.1
)
) |>
arrange(desc(population_Underserved / (population_Served + population_Underserved))) |>
select(
Borough = borough,
`Gap Tracts` = tracts_Underserved,
`Covered Tracts` = tracts_Served,
`Pop. in Gap` = population_Underserved,
`% in Gap`
)
kable(
borough_table,
format.args = list(big.mark = ","),
caption = "Industrial land proximity by borough (tract covered if any part lies within 1 mile of an industrial lot)"
)| Borough | Gap Tracts | Covered Tracts | Pop. in Gap | % in Gap |
|---|---|---|---|---|
| Queens | 13 | 712 | 32,622 | 1.4% |
| Brooklyn | 11 | 794 | 19,809 | 0.7% |
| Staten Island | 2 | 124 | 2,390 | 0.5% |
| Bronx | 4 | 357 | 6,605 | 0.4% |
| Manhattan | 1 | 309 | 0 | 0.0% |
The population density map establishes the baseline: where are people concentrated across the five boroughs? Manhattan’s core shows the highest densities, exceeding 150,000 persons per square mile in some tracts. The outer boroughs show more moderate densities with significant variation between neighborhoods.
knitr::include_graphics(here("output/maps/map1_population_density.png"))
The gap map overlays three layers of information: census tract proximity status (green = within 1 mile of an industrial lot, red = outside 1 mile), the merged 1-mile coverage zone (light blue halo), and individual industrial lot locations (blue points).
knitr::include_graphics(here("output/maps/map2_logistics_gaps.png"))
The following map was produced directly in QGIS 3.44 LTR using the same analysis layers exported from R. It demonstrates the full geospatial workflow — data processed in R, layers loaded and styled in QGIS, and exported as a cartographic output.
knitr::include_graphics(here("output/maps/qgis_map_export.png"))
underserved_by_borough <- borough_summary |>
filter(coverage == "Underserved", !is.na(borough)) |>
arrange(desc(population))
ggplot(underserved_by_borough,
aes(x = reorder(borough, population), y = population)) +
geom_col(fill = "#d6604d", alpha = 0.85) +
geom_text(
aes(label = comma(population)),
hjust = -0.1,
size = 3.5,
color = "gray30"
) +
coord_flip() +
scale_y_continuous(
labels = comma,
expand = expansion(mult = c(0, 0.15))
) +
labs(
title = "Residents Outside 1-Mile Industrial Land Radius, by Borough",
subtitle = "Census tracts with no industrial lot (PLUTO code 06) within 1 mile",
x = NULL,
y = "Population",
caption = "Source: U.S. Census Bureau (2020) | NYC PLUTO via Socrata API"
) +
theme_minimal() +
theme(
plot.title = element_text(face = "bold"),
panel.grid.major.y = element_blank()
)
The gap tracts are concentrated in outer Queens and Brooklyn. That raises an obvious question: are these areas systematically wealthier, and if so, does income explain why industrial land is absent – through zoning exclusion, land costs, or political influence? To examine this, I pull 2020 ACS median household income by tract and compare distributions across distance bands.
# Load pre-pulled income data (pulled in 01_data_pull.R)
# Avoids calling the Census API at render time.
income_data <- readRDS(here("data/processed/nyc_income.rds"))
nyc_tracts_income <- nyc_tracts |>
st_drop_geometry() |>
left_join(income_data, by = "GEOID") |>
filter(!is.na(median_income))band_table <- nyc_tracts_income |>
mutate(band = cut(
dist_to_nearest_miles,
breaks = c(0, 0.25, 0.5, 1, Inf),
labels = c("0-0.25 mi", "0.25-0.5 mi", "0.5-1 mi", ">1 mi"),
include.lowest = TRUE
)) |>
group_by(band) |>
summarise(
tracts = n(),
residents = sum(population, na.rm = TRUE),
median_income = median(median_income, na.rm = TRUE),
.groups = "drop"
)
kable(
band_table,
col.names = c("Distance Band", "Tracts", "Residents", "Median Household Income"),
format.args = list(big.mark = ","),
digits = 0,
caption = "Median household income by distance to nearest industrial lot (2020 ACS 5-year)"
)| Distance Band | Tracts | Residents | Median Household Income |
|---|---|---|---|
| 0-0.25 mi | 1,078 | 4,180,197 | 63,708 |
| 0.25-0.5 mi | 693 | 2,863,455 | 66,704 |
| 0.5-1 mi | 358 | 1,417,850 | 84,558 |
| >1 mi | 77 | 295,880 | 90,379 |
wt <- wilcox.test(
