Abstract
In 1938 a field agent for the Home Owners' Loan Corporation walked Chicago, scored its residential districts from A to D, and shaded a base map green, blue, yellow, and red. This paper asks one narrow question. Eighty-five years later, do the things that map sorted still sort with the grades it assigned. The work is a synthesis of the published redlining literature set beside a strictly descriptive analysis of real public data, namely the 1938 HOLC polygons from the University of Richmond Mapping Inequality project joined to 2023 American Community Survey tract estimates for Cook County [1][9]. We make no causal claim from our own cross-section, and we credit the cited studies for any attribution. The descriptive pattern is steady. Walking the grades in order, 2023 median household income falls from A through D, homeownership thins across the same descent, and vacancy stands highest on formerly D-graded ground [1]. Of the Cook County tracts whose center sits inside a 1938 zone, most fall on former C and D land, and present-day public subsidy pools there as well [1]. We read each number as a description of a present-day snapshot shaped by eighty years of compounded change, not as evidence the maps produced it. The honest reading respects a small grade-A sample, top-coded incomes, a tie in the middle of the vacancy gradient, and a boundary rule that files each tract under a single grade [1].
A Surveyor's Hand and Four Colors
Before it became a byword, the thing was a piece of paper. In 1938 the Home Owners' Loan Corporation produced what it called a residential security map of Chicago, and on it a field agent did what agents were doing that decade in dozens of American cities. He carved the residential city into districts, gave each district a letter from A to D, and colored it to match. Green for A. Blue for B. Yellow for C. Red for D. The University of Richmond's Mapping Inequality project, which digitized these documents and made them legible to anyone with a browser, preserves the one-word verdicts the grades carried [9]. A meant Best. B meant Still Desirable. C meant Definitely Declining. D meant Hazardous. The red sat at the bottom of that scale, and the red is where the word redlining comes from.
It helps to picture the object as an object. Someone sat over a base map of the city and decided, district by district, which ground a lender should trust. The agent wrote down what he saw and a good deal of what he assumed: the age of the housing, the trades and incomes of the people, the presence of what the appraisal language of the day called inharmonious or undesirable racial groups. He compressed all of it into one letter. The letter became a color. The color outlived the man who chose it. Chicago was not singled out for this. The HOLC graded residential districts from coast to coast, and the same four-color logic shaded maps in city after city. What recommends Chicago for a close look is narrower. Economists and sociologists have studied it hard, its 1938 map survives in clean digital form, and the present-day data needed to ask whether the old grades still describe anything are public and joinable.
The colors were a judgment about money. A grade told a lender, in a glance, where a mortgage was a sound bet and where it was a hazard, and the language was not subtle about why. Surveys in this program routinely keyed a downgrade to the arrival or proximity of Black residents, immigrants, and the poor, so the map did not merely rank housing stock. It ranked people and then priced the ground they stood on. A family in a red district could be refused a loan outright, or steered to terms a family in a green district never saw, and the gap that opened from that refusal did not stay put in 1938. It moved into who could buy, who could only rent, which blocks gained equity over the next three decades and which were starved of it. The wider history of federal housing policy in the same era, the work of the Federal Housing Administration alongside the HOLC, is by now well documented, and the through-line is that segregation in the United States was written into law and underwriting rather than left to private accident. The HOLC grade was one of the instruments that did the writing.
A little institutional history makes the map legible. The Home Owners' Loan Corporation was a New Deal agency, created in 1933 to refinance mortgages for owners at risk of foreclosure during the Depression, and the security maps came out of its effort to assess the risk attached to the neighborhoods where it held or might hold loans. The grades were not idle description. They were an appraisal instrument, and the appraisal logic of the period treated racial and ethnic homogeneity as a measure of stability and treated the arrival of Black families, in particular, as a sign that values were about to fall. That belief was self-fulfilling. When an appraisal manual instructs that the presence of a single Black family lowers the grade of a block, and lenders act on the grade, the value does fall, because credit dries up, and the prophecy is then cited as proof of the rule. The Mapping Inequality project, in digitizing the area description sheets that accompanied the maps, has put the actual language of these assessments back in front of readers, and the language is blunt about what the surveyors were measuring [9]. Phrases recording the racial and national makeup of a district, and recording it as a hazard, sit in the same forms as the notes on housing age and sewer access, scored as if they were the same kind of fact.
Chicago entered this system already sorted. By the late 1930s the city's Black population was concentrated on the South Side in the long, narrow corridor that came to be called the Black Belt, hemmed in by restrictive covenants, private agreements among white property owners not to sell or rent to Black buyers, which the courts enforced until the Supreme Court declined to in 1948. The HOLC surveyor did not invent that geography. He arrived to find it and graded it, and the corridor that the covenants had packed tight came back from his pen shaded yellow and red. This is the seam the scholarly debate later opened along, namely whether the map made the segregation or recorded a segregation already enforced by covenant and custom. For the purpose of reading the document as a document, the point is simpler. The grades were drawn onto a city that already had a color line, and they translated that line into the language of credit.
What this paper is matters, and what it is not matters more, because the subject invites overreach at every turn. It is an honest synthesis of published scholarship set beside our own descriptive analysis of real public data. The scholarship is the careful causal and historical work other researchers have already done, and we report it as theirs, with their numbers attributed to them in the text. The descriptive analysis is ours, and it rests on two real sources and nothing else. The first is the set of 1938 HOLC polygons digitized by Mapping Inequality and already present in the project's data files [9]. The second is the 2023 American Community Survey five-year tract estimates for Cook County, drawn from the U.S. Census Bureau and joined to those polygons [1]. The join yields a single table in which every present-day census tract that falls inside a 1938 graded zone carries the grade of that zone, so a 2023 number can be read against a 1938 letter. That is the whole machinery, and there is nothing more exotic beneath it.
What we did not do should be said just as plainly, because the temptations here are specific. We ran no experiment. Nobody assigned neighborhoods to treatment and control, and nobody could have, since the assignment happened in 1938 and the people who made it are gone. We did no fieldwork. We walked no blocks, knocked on no doors, measured no building. We fit no causal model. There is no regression here that tries to peel the effect of a grade away from everything else that has befallen a neighborhood in eighty-five years, and where this paper states a causal finding, that finding belongs to a cited study and is marked as such. We interviewed nobody. No resident, lender, or official speaks in these pages, because we spoke to none. The line between description and cause is the line this paper works hardest to hold, and a reader is entitled to test us against it on every page.
The organizing move is plain enough to put in a sentence. We walk the grades in order, A to D, and look at what stands on that graded ground today. The walk is literal. We begin in the green districts the surveyor called Best, pass through the blue and yellow he marked Still Desirable and Definitely Declining, and finish in the red he stamped Hazardous. At each stop we report a small set of present-day descriptive figures. What households there earn, how many own their homes, how much housing sits empty. At the end of the walk we step aside and count where public dollars have come to rest relative to those same old lines. The figures are ours and they describe a 2023 snapshot. For any claim about why the snapshot looks the way it does, the cited literature is where we send the reader, and we say so each time [1].
