Why do the poor live in cities? The role of public transportation

Here is the central empirical fact. In the year 2000, nearly 20 percent of people living in American central cities were in poverty. In the suburbs of those same metropolitan areas, the rate was 7.5 percent. That gap — roughly 12 percentage points — is not a small rounding error. It is a defining feature of American urban geography. Edward Glaeser, Matthew Kahn, and Jordan Rappaport wanted to know why. The obvious first guess is that poor people end up downtown because that's where the cheap land is. Cities are dense, and land near the center is expensive per square foot. But if the poor demand much less space, they might naturally cluster where density is highest. A second explanation points to amenities. Maybe the poor need urban services, social networks, or informal economies that suburbs don't offer. The third explanation, and the one Glaeser and colleagues end up defending, is public transportation. The poor don't own cars. Transit is concentrated near city centers, so the poor sort to where transit is available. The data pattern that motivates this whole inquiry is the income-distance gradient. Glaeser and colleagues use census-tract level data to measure how median household income changes as you move outward from the Central Business District — the CBD, the commercial core. In pooled regressions, income rises with distance from the center. The coefficient within three miles of the CBD is about 0.12 per mile in log terms, meaning richer neighborhoods tend to sit farther out. But here's where it gets interesting. Split cities by age, and the pattern comes apart. In old, transit-built cities like Boston, Chicago, New York, and Philadelphia, income actually falls as you move outward from the center for the first three miles, then rises. In newer Sun Belt cities like Atlanta, Houston, Los Angeles, and Phoenix, income rises monotonically with distance. The shape of poverty in space is not universal. It depends on how the city was built, and more specifically, on how people get around. Commuting data from the National Household Transportation Survey makes the stakes of transport mode concrete. A car commute costs about 1.6 minutes per mile once you're moving, with a fixed startup time of around 5.6 minutes. A bus commute costs roughly 2.95 minutes per mile with a fixed cost of 22 minutes. A subway is slightly faster per mile than a bus but still carries an 18-minute fixed cost just to get on. Transit commutes are not just slower — they're punishing in fixed time. And that fact shapes everything about where you'd choose to live if transit were your only option. The standard economic framework for thinking about this is the Alonso-Muth-Mills monocentric city model. The logic is simple: people trade off the cost of housing against the cost of commuting, and equilibrium sorts them by income. In the classic version, where everyone drives, the poor centralize only if they demand substantially less land than the rich — specifically, if the income elasticity of land demand exceeds one. An income elasticity of one means a one percent rise in income leads to a one percent rise in land consumed. Above one, the rich spread out enough that the center fills with poor. Below one, the model can't generate the observed pattern. Glaeser and colleagues estimate that elasticity, and the answer is damning for the classic story. Using household-level data from the American Housing Survey, the ordinary least squares estimate for single-family homeowners is just 0.08. Instrumenting income with years of education to correct for measurement error pushes it to 0.26. For apartment dwellers, estimates run from 0.34 to 0.55. The authors treat 0.25 as a plausible benchmark, with 0.5 as an upper bound. Even the most generous reading falls well short of one. Land demand differences alone cannot explain why the poor concentrate downtown. So the authors extend the standard model. Instead of one transport mode, they allow two: private cars and public transit. Cars cost money but save time. Transit saves money but costs time. Because the poor have lower wages, their time is worth less, which gives them a comparative advantage in the time-intensive but cash-cheap option: transit. And if transit is primarily available near the city center, the poor's optimal location pulls toward the center in a way that has nothing to do with land demand. The calibration sharpens this into a concrete condition: for the poor to concentrate centrally, the ratio of rich income to poor income must exceed 4.22. Under typical U.S. income distributions, that condition is easily met. Their calibration suggests transport modes account for roughly three-quarters of the observed centralization of poverty. Land demand elasticity accounts for about one quarter. That's a striking inversion of the usual story. Now, the empirical tests. In cross-sectional tract regressions across sixteen cities, controlling for distance to the CBD, the share of commuters using public transit is a powerful predictor of low tract income. The coefficient is nearly negative two. A ten point increase in transit share is associated with roughly 19 percent lower median income. When they include transit usage in the income-distance regressions, it accounts for about two-thirds of the positive relationship between distance and income within ten miles of the center. In instrumental variable specifications, where they instrument transit usage with distance to rail lines, access to public transportation explains essentially all of the income-distance relationship. The gradient largely disappears once transit is accounted for. The New York City subway analysis carries particular weight. Subway stop locations in the sampled boroughs haven't changed since 1942, which provides something close to exogenous variation. The transit coefficient in New York City is negative 1.2, and the patterns are consistent with the cross-city results. In panel data covering transit expansions between 1980 and 2000, tracts within one mile of rail transit have poverty rates about 4 percentage points higher than comparable tracts farther from rail. The treatment effect from gaining new rail access is smaller — statistically significant but modest — suggesting transit concentrates the poor rather than impoverishing them. The contrast between car-zone metros and subway cities is the empirical smoking gun. Glaeser and colleagues define car-zone metros as places where even the most transit-heavy tract between five and fifteen miles from the CBD has a transit commute share below 2.5 percent. There are 99 such metros. In those cities, the income-distance relationship is flat or reversed. The rich actually live closer to the center, consistent with a world where everyone drives, and the monocentric model operates cleanly. In the four old subway cities, the pattern flips. Poverty concentrates at the center, exactly where the transit network is densest. The correlation between log income and transit use within three miles of the CBD is positive 0.26 in old cities and negative 0.51 in new ones. Those numbers point in opposite directions because the underlying transport infrastructure points in opposite directions. Historical evidence reinforces the story. LeRoy and Sonstelie found that in 1907, 52 percent of workers earning less than ten dollars a week walked to work, while only 12 percent of workers earning twenty dollars a day did. Transport mode has always sorted income in space. As cars became affordable and highways expanded after World War II, the non-poor gained a new option — they could live far from downtown without punishing commutes. The poor, who couldn't afford that option, stayed where transit worked. In newer cities built around the car, this dynamic never took hold the same way: 81 percent of metro employment in new cities sits more than five miles from the CBD, compared with 55 percent in old cities. The suburbanization of jobs followed the suburbanization of the affluent. Paris stands as the notable international exception. Haussmann's nineteenth-century urban clearance pushed the poor to the periphery. Massive suburban public housing reinforced that pattern, and heavily subsidized transit made the periphery livable. The result is a city where the wealthy live centrally and the poor live in the banlieues — the opposite of the American pattern. The same transit-and-sorting logic applies, but different historical and policy choices produced a mirror image. The takeaway is this: the geography of poverty in American cities is not the residue of random history or urban decay. It is a rational spatial equilibrium. The poor have sorted into central cities because that's where the infrastructure that substitutes for car ownership is located. This means that cutting transit — often framed as a cost-saving measure that doesn't hurt growth — hits the people who chose their homes specifically to access it. Glaeser, Kahn, and Rappaport are careful to note that housing markets, urban amenities, and the social consequences of poverty also shape where people end up. Transit doesn't explain everything. But it explains more than the land market alone, more than amenity preferences, and more than most urban economists had previously accounted for. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.