median_income ~ I(dist_to_nearest_miles > 0.5),
data = nyc_tracts_income
)income_summary <- nyc_tracts_income |>
group_by(served) |>
summarise(
tracts = n(),
median_income = median(median_income, na.rm = TRUE),
.groups = "drop"
) |>
mutate(group = ifelse(served, "Within 1 mile of industrial land", "Outside 1 mile (gap)"))
kable(
income_summary |> select(group, tracts, median_income),
col.names = c("Group", "Tracts", "Median Household Income"),
format.args = list(big.mark = ","),
caption = "Median household income: gap tracts vs. covered tracts (2020 ACS 5-year)"
)| Group | Tracts | Median Household Income |
|---|---|---|
| Outside 1 mile (gap) | 17 | 101,461 |
| Within 1 mile of industrial land | 2,189 | 69,275 |
ggplot(nyc_tracts_income, aes(x = median_income, fill = served)) +
geom_density(alpha = 0.6) +
scale_fill_manual(
values = c("FALSE" = "#d6604d", "TRUE" = "#7fbf7b"),
labels = c("FALSE" = "Outside 1 mile (gap)", "TRUE" = "Within 1 mile")
) +
scale_x_continuous(labels = dollar_format(scale = 1/1000, suffix = "k")) +
labs(
title = "Income Distribution: Gap Tracts vs. Covered Tracts",
subtitle = "2020 ACS 5-year median household income",
x = "Median Household Income",
y = "Density",
fill = "Coverage Status",
caption = "Source: U.S. Census Bureau, ACS 2020 5-year estimates"
) +
theme_minimal() +
theme(plot.title = element_text(face = "bold"))
all_gap_median <- median(
nyc_tracts_income$median_income[!nyc_tracts_income$served], na.rm = TRUE
)
served_median <- median(
nyc_tracts_income$median_income[nyc_tracts_income$served], na.rm = TRUE
)
cat(
"Median income, gap tracts (all NYC): $", comma(round(all_gap_median)), "\n",
"Median income, covered tracts: $", comma(round(served_median)), "\n",
"Gap tracts with income data: ",
sum(!nyc_tracts_income$served), "of", sum(!nyc_tracts$served), "\n"
)Median income, gap tracts (all NYC): $ 101,461
Median income, covered tracts: $ 69,275
Gap tracts with income data: 17 of 31 Median household income rises monotonically with distance from industrial land across all four distance bands: $63,708 within 0.25 miles, $66,704 at 0.25–0.5 miles, $84,558 at 0.5–1 mile, and $90,379 beyond 1 mile. The largest step is between the 0.25–0.5 mile band ($66,704) and the 0.5–1 mile band ($84,558), a gap of roughly $18,000. A Wilcoxon rank-sum test comparing tracts within 0.5 miles to tracts beyond 0.5 miles confirms the difference is not noise (p < 0.001, across 2,206 tracts with income data).
The binary gap comparison (the 31 gap tracts vs. the rest) tells the same story: gap tracts median $101,461 vs. covered tracts $69,275. But the distance-band table is more informative because it shows the gradient holds continuously, not just at the 1-mile cutoff.
The data cannot distinguish between two explanations: affluent neighborhoods may have successfully excluded industrial uses through zoning, or those neighborhoods may simply sit at the outer edges of where the city’s industrial corridors historically developed. Separating geography from politics would require zoning history and land use change data over time. What the data does establish is that the income gradient is real, city-wide, and statistically robust.
Queens has the largest gap population (32,622 residents across 13 tracts), followed by Brooklyn (19,809 across 11 tracts). Both sets of gap tracts sit at the outer edges of their boroughs, distant from the industrial corridors concentrated along the waterfront and freight routes closer to the urban core.
Manhattan has only one gap tract (GEOID 36061000100, Census Tract 1), with zero counted residents. Its centroid falls at approximately 40.695°N, 74.043°W, in the harbor southwest of the Battery – consistent with harbor islands (Governors Island, Ellis Island, or Liberty Island) that carry no permanent residents. Zero is most likely a true count, not a differential privacy artifact. Manhattan’s near-complete industrial land coverage otherwise reflects the density of lots in specific corridors (West Side railyards, Meatpacking District, Hudson Yards area).