A word on tone before the numbers begin. Writing about redlining tempts a person to let the moral weight of the subject carry the analytic load, to treat the match between an old red zone and a present hardship as if the match were itself the proof. It is not, and pretending otherwise would make this paper less trustworthy rather than more. The grades were drawn onto a city that already had a racial and economic geography of its own. The decades since have stacked highway routing, public-housing siting, deindustrialization, blockbusting, urban renewal, and mortgage discrimination that ran long after the HOLC stopped grading, all on top of the original lines. A 2023 income figure inside a former D zone is the sum of all of that. We report the sum honestly and refuse to assign it to any single cause from our own data. The restraint is not decoration. It is what lets the descriptive pattern, where it appears, carry any weight at all.
What the Record Already Settles
A reader might fairly ask why a descriptive paper spends its second section on other people's work. The reason is that the question this paper poses, whether the 1938 grades still track present-day outcomes, sits on top of a large and careful literature, and honesty puts that literature first. The causal claims we refuse to make from our own data have, in many cases, already been made responsibly by researchers whose designs were built for the purpose. Reporting their findings as theirs is how a descriptive synthesis earns the right to add a local snapshot at the end. The literature also tells us what to expect before we look, which is its own discipline. A descriptive result that matched nothing in the published record would be a reason for suspicion about the data or the join. A result that lines up with what boundary studies and national overlays have already found is, at least, internally coherent with work that was built to attribute cause and did the harder thing.
The anchor is the work of Daniel Aaronson, Daniel Hartley, and Bhashkar Mazumder, published in the American Economic Journal: Economic Policy in 2021 [2]. The design is the reason it carries weight. Rather than compare graded neighborhoods to the rest of the country, which would tangle the grade together with everything that made a neighborhood gradable in the first place, they compared ground on opposite sides of HOLC boundaries and used a propensity-score approach to sharpen the match. The boundaries are the useful feature. Land a block apart was often alike before the surveyor's pen split it into different colors, so a difference that opens up afterward is harder to wave away. The intuition is worth dwelling on, because it is the move that separates a causal claim from a descriptive one. If two stretches of a city were nearly identical in 1935, similar housing stock, similar residents, similar prices, and a surveyor in 1938 happened to draw a grade line between them, then the two stretches form something close to a natural experiment. Whatever they shared before the line should have kept them on the same trajectory. A gap that opens after the line is drawn, and that grows as the years pass, is hard to pin on anything the two sides did not already share. That is the purchase a boundary gives a researcher, and it is exactly the purchase our centroid-binned cross-section throws away. Reading across those lines, the authors find that the HOLC maps lowered later homeownership, depressed house values, reduced rents, and raised segregation in the decades that followed [2]. That the study comes from economists at the Federal Reserve Bank of Chicago, working partly in the city this paper examines, makes it a natural local anchor as well as a national one.
The same authors circulated a working-paper version through the Chicago Fed that engages Chicago directly and reaches outcomes past housing [11]. That version reports effects of HOLC grading and of yellowlining, the treatment of the C-graded ground just above red, on adult income, on upward mobility for the children who grew up in graded neighborhoods, and on modern credit scores [11]. The reach there should stop a reader. A line drawn to ration mortgage credit in 1938 shows up, in this work, in the credit files of people born long after the maps were boxed away. That is the sort of long shadow a careful causal design can document and a cross-section like ours cannot.
Boundary logic returns in Jacob Krimmel's working paper, which sets redlined tracts against the yellowlined tracts right next to them [5]. The comparison is deliberately conservative, since C and D ground were often neighbors and often similar to begin with, so comparing red to yellow asks a harder question than comparing red to green. Krimmel finds persistently lower housing supply and lower density on the redlined side of those lines, and, in the result that does the most to isolate the map's own hand, he finds that the difference holds even for boundary pairs that started out all white [5]. That last point bears on the debate taken up below, and the reason needs spelling out. The strongest version of the skeptical argument is that the maps did nothing but trace a color line already drawn by covenant, custom, and prior lending, so that any later gap merely reflects who was living where before the surveyor arrived. The all-white boundary pairs are the test case that argument has to answer. If two adjacent stretches were both entirely white in the 1930s, then prior racial sorting cannot explain a gap that opens between them, because there was no racial difference to sort. A persistent gap in those pairs points to the grade itself, acting through credit, as the thing that pried them apart. Krimmel reports exactly such a gap, which is why his finding weighs against the pure documentation story even though it does not refute it across the board [5].
Ian Appel and Jordan Nickerson bring a regression-discontinuity design to the 1940 HOLC boundaries [6]. The approach watches what happens to outcomes as you cross a grade line, treating the line as a sharp break the underlying neighborhood did not necessarily share. The logic is close kin to the boundary comparison but more local. A discontinuity design looks at the immediate neighborhood of the line itself, the parcels just inside and just outside, on the reasoning that the closer you stand to the boundary the more alike the two sides are in everything except their grade. They estimate that redlined neighborhoods carried roughly 4.8 percent lower home prices in 1990, half a century after the maps were drawn, and they tie that price gap to two concrete housing facts, fewer owner-occupied homes and more vacant structures [6]. The vacancy result will return later in this paper, since vacancy is one of the present-day figures our own walk reports, and it helps to have a causal study on the record that read disinvestment off the same kind of boundary. The 4.8 percent figure is also a useful corrective to anyone tempted to read the redlining literature as claiming enormous overnight effects. The estimated price gap at the boundary is real and durable, and it is also modest in size, the kind of difference that compounds quietly across decades rather than the kind that announces itself in a single year.
Jacob Faber widens the lens to most of a century [3]. Working with census data spanning decades, Faber compares cities the HOLC appraised against cities it did not, and finds that the appraised cities grew more segregated than the un-appraised ones [3]. His timing is specific, and the timing is the argument. The gap between appraised and un-appraised cities opened between 1930 and 1950, the years that bracket the HOLC's own work, and it had not closed by 2010 [3]. That shape is what a causal story predicts and a coincidence does not. If the maps were merely incidental, there would be no reason for the divergence to begin precisely when the appraisals happened, and no reason for it to persist for sixty years after. The publisher's page for this article sits behind a paywall, which we note in the interest of full disclosure, but the finding and its framing are well attested across the secondary record. Faber's contribution here is chronological. He places the divergence in the window when the maps were made and shows it did not heal on its own across the decades that followed, which matters for a paper like ours because it means the 2023 pattern we describe is the far end of a trend that other researchers have already traced back to its beginning.
The closest model for what this paper actually does is the National Community Reinvestment Coalition overlay by Bruce Mitchell and Juan Franco [4]. Their analysis lays present-day demographic and economic data over the old HOLC grades at national scale and reports the pattern descriptively, exactly as we do. Two of their figures anchor the comparison. About 74 percent of formerly D-graded, hazardous areas are low-to-moderate income today, and roughly 64 percent are majority-minority [4]. Those numbers are theirs, national in scope, descriptive in kind, and we cite them as the benchmark against which our Cook County snapshot can be read. When our own footprint turns out to be dominated by former C and D ground, the NCRC overlay is the reason that result reads as of a piece with the national picture rather than as a Chicago oddity.
