In July 1995, more than 700 people died in a Chicago heat wave. The sociologist Eric Klinenberg argued the deaths were social rather than meteorological: isolated people in neighborhoods that had lost the everyday businesses which gave them a reason to go outside and be seen. These maps test that idea using satellite thermal imagery and business records from 1995. Click any community area to see its numbers.
Darker = more vulnerable
Hover to preview, click to pin an area's full profile.
Between July 13 and July 20, 1995, Chicago recorded temperatures up to 106 degrees. The Cook County Medical Examiner certified 465 heat related deaths in that period, and researchers later estimated close to 700 excess deaths overall. The victims were overwhelmingly elderly, and they died indoors, alone, in apartments that had absorbed heat for days.
The obvious explanation is that some neighborhoods were simply hotter. Klinenberg's book Heat Wave argued something different. He compared two adjacent West Side neighborhoods with similar poverty rates and very different death rates, and concluded that what protected people was social contact. In places with busy shops, cafes and churches, older residents had ordinary reasons to leave the house, and neighbors noticed when they stopped appearing.
There is no public record of where people died in 1995 at the neighborhood level. Death records that detailed are not released, and the county's open data archive only starts in 2008. So this project does not try to model deaths. It measures the two things Klinenberg said mattered, and asks whether they line up with the geography the historical record already describes.
The first is heat. Satellites measure the temperature of the ground and rooftops, not the air, and they pass over at fixed times of day. That distinction turned out to be the whole story, and it is explained below.
The second is social infrastructure. For 1995, I used the Census Bureau's count of businesses by ZIP code, specifically restaurants and bars, which is as close as public data gets to "places where people gather." For today, I used OpenStreetMap's record of cafes, libraries, community centres and social facilities.
Daytime satellite images say the hottest parts of Chicago are the bungalow neighborhoods on the Southwest and Northwest sides. By that measure the neighborhoods that suffered most in 1995 look unremarkable, and some look cool. Washington Park came out as the ninth coolest of the city's 77 community areas.
That is because the daytime satellite passes over at about 10:30 in the morning, and Lake Michigan keeps the lakefront cool during the day. But heat does not kill people at 10:30 in the morning. It kills when buildings hold their heat through the night and the body never gets a chance to recover.
Switching to a satellite that passes over at 1:30 in the morning reverses the picture almost completely. North Lawndale moves from 41st hottest to 2nd. East Garfield Park moves from 44th to 3rd. Washington Park moves from 69th to 16th. The West Side corridor takes four of the five hottest positions at night. Daytime and nighttime temperature agree only about half the time, and it is the nighttime measurement that matches where people actually died.
Chicago's commercial geography has barely moved in thirty years. Where there were bars and restaurants in 1995, there are cafes now, and where there were not, there still are not. The statistical relationship is strong, around 0.7 on a scale where 1.0 would be a perfect match.
That matters because it means Klinenberg was not describing a moment that has since passed. The pattern he identified is still in place.
Heat and social infrastructure turn out to be almost unrelated to each other. Some hot neighborhoods are full of cafes, and some cool ones have almost none. Neither map alone predicts where the 1995 deaths clustered. The overlap does.
Five community areas fall in the worst quarter of the city on all three measures, hot at night, short of businesses in 1995, and short of gathering places today: South Lawndale, West Garfield Park, West Englewood, Chicago Lawn and Austin. Roughly 7,000 residents over 65 live alone in those five areas.
OpenStreetMap is built by volunteers, so a neighborhood can look empty simply because nobody has mapped it. If volunteer mapping were thinnest in poor neighborhoods, this entire analysis would be measuring mapping effort rather than real amenities.
So I checked, using the share of buildings in each area that carry a street address in OpenStreetMap as a measure of how thoroughly it has been surveyed. The neighborhoods this finding depends on are mapped as well as or better than the city average. West Garfield Park is twelve points above the median. The city's least mapped areas are the outlying bungalow belt, not the South and West sides. The pattern is real.
No death records appear anywhere in this analysis. This is a map of vulnerability, not of what happened. Its claim to be meaningful rests on independently reproducing a pattern that historians and sociologists documented by other means.
There is also one comparison it cannot make. The two neighborhoods at the center of Klinenberg's argument, North and South Lawndale, share a single ZIP code, so the 1995 business data cannot tell them apart. That specific test needs a different source.
| Community area | Vuln. | Night | Day | 1995 commerce | Third places | OSM % | 65+ alone |
|---|
Night heat is median land surface temperature from 75 MODIS Aqua 8-day composites (summers 2019–2025, ~1:30 AM overpass), 1 km native resolution. Temperatures display in °F by default; use the toggle above the map for °C. Day heat is 46 Landsat 8/9 scenes at 30 m, ~10:30 AM. They correlate only 0.50, and night is the one that matters: heat kills when buildings never cool overnight.
1995 commerce is eating and drinking establishments per 10,000 residents from the Census ZIP Code Business Patterns for 1995 (SIC 5812 and 5813), interpolated from ZIP codes onto community areas by population-weighted areal apportionment. The interpolation recovers 2,887,642 of Chicago's actual 2,896,016 residents, an error of 0.3%.
Third places today are OpenStreetMap cafes, libraries, community centres and
social facilities. OSM completeness (the share of building footprints
carrying addr:housenumber) is the honesty check: a neighborhood can look
amenity-poor simply because nobody mapped it. Chicago's least-mapped areas are the outlying
bungalow belt, not the South and West Sides, so the pattern below is not a mapping artifact.
Combined vulnerability is the mean percentile rank of three axes: night heat, scarcity of 1995 commerce, and scarcity of third places today. Equal weights, chosen for legibility rather than fitted to any outcome.
What this is not. No mortality data appears anywhere in this analysis. Death records from 1995 with neighborhood detail are not public; the Cook County Medical Examiner's open archive begins in 2008. This is a vulnerability model, and its validation is that it independently reproduces a pattern the historical record already documented.
One known weakness: ZIP 60623 spans both North and South Lawndale, the pair at the heart of Klinenberg's account, so this data cannot resolve that specific comparison. Small community areas such as Fuller Park are also where areal interpolation is least reliable.
Built with Python (GeoPandas, Rasterio, PySTAC) and Leaflet. Basemap by Esri. Analysis and map by Brian Bergstrom, September 2026.