Median household income rises monotonically with distance from industrial land across all four distance bands, from $63,708 within a quarter mile to $90,379 beyond a mile. The largest step is between the 0.25–0.5 mile band ($66,704) and the 0.5–1 mile band ($84,558). A Wilcoxon rank-sum test comparing tracts within 0.5 miles to tracts beyond 0.5 miles confirms the gradient is statistically significant (p < 0.001, across 2,206 tracts). The data cannot say whether this reflects deliberate zoning exclusion or simply the geography of where industrial corridors developed, but the pattern itself is robust. The Manhattan gap tract has no income data, since it has no counted population.
The distance-band analysis adds resolution beyond the binary gap classification. About 16% of New Yorkers (96.6% within a mile minus 80.4% within half a mile) live in the 0.5–1 mile band. Staten Island stands out with the longest median tract distance (0.50 miles, vs. 0.21–0.31 for the other four boroughs). Queens has a shorter median (0.27 miles) but a longer right tail: about 6% of Queens residents are more than a mile from industrial land by the centroid measure, the second-highest share after Staten Island.
Broad land use category. According to the NYC DCP PLUTO Data Dictionary, land use code 06 maps to building classes E* (warehouses), F* (factories), and L* (loft buildings). Factories and lofts are included alongside distribution-oriented warehouses, and the category gives no indication of whether a lot hosts an active logistics operation. L* specifically includes converted loft buildings and self-storage facilities that generate little noise or truck traffic – meaningfully different from a large active warehouse. This analysis measures proximity to industrial land as a class – it does not distinguish a 500,000 sq ft fulfillment center from a converted loft building or an inactive factory floor. Including L* lots weakens the “industrial facility” proxy and likely understates the income gradient around genuinely heavy industrial uses. A more operationally precise analysis would require commercial real estate data (CoStar, CBRE) or cross-referencing with business license records to isolate active logistics facilities.
Edge effect. This analysis counts only NYC-classified lots. Major distribution hubs in New Jersey and just outside the city line in Nassau and Westchester are excluded. The effect is largest for Staten Island, which sits adjacent to major New Jersey logistics corridors in Elizabeth and Linden; its distance numbers would look materially different if those lots were included.
The 1-mile threshold. A tract 1.1 miles from a lot is counted as a gap; one at 0.9 miles is not. The distance-band table partially addresses this by reporting continuous bands, but the binary classification in the maps and borough table still uses the fixed threshold.
Straight-line distance is not travel time. Proximity to industrial land says nothing about road network connectivity, traffic conditions, loading access, or building entry – the actual constraints that determine delivery reach in an urban environment.
The core finding is that NYC census tracts closer to industrial land have lower median household incomes, and the gradient is continuous and statistically robust. This does not prove that industrial proximity causes lower incomes – the causal story runs in at least two directions. Industrial corridors may have developed where land was already cheap (lower-income areas). Or affluent neighborhoods may have used zoning to exclude industrial uses over time, making the correlation partly a product of political power. Distinguishing those explanations requires zoning history data this analysis does not include.
What the gradient does support is a directional reading: across 2,206 NYC tracts, proximity to industrial land is consistently associated with lower household income. People with more resources tend to live farther from industrial land. That pattern is consistent with the Netherlands research finding a 15% property value discount within 250 meters of industrial sites, though the two measures capture different things (income sorting vs. asset value).
Back to the $10,000 question. NorthPoint offered Hazle Township residents $10,000 – a one-time payment – to accept a 1,300-acre data center. The Netherlands study gives the clearest empirical benchmark: homes within 250 meters of industrial sites sold for roughly 15% less. If a home near the facility is worth $200,000, that is a $30,000 loss in asset value – three times the NorthPoint offer. The discount fades with distance; by roughly 1.4 miles it is near zero, so homes closest to the facility bear the largest loss. A flat $10,000 to everyone regardless of distance ignores that geography entirely.
The NYC income gradient points the same direction. Tracts within 0.25 miles of industrial land have a median income of $63,708; tracts beyond 1 mile reach $90,379. That $26,671 gap reflects sorting rather than a direct cost to any individual household, and should not be read as an annual figure. But it reinforces the Dutch finding: proximity to industrial land is something people with options choose to avoid.
None of this means developers should not build. It means the compensation question deserves an empirical basis rather than an arbitrary one. Relevant inputs would include: the distance from the facility to each household, local property values before and after comparable facilities opened nearby, the expected duration of impacts (noise, traffic, visual), and whether payments should be structured as one-time checks or ongoing annual compensation tied to the facility’s operating life.
The current conflict over data center siting is partly a zoning fight, partly an AI anxiety fight, and partly a legitimate dispute over who pays for infrastructure that benefits a much larger population. In Q2 2026 alone, community resistance stalled $68 billion in data center projects across the U.S. TechCrunch reported in September 2026 that “everyone can find a reason to dislike data center construction” – from noise and water use to AI ethics to simple neighborhood character concerns.