The NCRC work also matters because it is honest about its own genre, and that honesty is the model this paper tries to follow. Mitchell and Franco do not claim their overlay proves the maps caused present-day low income or present-day racial composition. They report an association at national scale and let the causal studies carry the rest, which is exactly the division of labor adopted here. A descriptive overlay has a real job even though it cannot attribute cause. It establishes the scope and shape of the present-day pattern, it shows whether the association is broad or local, and it gives the causal studies something to be the explanation of. The boundary and discontinuity papers tell us that the grade left a mark in the places where the mark can be isolated. The national overlay tells us that the pattern those papers explain is not confined to their study sites but is visible across the formerly graded country. Our Chicago read sits one level down from the NCRC's national one, the same kind of overlay applied to a single county, and it inherits both the value and the limits of the genre.
The legacy the literature documents does not stop at dollars and tenure. Jeremy Hoffman, Vivek Shandas, and Nicholas Pendleton studied intra-urban heat across 108 U.S. urban areas and found that formerly redlined areas tend to run hotter than their non-redlined neighbors, by as much as roughly seven degrees Celsius in the most extreme cases and by about 2.6 degrees on average across the cities they examined [7]. In their account, 94 percent of the studied cities show the redlined-hotter pattern, which is a near-universal result rather than a scattered handful of examples [7]. The built environment carries the difference. Decades of uneven investment left some neighborhoods with more pavement and asphalt and less tree canopy and green space, and the thermometer reads the result back out, since dark surfaces hold heat and trees shed it. That a 1938 credit grade should turn up in a present-day summer temperature is a sharp illustration of how far these inheritances travel from their origin. A line drawn to ration mortgage money now registers as degrees of heat that fall hardest on the residents least able to cool their homes, and it stretches the meaning of the word outcomes well past the housing variables our own data can see.
Health is the other broad frontier the literature has opened, and it is where the stakes of the whole inquiry become bodily. A systematic review by Kraus and colleagues gathers a growing body of studies tying worse HOLC grades to worse present-day health [10]. Across the work they survey, formerly lower-graded areas show higher rates of asthma, higher rates of preterm birth, and higher cardiovascular disease, and the review reports a life-expectancy gap on the order of 3.6 years associated with worse grades [10]. The pathways are not mysterious once the earlier findings are in view. Hotter neighborhoods with less canopy, older housing with more lead and mold, fewer grocery stores and clinics, more exposure to industrial land use, and the chronic stress of disinvestment all plausibly feed those health gaps, and they are themselves downstream of the credit and investment patterns the other studies trace. We report these as the review's synthesis of many studies, not as a finding of our own, since our data hold no health variables at all. The point of including them is to mark the full reach of what other researchers have connected to these maps, so that a reader does not mistake income, tenure, and vacancy for the whole of the story. A gap of 3.6 years of life expectancy is, in the end, what gives the income and vacancy gradients their weight. The housing numbers measure the conditions. The health numbers measure what those conditions do to the people living inside them.
A synthesis that presented only the studies supporting a strong map effect would be a brief, not a review. The skeptical case deserves real room, and it has a serious advocate in Amy Hillier [8]. Working in Philadelphia with address-level mortgage data, Hillier argues that the HOLC grades on their own do not account for most of the lending differences she observes across the city [8]. Her central claim is about sequence. Lenders, on her reading, were already steering clear of the areas that would later be shaded red before the HOLC maps existed, which suggests the maps may have documented and hardened a discriminatory pattern the private market had already set in motion rather than originating it [8]. This is not a fringe position and it does not fold under pressure. It frames the live scholarly disagreement with some precision, namely how much the maps caused and how much they recorded a city other hands were already sorting.
Hillier's evidence is worth taking on its own terms rather than as a foil. Address-level mortgage records are a finer instrument than the tract or boundary data most of the other studies rely on, and they let her ask not whether redlined areas had less lending, which is not in dispute, but whether the grade was the thing that produced the shortfall once everything else about a property is accounted for. Her answer, in Philadelphia, is that much of the lending difference was already in place and traceable to factors other than the colored line [8]. The caution that follows is about reading the maps as a simple instruction that lenders received and obeyed. If the avoidance of red areas predated the maps, then treating the grade as the lever that moved the credit can mistake a symptom for a cause. A careful reader should hold this alongside the boundary findings rather than choosing between them. It is entirely possible, and probably correct, that the private market was already discriminating before 1938, that the maps then codified and amplified that discrimination, and that the relative size of the prior pattern and the added push varies from city to city. Hillier's Philadelphia is one city studied with unusually fine data, and her caution travels.
Step back from the individual studies and a pattern in the methods themselves comes into view. The strongest evidence for a map effect does not come from comparing graded cities to ungraded ones, or red ground to the national average, because those comparisons fold in everything that made a place gradable to begin with. It comes from the studies that found a way to compare like with like, ground on opposite sides of a single line that the map split and history had not. Aaronson, Hartley, and Mazumder build that comparison with boundaries and propensity scores [2]. Krimmel builds it by pairing red ground with the yellow ground next door, and tightens it further with the all-white pairs [5]. Appel and Nickerson build it with a discontinuity that looks only at the parcels hugging the line [6]. Three research teams, three designs, one shared logic, and a convergent answer that the grade left a mark. That convergence is what makes the causal literature hard to dismiss, and it is also precisely the kind of footing our cross-section does not have. We compare whole tracts binned by grade across the entire mapped city, which means our gradient carries all the confounding the boundary studies worked to strip out. The difference is not a flaw in our data. It is the difference between a description and an experiment, and it is why the attribution stays with them.
The honest reading is that the disagreement is genuine and the evidence does not collapse onto one side. Krimmel's finding that the redlined-versus-yellowlined gap holds even for initially all-white boundary pairs cuts against a pure documentation story, since it locates an effect where prior racial sorting cannot explain it [5]. The boundary and propensity-score design of Aaronson, Hartley, and Mazumder, and the discontinuity of Appel and Nickerson, are both built to separate the map's break from the neighborhood's prior trend, and both find a break [2][6]. Hillier's address-level sequence argument stands on the other side as a real constraint on how much any of those breaks can be charged to the colored lines alone [8]. A reader can hold both at once. The maps appear to have done something, in the places and the designs where researchers have looked with care, and the maps also fell on a city whose discrimination ran ahead of them. This paper does not settle that question, because its cross-section cannot. It records the debate, credits both sides, and turns to its own modest descriptive task with the causal question left open on purpose.
So the record settles a great deal and not everything. It settles that the association between old grades and present outcomes is large, replicated, and visible across income, housing, segregation, temperature, and health [2][11][5][6][3][4][7][10]. It settles that at least some of that association reflects the maps working on markets rather than merely mirroring earlier prejudice, in the particular designs that can test the difference [2][5][6]. It leaves open, honestly, the share of the whole that the maps caused as against documented [8]. With that record in hand we can lay the old map over the new city and start the walk, knowing the description to come is a local snapshot resting on a national foundation we did not have to rebuild.
Laying the Old Map Over the New City
The walk is only as trustworthy as the join beneath it, so the join gets explained in plain terms here, with its limits on the table, before any gradient is reported. The procedure is deliberately simple, and that simplicity is both its strength and the source of its main cautions.
Start with the two layers. The bottom layer is the 1938 HOLC map, a set of polygons, each a closed shape over the city carrying a grade. The top layer is the present-day census geography of Cook County, a set of tracts, each a small statistical area for which the American Community Survey reports estimates. The task is to decide which 1938 grade, if any, lies under each 2023 tract. We settle it with one rule. The Census Bureau publishes for every tract an official internal point, a representative latitude and longitude the Bureau guarantees falls inside the tract, recorded in the fields it labels INTPTLAT and INTPTLON. We test whether that internal point falls inside an A, B, C, or D residential polygon. If it does, the tract takes that grade. If the point lands inside no graded residential polygon, the tract takes no grade and drops out of the analysis [1].
The present-day figures come from a small, named set of American Community Survey variables, and naming them is part of keeping the analysis checkable. Median household income is the table the Census publishes as B19013, reported once per tract. Tenure comes from table B25003, which splits occupied units into owner-occupied and renter-occupied, so the homeownership rate is owner-occupied units divided by all occupied units. Vacancy comes from table B25002, which separates a tract's total housing units into occupied and vacant, so the vacancy rate is vacant units divided by total units. Each figure is a 2023 five-year estimate, which means it pools survey responses gathered from 2019 through 2023 rather than measuring a single year, a feature of the ACS that buys smaller margins of error at small geographies in exchange for describing a five-year average rather than a snapshot. We aggregate within each grade in the way that respects what each statistic is. For income, where a tract-level median cannot be meaningfully summed, we report the median of the tract medians and its quartiles. For tenure and vacancy, which are built from counts, we sum the underlying units across all tracts in a grade and form the rate from the totals, so that larger tracts carry the weight their size warrants rather than every tract counting equally regardless of how many households it holds.
Two kinds of 1938 ground fall out by construction. The HOLC map carries the four residential grades, but it also carries zones marked Commercial and Industrial, along with a set of polygons bearing no grade at all. Those ungraded zones, twenty polygons including the Loop, are not a grade and are dropped, since a Commercial designation says nothing about residential desirability on the A-to-D scale, and folding it into the walk would only blur the gradient we are trying to read [1]. The residential grades we keep came from clean A, B, C, and D labels once that filtering was done, so the letters going into the join are unambiguous [1].
The footprint that results is the natural place to set the scene, because it tells the reader how much of today's county the old map actually touches and which grades dominate where it does. Of the 1,332 Cook County tracts in the 2023 data, 769 have a center that falls inside a 1938 HOLC residential zone [1]. The rest carry no grade, and the reason is geographic rather than mysterious. The 1938 survey covered far less ground than present-day Cook County, so most of the county simply lies outside the mapped area, and those tracts are excluded as expected rather than treated as a fifth category [1]. A little over half the county's tracts, then, sit on graded ground, and the rest lie beyond the edge of anything the surveyor ever shaded.
The gap between the 1938 map and the 2023 county is itself a piece of history, and it shapes what the walk can claim. The HOLC surveyed the developed city as it stood in the late 1930s, which in Cook County meant Chicago proper and a ring of older suburbs, not the vast tract-house expansion that would fill the postwar decades. Most of that postwar growth happened on ground the 1938 map never reached, and much of it happened under the next generation of federal housing policy, the FHA and VA mortgage guarantees that built the white suburbs while the explicit HOLC grades were being retired. So the tracts that fall outside our footprint are not a random remainder. They are disproportionately the newer, outer, postwar parts of the county, developed under a different set of rules, and our analysis is silent about them by construction. When this paper says the grades still track outcomes, it means within the older, mapped core of the county, the part of Chicago that already existed to be graded in 1938. That is a real limit on reach, and it is also a reminder that the redlining map is only one chapter in a longer federal housing story whose later chapters were written somewhere else on the map.
Inside that 769-tract footprint, the grades are anything but evenly spread, and the imbalance shapes what the walk can and cannot say. Former C zones dominate, with 416 tracts, 54.1 percent of the graded footprint, more than every other grade combined [1]. Former D zones come next at 247 tracts, 32.1 percent [1]. Former B zones hold a smaller share at 91 tracts, 11.8 percent, and former A zones are a sliver, 15 tracts, 2.0 percent of the footprint [1]. Put those together and more than eight in ten graded tracts in present-day Cook County sit on ground the 1938 surveyor called Definitely Declining or Hazardous. The green at the top of the scale, the Best ground, barely registers in the present-day tract count. That distribution is a quiet finding in its own right, since it means the city the HOLC graded was, across its mapped extent, mostly graded down rather than up. It is tempting to read that as a verdict on the surveyor's pessimism, but the more careful reading is about what the program was for. The HOLC was assessing risk in an already-built industrial city during the Depression, a city of aging housing stock, dense working-class wards, and a tightly bounded Black population, and the appraisal categories of the day treated most of that as a hazard or a decline waiting to happen. A grading system built on those assumptions, applied to that city, was always going to shade more ground yellow and red than green. The skew is a property of the instrument meeting the place, and it is the backdrop against which the present-day gradient has to be read. For method it is also a warning. The grade-A group is small, and small groups make less stable statistics.
The land-area companion figures tell a slightly different version of the same story and belong alongside the tract counts. By area, former C zones cover 280.4 square kilometers, the largest share at 48.5 percent, and former D zones cover 162.5 square kilometers, 28.1 percent [1]. Former B zones cover 93.2 square kilometers, 16.1 percent, and former A zones cover 42.5 square kilometers, 7.3 percent of the graded land [1]. The A share of land, 7.3 percent, runs larger than the A share of tracts, 2.0 percent, the kind of small discrepancy you would expect when high-grade districts hold physically larger, less densely subdivided tracts. The whole graded footprint comes to 578.7 square kilometers across the 769 tracts [1]. We report both shares because they answer slightly different questions, one about how many present-day statistical communities sit on graded ground and one about how much physical city does.
Two honesty points fall directly out of the rule, and the later sections lean on both, so they belong here and not in a closing aside. The first is about boundaries. Because each tract is filed by a single internal point, a tract whose area straddles a 1938 grade line is counted entirely in whichever grade its center happens to land in, and the partial overlap is not split between the two grades [1]. A tract that is sixty percent former C and forty percent former B is recorded as wholly C if its center sits in the C zone, and nothing of the forty percent survives the assignment. This keeps the method transparent and repeatable, since an internal point is a fixed official coordinate rather than a judgment call, but it coarsens the picture near every boundary. The reader should imagine the gradient as built from whole tracts sorted into single bins, not from finely partitioned land.