The Brookings Institution framed it as a democratic accountability problem: if AI infrastructure is built wherever companies choose, with local governments as the only check, communities end up negotiating individually against well-resourced developers without any common framework for what fair compensation looks like.
NYC’s industrial land income gradient does not resolve that, but it offers one empirical anchor. Across 2,206 tracts, income rises consistently with distance from industrial land. That sorting pattern, combined with the Dutch property value evidence, suggests the question of what residents are owed is empirical, not just political – and that a flat $10,000 is not derived from any such calculation.
Whether the comparison between NYC’s industrial lots and a Pennsylvania data center is close enough to matter is a judgment call. The land uses are different. The geography is different. But the core mechanism – a large industrial facility arrives near a residential area, generating noise, traffic, and visual impact – is similar enough that the income gradient finding is at minimum a reasonable order-of-magnitude reference, and significantly better than no empirical basis at all.
All data is pulled from public APIs at runtime. To reproduce this analysis from scratch:
# 1. Install packages and create folder structure
source("setup.R")
# 2. Pull all data from APIs
source("analysis/01_data_pull.R")
# 3. Run spatial analysis (computes covered/gap status and dist_to_nearest_miles)
source("analysis/02_spatial_analysis.R")
# 4. Export maps
source("analysis/03_maps.R")
# 5. Render this report
quarto::quarto_render("report/index.qmd")R version: R version 4.5.2 (2025-10-31 ucrt) Key packages: tidycensus 1.7.5, sf 1.1.0, ggplot2 4.0.2
U.S. Census Bureau. (2020). 2020 Decennial Census Redistricting Data (Public Law 94-171) Summary File. Retrieved via the tidycensus R package.
U.S. Census Bureau. (2020). American Community Survey 5-Year Estimates: Median Household Income (B19013). Retrieved via the tidycensus R package.
NYC Department of City Planning. (2024). PLUTO: Primary Land Use Tax Lot Output. NYC Open Data. https://data.cityofnewyork.us/resource/64uk-42ks.json
U.S. Census Bureau. (2024). TIGER/Line Shapefiles: Counties. Retrieved via the tigris R package.
de Vor, F., & de Groot, H. L. F. (2011). The Impact of Industrial Sites on Residential Property Values: A Hedonic Pricing Analysis from the Netherlands. Regional Studies, 45(5), 609–623. Key finding: 15% price discount within 250 meters of industrial sites, fading to near zero at 2,250 meters (roughly 1.4 miles). Based on 70,684 home sales. https://papers.tinbergen.nl/09035.pdf
Data Center Watch / Parameter.io. (2026). Community Resistance Stalls $68 Billion Worth of AI Data Center Developments. https://parameter.io/community-resistance-stalls-68-billion-worth-of-ai-data-center-developments/
Wheeler, T. (2026). Data Center Backlash Signals a Fight Over AI Power. Brookings Institution. https://www.brookings.edu/articles/data-center-backlash-signals-a-fight-over-ai-power/
Alcorn, C. (2026). Data center developer offers $10,000 checks to 4,500 households if 1,300-acre facility is approved. Tom’s Hardware. https://www.tomshardware.com/tech-industry/data-centers/data-center-developer-offers-usd10-000-checks-to-4-500-households-if-the-1-300-acre-facility-is-approved-locals-push-back-over-noise-and-bribe-concerns
George Mason University, Schar School of Policy and Government. (2025). Study: Home Prices Are Higher When the House Is Near a Data Center. https://schar.gmu.edu/news/2025-11/study-home-prices-are-higher-when-house-near-data-center
Coldewey, D. (2026, September 22). Everyone can find a reason to dislike data center construction. TechCrunch. https://techcrunch.com/2026/09/22/everyone-can-find-a-reason-to-dislike-data-center-construction/
Haber, M. (2026). How Data Centers Are Set to Impact the Value of Your Home. Newsweek. https://www.newsweek.com/how-data-centers-are-set-to-impact-the-value-of-your-home-11939599
This project was developed with assistance from Claude (Anthropic) as a collaborative coding and methodology tool. Claude helped structure the R scripts, debug API connectivity issues during data acquisition, and draft the prose in this report. All analytical decisions — including the choice of 1-mile service radius, the PLUTO land use classification, and the interpretation of results — were made and reviewed by the author. All code was tested and executed locally by the author prior to publication.