A word on why we assign by centroid rather than by area is owed here, since the choice is real and a different analyst might choose otherwise. Two methods were available. We could split each tract's statistics across every grade its area touches, weighting by how much of the tract sits in each zone, or we could assign the whole tract to the single grade under its official center point. We chose the second. The reasons are practical and they are also about honesty. An area-weighted split would require apportioning a tract's households and housing units across grade boundaries as though they were spread evenly over the tract's land, which they are not, since population clusters and empty land does not. That apportionment would manufacture a precision the underlying data do not have, dressing an assumption about where people live inside a tract as if it were a measurement. The centroid rule makes no such assumption. It uses a fixed, official coordinate the Census already publishes, assigns each tract once, and leaves the result fully reproducible by anyone with the same two files. The cost is the coarsening near boundaries already described. The benefit is that no figure in this paper depends on a guess about the internal distribution of a tract, and a reader who prefers area-weighting can re-run the join the other way against the posted files and see how much it moves [1].
The second point is about the grade-A group. Fifteen tracts is a thin base on which to compute a median, a quartile, or a rate, and the statistics for grade A are accordingly the least stable in the study [1]. A few unusual tracts can move an A-grade figure in a way they could never move a C-grade figure built from 416. We flag this now because the very first stop on the walk is grade A, and its headline numbers, real as they are, deserve to be read with that instability in mind rather than treated as precisely as the larger groups' figures. The walk reports grade A honestly and then says, each time it matters, that the group is small.
One practical note closes the method. Every figure in this paper is reproducible. The 1938 HOLC polygons, the joined 2023 ACS tract estimates, and the computed grade assignments are posted at the project's research data archive, and a reader who wants to check a number, re-run the join, or test a different boundary rule can do so against the same files we used [1]. We would rather invite that scrutiny than ask for trust. With the layers aligned, the exclusions named, the footprint described, and the two load-bearing cautions stated, the map is laid over the city and the walk can start where the surveyor started, in the green.
Green Ground, Eighty-Five Years On
The walk begins where the surveyor's confidence ran highest. Grade A, shaded green, carried the verdict Best, and these were the districts the HOLC judged safest to lend against in 1938 [9]. The A grade went to neighborhoods the appraisers read as new, homogeneous, and in demand, the kind of ground where they expected values to hold or climb and a mortgage to be repaid without trouble. In the racial logic of the period, homogeneous meant white and meant insulated from the arrival of anyone the surveys treated as a hazard, so the green at the top of the scale was an economic judgment and a racial one at the same time, the same judgment that pushed other districts down. Eighty-five years later we can ask what stands on that green ground, and income is the first figure to report, because income opens the descending pattern the rest of the walk will trace.
On former A ground, 2023 median household income stands at $207,434, taken as the median of the tract medians for the fifteen grade-A tracts [1]. The figure sits at the top of the walk, well clear of every grade below it, and it is the anchor against which the descent is measured. The spread around it is wide. The interquartile range runs from a first quartile of $153,006 to a third quartile of $250,001, so the middle half of grade-A tracts report median household incomes between roughly one hundred fifty-three thousand dollars and the very top of the scale [1]. Even the bottom of that middle half, $153,006, would stand above the central figure for every other grade, which is the first concrete sign that the green ground of 1938 is, in 2023, the affluent ground.
The figure has to be read with its honesty attached, and two cautions follow from the method. One carries over from the join. Grade A is only fifteen tracts, its statistics the least stable in the study, so the $207,434 median rests on a thin base, and a few atypical tracts could move it in a way they could not move the larger grades [1]. The other is specific to income and to the top of the scale. The American Community Survey top-codes its highest median-income bracket at $250,001, which is to say the Census does not report a tract median above that ceiling and records anything at or beyond it as $250,001 [1]. The grade-A third quartile sits at exactly $250,001, the cap itself [1]. So the true top of the grade-A distribution is censored. At least a quarter of grade-A tracts press against the ceiling of what the data can express, and their real medians, whatever they are, run higher than the number we are allowed to see. The green ground is therefore at least as affluent as the figure says and very possibly more so. We report the censored number and decline to guess past it.
Median household income falls sharply below the grade-A districts
Reading the bars in order makes the shape plain. Grade A stands alone at the top at $207,434 [1]. The fall from A to B is the largest single step in the whole walk, since grade B comes in at $80,814, far below the green ground rather than just beneath it, and the bars for B, C, and D then ease down more gently from there toward the floor [1]. The visual story is not a smooth slope from top to bottom but a cliff right below the green, followed by a longer, shallower decline through the blue, yellow, and red. That shape matters before the middle grades are taken up in detail, because it puts most of the distance between best and worst into the gap between A and everything else, and it previews how far the formerly Best ground has pulled away from the rest of the graded city.
Holding to the discipline of the paper, the green-ground figures describe a present-day snapshot and nothing more. A median of $207,434 across fifteen censored, small-sample tracts is a fact about Chicago in 2023, the sum of eighty-five years of compounded change the figure cannot pull apart [1]. What it earns is the first stake for the descent, the top of a walk whose lower stops, through the blue and yellow of the middle grades and into the red at the bottom, carry the homeownership and vacancy figures that fill out the picture. The green is the high end. The rest of the walk follows the colors down.
The Long Descent Through Blue and Yellow
The walk does not drop off a cliff after the green. It slopes. Step from the A districts into the blue ones the 1938 surveyor called Still Desirable, then into the yellow he marked Definitely Declining, and the ground gives way by degrees. This is the middle of the gradient, and it is where most of the graded city actually lives. Of the 769 Cook County tracts inside a 1938 zone, 91 fall on former B ground and 416 on former C, so the two middle grades together hold 507 tracts, well over half the footprint and the clear majority of its present-day population [1]. Whatever the colors track, they track it here most broadly.
Read tenure down the slope and the descent is plain. Homeownership, computed as owner-occupied units over occupied units and aggregated within each grade, stands at 90.0 percent on former A ground, 57.6 percent on B, 48.5 percent on C, and 44.1 percent on D [1]. The largest single step is the one off the green. From A to B the rate drops 32.5 percentage points, from nine in ten households owning to fewer than six in ten [1]. The B-to-C step takes off another nine points and change, C-to-D another four, so by the bottom of the walk the gap from the top has widened to 46.0 percentage points, with C sitting 41.5 points below A and B 32.5 points below [1]. The blue and the yellow are not a calm plateau between a high green and a low red. They are the steep face of the slide, the stretch where ownership turns into tenancy for most of the people the old map still shadows.
The counts underneath the percentages carry their own weight, since the rates are built from real household totals. On former A ground, of 24,464 occupied units, 22,028 are owner-occupied [1]. On former B ground the count is 92,738 owners out of 161,088 occupied units, on C ground 298,332 out of 614,953, and on D ground 131,290 out of 297,824 [1]. Two things the percentages hide jump out of the raw counts. One is scale. The C grade sits second from the bottom on the ownership ladder, yet it is by far the largest housing market of the four, with more occupied units than A, B, and D combined, so the 48.5 percent ownership rate there describes the tenure of more households than any other figure in the walk [1]. The other is the thinness of the green. Grade A holds about twenty-four thousand occupied units across its fifteen tracts, an order of magnitude fewer than the C grade, so the 90.0 percent figure, high as it is, rests on a small slice of the city's housing [1]. The descent in tenure is steep, and it is lopsided, with most of the households sitting at the bottom three rungs and only a sliver at the top.
Homeownership thins steadily from the green grades to the red
The income gradient corroborates the same descent on its own scale. The median of tract medians is $80,814 on former B ground and $71,898 on former C, which works out to 39.0 percent and 34.7 percent of the grade-A median [1]. The B figure sits a little above a third of the green-ground median reported in the last section, the C figure a hair lower, and the two middle grades bracket the great mass of the graded county between roughly a third and two-fifths of what the A tracts post today [1]. Income and tenure are different measurements, and they need not move together, yet here they walk down in step. Each color the surveyor laid down in 1938 sits today on a lower rung of both ladders than the color above it.
The spreads tell as much as the medians, and the middle grades spread wide. On former B ground the interquartile range of tract medians runs from $56,724 at the first quartile to $116,271 at the third [1]. On former C ground it runs from $51,100 to $94,092 [1]. The ranges overlap heavily, which is the honest texture under the clean step from 80,814 to 71,898. The blue and the yellow are not two tidy income bands stacked one atop the other. They are two broad, overlapping distributions whose centers sit close together and a little below the green, so that a particular C tract can easily be richer than a particular B tract even though the typical C tract is poorer than the typical B tract. The gradient is a statement about central tendency across hundreds of tracts, not a guarantee about any single block, and the wide quartiles are the reminder of that. They also explain why the largest income step in the walk is the one off the green and not any step within the middle. The distance from A to B is a cliff. The distance from B to C is a short slope between two ranges that share most of their span.
Set against this descriptive Chicago slide is a body of work that studied tenure with designs built to isolate cause, and it should be read on its own terms rather than folded into ours. Aaronson, Hartley, and Mazumder, comparing neighborhoods that landed on opposite sides of a grade line and sharpening the match with propensity scores, find that the HOLC maps lowered later homeownership along with house values and rents and raised segregation in the decades that followed [2]. Appel and Nickerson, running a regression discontinuity on 1940 HOLC boundaries, tie redlined ground to fewer owner-occupied homes and report roughly 4.8 percent lower home prices in 1990 [6]. Krimmel, setting redlined tracts against the yellowlined tracts immediately next to them, finds persistently lower housing supply and density, an effect that holds even for boundary pairs that began all white, which pushes the reading toward the map's own market hand rather than a proxy for who already lived there [5]. Those studies reach for cause with instruments our cross-section does not have. Our numbers do something narrower and still worth saying. In 2023 Cook County, the tenure gap the literature traced from grade lines is exactly the shape you find walking the blue and yellow ground today, ownership thinning as the grades fall. We describe the present-day pattern. The causal weight stays with the cited work.
The blue and yellow grades carry one more thing the bottom of the walk does not, which is sheer numbers. Grade D is the steepest single rung, but the C tracts alone, at 416, outnumber every other graded tract put together, and C by itself is 54.1 percent of the footprint [1]. The middle of the gradient is therefore where the descent touches the most ground and the most households. A pattern that held only in a handful of tracts would be a curiosity. This one runs through the majority of the formerly graded city, which is part of why it earns a careful look rather than a wave of the hand.
Red Ground and the Fingerprint of Disinvestment
At the foot of the walk is the red, the districts the 1938 surveyor stamped Hazardous and colored for warning. These were the areas the appraisers wrote off, and in Chicago the red fell heavily on the South and West Side neighborhoods where Black residents and recent immigrants were concentrated, the Black Belt foremost among them. The word Hazardous did real work. It told a lender that a mortgage here was a loss waiting to happen, and the refusal of credit that followed helped make the prophecy true, since a neighborhood denied home loans could not easily maintain or transfer its housing, and decline followed the denial. Of the graded footprint, 247 tracts sit on former D ground, 32.1 percent of the total, the second-largest group after the yellow and the bottom of every gradient this paper traces [1]. Income reaches its floor here. The median of tract medians on D ground is $65,240, which comes to 31.5 percent of the grade-A median, a shade under a third of what the green tracts post today and the lowest of the four colors [1]. The descent that began off the green and steepened through the blue and yellow ends on the red, where the median household earns roughly a third of its counterpart in the districts the surveyor called Best.
A reading of the income floor needs the same honesty the green ground got, and it points the opposite way. Where grade A was censored at the top by the $250,001 cap, grade D is uncensored and unambiguous at the bottom. Its first quartile of tract medians sits at $42,639 and its third quartile at $96,518, a range that, like the middle grades, is wide and overlaps with the grades above it [1]. The third quartile of D, near ninety-seven thousand dollars, sits above the median of B, which is a reminder that the better-off quarter of red-graded tracts out-earns the typical blue-graded one, and that the gradient is a tendency rather than a wall. What the red ground does not have is a thin top quartile pressed against a cap. Its distribution is fully visible, and its center is the lowest of the four, $65,240, a real number drawn from the 241 D tracts that carry a median rather than the Census not-available sentinel [1]. The floor of the walk is as legible as any stop on it.
Vacancy is the clearest reading of disinvestment in the data, and it is the one the red ground carries hardest. Computed as vacant units over total housing units and aggregated within grade, the vacancy rate runs 5.1 percent on former A ground, 8.6 percent on B, 8.6 percent on C, and 11.3 percent on D [1]. The red is the high end, 6.2 percentage points above the green and well above the middle grades [1]. Roughly one in nine housing units on former D ground sits vacant in the 2023 pooled estimate, against about one in twenty on former A ground [1]. Empty housing is the most legible mark disinvestment leaves on a map. A unit stands vacant because the demand to fill it, the capital to repair it, or the value to justify either has thinned, and on the red ground all three have thinned the most.
The unit counts give the rate its size. On former D ground, 38,049 units stand vacant out of 335,873 total housing units [1]. On former A ground the count is 1,310 vacant of 25,774, on former B ground 15,237 of 176,325, and on former C ground 58,146 of 673,099 [1]. The C grade again carries the largest absolute number of empty units, because it carries the largest housing stock, but its rate sits in the middle of the walk rather than at the bottom, which is the difference between counting empty buildings and measuring the share that stand empty. The D rate is the one that registers as disinvestment, since it is the highest share, and the share is what tells you whether a neighborhood is losing its grip on its own housing rather than simply being large. Thirty-eight thousand empty units on the red ground is a great deal of housing sitting idle in a city with no shortage of households that need it, and the 11.3 percent figure says that idleness is concentrated where the surveyor once wrote Hazardous.
Vacancy is highest on the formerly hazardous ground
Honesty requires naming what the vacancy numbers do not do, which is climb in a clean straight line. B and C tie at 8.6 percent [1]. The gradient is not strictly monotonic, and the tie falls right in the middle of the walk, where the most tracts are. It would be easy to round the story into a tidy A-through-D staircase, and the data do not support that. What they support is narrower and still pointed. Vacancy is lowest on the green, highest on the red, and elevated across the whole middle, with the two middle grades indistinguishable from each other on this measure [1]. The disinvestment signal is sharp at the bottom of the walk and muddy in the middle, and saying so is more useful than smoothing it.
The literature that read vacancy and disinvestment as a long-run inheritance did so with designs our snapshot lacks, and the distinction matters at the bottom of the walk as much as anywhere. Appel and Nickerson, in the discontinuity work already cited, tie redlined ground to lower 1990 prices and fewer owner-occupied homes, and to more vacant structures as well, the same fingerprint our 2023 cross-section shows on D ground, reached by a method built to argue cause [6]. The built-environment legacies run wider than housing alone. Hoffman, Shandas, and Pendleton document that formerly redlined areas tend to run hotter than their neighbors, a physical inheritance written into pavement and missing tree canopy [7]. The Kraus review ties worse HOLC grades to higher present-day asthma, preterm birth, and cardiovascular disease, and to life expectancy lower by roughly 3.6 years [10]. Those are findings from the cited literature, attributed to it, describing inheritances our data do not measure. Our vacancy number stays in its lane. It is a description of the 2023 housing stock on ground that was graded D in 1938, the highest vacancy of the four colors, offered as a present-day pattern and not as proof that the grade produced it.
Where the Public Dollars Came to Rest
Step off the walk down income, tenure, and vacancy and ask a different kind of question, geographic rather than economic. Where, relative to the 1938 grades, does present-day public investment sit. The question is worth asking because it turns the analysis around. The income, tenure, and vacancy figures describe what private markets and eighty years of history left behind on the graded ground. The location of public dollars describes where the city, today, has chosen to intervene, which is a different kind of fact, a record of present-day policy rather than accumulated private outcome. If the old grades still organize where the city spends, that is a sign the geography the HOLC mapped is still the geography Chicago governs around, whether or not anyone in the relevant offices has ever looked at a 1938 security map. The answer is a plain point-in-polygon count from City of Chicago open data, each subsidized site placed by its coordinates inside whichever 1938 zone contains it, and the count leans hard toward the colors the surveyor marked down.
Take the affordable rental developments first. Of those that fall inside a graded zone, none sit on former A ground, 16 sit on B, 246 on C, and 274 on D [1]. In share terms that is 3.0 percent in the blue, 45.9 percent in the yellow, and 51.1 percent in the red, so about 97 percent of subsidized affordable rental housing inside the graded footprint sits on former C or D ground, and not one development on former A [1]. The TIF and RDA funded projects, the city's tax-increment financing and redevelopment dollars, fall the same way. Inside a graded zone, 1 project sits on former A ground, 38 on B, 213 on C, and 286 on D, which is 0.2 percent in the green, 7.1 percent in the blue, 39.6 percent in the yellow, and 53.2 percent in the red, roughly 93 percent landing in C or D [1]. Two separate streams of public money, one for housing and one for redevelopment, pool overwhelmingly on the ground the 1938 map graded worst.
Public housing and TIF dollars concentrate on the downgraded grades
The denominators deserve to be stated plainly, since the honest count is smaller than a careless one. The analysis reads 598 affordable rental developments, all 598 geocoded, of which 536 fall inside a graded zone and 62 fall outside any 1938 zone [1]. It reads 768 TIF and RDA projects, 759 of them geocoded, of which 538 fall inside a graded zone and 221 outside [1]. Those are the true record counts, pulled with a parser that correctly handles the embedded newlines inside description and address fields rather than the larger physical line counts that naive counting would report from the same files [1]. The shares above are computed against the sites that actually land inside a graded zone, not against the full citywide rolls, and the substantial counts sitting outside any 1938 zone are a reminder that the graded footprint is only part of the present-day city.
The two streams are worth holding apart for a moment, because they answer to different logics and still land in the same place. Affordable rental housing is built where land and need line up, and its near-total absence from former A ground, zero developments of the 536 inside the footprint, partly reflects that the green tracts are few and expensive and not where subsidized rental gets sited [1]. Tax-increment financing is a different animal. A TIF district captures the growth in property-tax revenue within its bounds and plows it back into the area, and it is nominally a tool for spurring development in places judged blighted or underinvested. That 53.2 percent of the graded-zone TIF and RDA projects sit on former D ground, and 39.6 percent more on former C, says the city's designation of where development needs a public thumb on the scale has tracked, decades later, the same ground the HOLC marked down [1]. One instrument follows need, the other follows a blight designation, and both arrive at the red and yellow. The convergence is descriptive, not causal, but it is striking that two separate bureaucratic judgments about where public help belongs both point back at the surveyor's worst grades.
Read the geography with care and it cuts more than one way. Public money concentrating on formerly downgraded ground is exactly what you would expect if subsidy follows present-day need, and the earlier sections establish that the need is real there, with lower incomes, thinner ownership, and higher vacancy on the C and D ground [1]. A dollar of affordable housing or redevelopment financing is meant to flow toward the places private capital has left, and on this map those places line up with the old red and yellow. The count describes where the dollars sit. It does not establish that the 1938 grades routed them there, nor whether the subsidy is adequate to what it meets, and a single point-in-polygon snapshot cannot tell those stories apart. Nor does a count of projects measure money, since one large development and one small grant each register as a single point regardless of how many dollars or units stand behind them, so the geographic concentration we report is a concentration of sites and not a tally of spending. A reader should take the pattern for what it is, a map of where the city has placed its housing and redevelopment efforts, and not as a budget. The national descriptive overlay runs parallel to this local count. Mitchell and Franco report that about 74 percent of formerly D-graded areas are low-to-moderate income today and roughly 64 percent are majority-minority, the present-day profile that present-day subsidy is built to reach [4]. The Chicago count is one city's version of that picture, drawn from its own open data and held to the same descriptive standard.
What a Gradient Cannot Prove
A descriptive cross-section earns its keep only if it stays honest about what it is, so the limits get their own room here rather than a parting line. The central limit governs everything above. This is a snapshot, not a causal estimate, and the A-to-D gradient it traces reflects eighty years of compounded policy, capital, and demographic change rather than the 1938 maps acting alone [1]. Every causal claim in this paper belongs to the cited literature, reached by boundary comparisons, propensity scores, and regression discontinuities our single cross-section does not attempt [2][5][6]. When we say income and homeownership fall and vacancy is highest on the red, we are describing the shape of Chicago in 2023. We are not saying the surveyor's pen drew that shape.
The specifics deserve to be named one by one, the inconvenient ones included. The grade-A group is only 15 tracts, the smallest of the four by a wide margin, so its statistics are the least stable in the study and the top of the walk rests on the thinnest sample [1]. ACS median household income is top-coded at $250,001, and the grade-A upper quartile sits exactly at that cap, so the true top of the A distribution is censored and the green-ground income reported earlier understates rather than overstates the real spread [1]. Fifteen tracts carry the Census not-available sentinel, which we set to missing rather than impute, so the income statistics rest on the 761 graded tracts that carry a real median [1]. The vacancy gradient is not strictly monotonic, with B and C tied at 8.6 percent, a tie we declined to smooth into a cleaner staircase [1]. Each tract is filed under a single grade by whether its official Census internal point falls inside an A, B, C, or D polygon, so tracts straddling a zone boundary are counted whole in one grade and partial overlaps are not split [1]. Only 769 of 1,332 Cook County tracts fall inside any 1938 zone, since the survey covered far less ground than today's county, so the whole walk describes the mapped part of Chicago and is silent on the rest [1]. The ACS figures are 2019 to 2023 pooled five-year estimates carrying sampling error, and the margins of error present in the source files are not carried into our aggregates, so the point estimates above should be read as central tendencies rather than exact values [1]. This deserves a moment, because sampling error cuts hardest exactly where the sample is thinnest. A median household income for a single grade-A tract is estimated from a modest number of surveyed households over five years, and it arrives with a margin of error the published table reports and our aggregate drops. For the large grades, built from hundreds of tracts and hundreds of thousands of units, the errors in individual tracts substantially wash out in the aggregate. For grade A, with fifteen tracts, they do not wash out as cleanly, which compounds the small-sample caution already raised. None of this is hidden. The margins sit in the source files we posted, and a reader who wants to propagate them through to the grade-level figures can do so. We chose to report clean point estimates with the caveat stated plainly rather than to carry error bands through aggregations the ACS was not designed to support, but the honest reader should picture every number in the walk with a cushion of uncertainty around it, wider at the top of the gradient than at the bottom. The public-investment counts use the corrected record totals, 598 affordable rental developments and 768 TIF and RDA projects, not the inflated physical line counts careless parsing would produce [1]. None of these cautions sink the descriptive pattern. All of them bound how far it can be pushed.
Turn outward to the scholarly debate and the same restraint holds, since the literature itself does not speak with one voice on cause. Hillier, working at the address level in Philadelphia, argues that lenders were already steering clear of the areas that would be graded red before the HOLC maps existed, so the grades alone do not explain most of the lending differences that followed, and on her reading the maps may have documented a discriminatory market more than they created one [8]. That argument deserves its full weight, not a footnote. Set against it is the boundary and discontinuity evidence of a genuine map effect, the finding that neighborhoods on opposite sides of a grade line diverged afterward in ways consistent with the line itself mattering, and the divergence that holds even for pairs which started all white [2][5][6]. The open question is how much the maps caused and how much they recorded, a real disagreement among careful people rather than a settled point. Our cross-section cannot adjudicate it and does not try. What our numbers add is a present-day description consistent with either reading, since a gradient this clear in 2023 is what you would expect whether the maps made the divergence or faithfully marked one already underway. The claim narrows to that.
One methodological caution deserves to be stated in its own right, because it governs how any reader should generalize from these numbers. Everything here is measured at the level of the tract, not the household, and a pattern that holds across tracts does not automatically hold across the people inside them. This is the ecological inference problem, and it is easy to state and easy to forget. When we report that former D ground has lower median income and higher vacancy, we are describing tracts, areas of a few thousand residents, not making a claim about any individual. A particular household on former D ground may be wealthier than a particular household on former A ground, and nothing in the gradient rules that out. The walk describes the central tendency of places sorted by grade. It does not describe destinies of persons, and reading it as though it did would import a determinism the data cannot support and the history does not warrant. People move, neighborhoods change hands, and a tract is a container, not a fate. The honest claim is about the containers.
There is a further reason for caution that has nothing to do with our method and everything to do with the city. The grades of 1938 were not the last word the federal government, the banks, or the local market said about these neighborhoods. The forces that came after did not arrive at random. The Dan Ryan Expressway, routed in the late 1950s, ran its trench along the long-standing boundary between Black and white neighborhoods on the South Side rather than the straighter path a few blocks east, and the homes it took were largely Black-owned. The Chicago Housing Authority stacked its high-rises in a long wall down State Street, deepening the concentration of poverty on ground the HOLC had already marked down. Blockbusting and contract selling stripped equity from West and South Side families through the 1950s and 1960s, a well-documented predatory machinery in which a family bought a home on installment, paid like an owner while holding no deed, and could be evicted with nothing after a single missed payment, the seller free to start over with the next family. Each of those later interventions tended to land on or near the ground the 1938 map had graded C and D, so the gradient we read in 2023 is layered, the original lines plus everything that was poured into the same channels afterward. That layering is precisely why our snapshot cannot establish cause, and it is also why the snapshot looks the way it does.
What an Old Map Still Asks
Return, at the end, to the document the paper opened on. A field agent walked Chicago in 1938, scored its districts from A to D, and shaded them green, blue, yellow, and red. Walk those grades in order eighty-five years later and the pattern comes out in one breath. Median household income falls from the green through the red, homeownership thins along the same descent, vacancy stands highest on the red ground the surveyor called Hazardous, and two streams of public subsidy pool where the old map drew its worst lines [1]. All of that is description of a present-day snapshot, the shape of Chicago in 2023 laid over the shape someone assigned the city in 1938, and none of it is offered as proof of cause.
The published causal literature did work this paper deliberately did not repeat, and it keeps its own credit. The boundary, propensity-score, and discontinuity studies argue that the maps reached forward into later homeownership, property values, segregation, adult income, upward mobility, and modern credit scores, and the national overlay shows the descriptive pattern is no Chicago peculiarity but a feature of the formerly graded country [2][11][4]. Those findings live in a different lane from ours, reached by methods built to attribute, and the honest move is to credit them for what they show rather than borrow their authority for our cross-section. Our contribution is smaller and, kept in its place, sturdy. In one large American county, with real public data anyone can pull, the 1938 grades still track the contours of the present.
There is a practical reason the descriptive result is worth having even with the causal question set aside. Policy does not run on causal estimates alone. It runs on knowing where the need is, and a descriptive overlay answers that directly. Whatever produced the gradient, the gradient is where Chicago's lower incomes, thinner ownership, higher vacancy, and concentrated public subsidy sit today, and they sit on ground a federal agency marked down in 1938. A city deciding where to direct affordable housing, where to repair the housing stock, where to put a clinic or plant a tree does not need to settle the historians' debate to use the map. The places that were graded worst are, by a wide and consistent margin, the places that carry the deficits now. That is a fact about the present, established without any claim about the past, and it is the kind of fact that the descriptive genre exists to provide.
What stays open is the question the gradient cannot close. A single surveyor's verdict from 1938 still tracks the shape of Chicago in 2023, and whether that verdict made the city or merely recorded a city already being made is something this cross-section cannot answer and the scholarship still debates [8]. The files behind every number here sit at the linked archive, the HOLC polygons, the ACS join, and the point-in-polygon counts, posted to be checked rather than taken on faith [1]. The map asked a question of the ground when it laid down those four colors. The ground, eight decades on, has not finished answering.
Citations
11 sources cited.
Primary Sources
- 1.Author analysis of 1938 Chicago HOLC graded zones (Mapping Inequality) joined to ACS 2023 5-year tract estimates for Cook County. HOLC polygons: University of Richmond Digital Scholarship Lab, Mapping Inequality (already in repo). ACS estimates: U.S. Census Bureau, American Community Survey 5-Year Detailed Tables.. Reproducible files at rooted-forward.org/research/data/holc-redlining-present-day-outcomes-chicago.
- 2.
Secondary Sources
- 3.
- 4.
- 5.
- 6.
- 7.
- 8.
- 9.
- 10.
- 11.