Evaluating gentrification’s relation to neighborhood and city health

Joseph Gibbons, Michael S. Barton, Elizabeth Brault•View original
OverviewTechnicalfable voice
Let's start with the core tension this research is built around. Gentrification is often framed in public discourse as either neighborhood renewal or neighborhood destruction, but what does it actually do to health? And more specifically, does whatever it does at the block level ever add up to something meaningful at the scale of the whole city? That is the precise causal question that Gibbons, Barton, and Brault set out to answer, and the way they decompose it across two analytical scales is what makes the study worth your attention. The background they're working against is well-established. Neighborhood context shapes self-rated health. Poverty, environmental hazards, and social disorder tend to worsen health outcomes. Gentrification could plausibly improve those conditions by bringing new resources and amenities into previously disinvested areas. But displacement and selective migration could offset or reverse any gains. Despite a long theoretical discussion of these competing mechanisms, the authors note there's no conclusive nationwide evidence linking gentrification to neighborhood health outcomes, and essentially no clear story at the city level. So they set out to test three hypotheses. First, that gentrification is negatively associated with poor self-rated physical health at the neighborhood level. Second, that the proportion of gentrifying tracts in a city is not positively associated with city-level poor health rates, meaning no city-wide benefit or penalty. Third, that these patterns hold regardless of city size. That third hypothesis matters more than it might seem at first. Large cities, defined here as those at or above the seventy-fifth percentile of the five hundred city sample, think New York on the high end, and Honolulu on the low end of that threshold, may have gentrification concentrated in a handful of high-profile neighborhoods. So if neighborhood effects are real but geographically contained, you'd expect them to wash out when you zoom to the city level. Testing that explicitly is a meaningful design choice. Now, the data and measurement. The health outcome comes from the centers for disease control and prevention's five hundred Cities project, which provides tract and city-level estimates for the 2014 wave of the Behavioral Risk Factor Surveillance System, or BRFSS. These small-area estimates were generated through a multilevel method linking geocoded county BRFSS data to block-level demographics from the 2010 Census. The centers for disease control and prevention validated these estimates by comparing county-level reconstructions against raw BRFSS county data in Missouri and Massachusetts, and the measures tracked closely. The dependent variable is the percentage of adults eighteen and over reporting fourteen or more days of poor physical health in the past thirty days, a crude rate, not post-stratified by age, race, or income, which the authors flag as a limitation. The tract-level analytic sample covers twenty-six thousand six hundred twenty observations drawn from five hundred cities, with a parallel city-level ordinary least squares analysis running at N equals five hundred. The gentrification measure follows the threshold approach developed by Ding, Hwang, and Divringi, itself influenced by Freeman. Tracts are first classified as gentrifiable, meaning their median household income started below the city median in 1990 or 2000. A gentrifiable tract is coded as gentrifying if it subsequently sees increases in gross rent or median home value above the city median and also gains in college-educated residents above the city median. What distinguishes this study is its commitment to stage. Following Brown-Saracino's argument that stage matters, the authors use four mutually exclusive categories: old gentrification, meaning the tract gentrified in the 1990s but was no longer gentrifiable by 2000; recent gentrification, meaning it gentrified in the 2000s but not the 1990s; continued gentrification, meaning it gentrified in both decades; and never gentrified, meaning gentrifiable throughout but never meeting the criteria. These stages cover the arc from 1990 through 2014 and are the primary neighborhood-level predictors in the models. Covariates at the tract level include racial and ethnic composition, percentage Black, percentage Asian, percentage Hispanic, along with residential stability, homeownership, unemployment, and median age. At the city level, the models add Theil's H for multigroup segregation across White, Black, Hispanic, and Asian populations, the Gini coefficient for economic inequality, region indicators, and the large city indicator. All continuous predictors are grand-mean centered, and tract-level models are estimated as multilevel regressions with tracts nested in cities, proceeding through four models of increasing adjustment. Now to the results, and this is where the two-scale design pays off. At the neighborhood level, the findings are clear and consistent. Across all four progressively adjusted models, every gentrification category shows a statistically significant negative association with poor self-rated health. In Model 1, unadjusted, the magnitudes are large: Continued Gentrification comes in at negative 3.477, Non-Gentrifiable at negative 5.195, Old Gentrification at negative 2.408, and Recent Gentrification at negative 1.268, all relative to the never gentrified reference group. As tract and city controls are added, these coefficients attenuate substantially, which is expected. But they don't disappear. In the fully adjusted Model 4, Continued Gentrification is negative 1.400, Non-Gentrifiable is negative 2.161, Old Gentrification is negative 0.800, and Recent Gentrification is negative 0.680, all significant at a p-value below 0.001. A gradient is visible within the gentrifying categories. Continued Gentrification, sustained gentrification across both the 1990s and 2000s, shows the strongest health advantage among the gentrifying tracts in the full model. Old and Recent Gentrification are smaller in magnitude but still significant. The Non-Gentrifiable category, which encompasses tracts that started above the city median and were never at risk of gentrification, shows the largest health advantage overall, which makes intuitive sense — these are the wealthiest, most stable tracts. The control variables align with prior literature: Black and Hispanic shares predict higher rates of poor health; residential stability and homeownership predict lower rates; unemployment and median tract age push the outcome upward. Theil's H and Gini at the city level both associate with higher poor health in more segregated or unequal cities. Critically, the large city indicator in Model 4 is not significant at the tract level. That supports Hypothesis C — the neighborhood-level health associations with gentrification don't vary by city size. Whether you're looking at a major metro or a mid-sized city, gentrifying tracts show lower rates of poor self-rated health. But here is where the city-level story diverges sharply from the neighborhood story, and this divergence is the paper's central empirical finding. When Gibbons, Barton, and Brault run the city-level ordinary least squares models, asking whether the proportion of gentrifying tracts in a city predicts city-level poor health rates, they find nothing. The four gentrification proportion coefficients are small and statistically non-significant across the board: Proportion Old Gentrifying is 0.123, Proportion Continued Gentrifying is negative 0.030, Proportion Recent Gentrifying is negative 0.080, and Proportion Non-Gentrifiable is negative 0.031. None of them move the needle on citywide health. The city-level model is not a null model overall; it explains about thirty-six percent of the variance in city-level poor health rates. The covariates that do matter at that scale tell a coherent story about structural inequality. Theil's H is positive and significant at 1.083, meaning more racially and ethnically segregated cities have worse health outcomes. Percentage Hispanic is positive at 0.742, percentage Asian is negative at negative 0.412. Residential stability and homeownership are both negative, unemployment is positive. The large city indicator is negative and significant at negative 0.309 — larger cities tend to have lower reported rates of poor health at the city level, even controlling for everything else. But gentrification? It doesn't register. The authors interpret this cleanly: gentrification's health benefits, to the extent they exist, are highly localized. They appear within gentrifying tracts. They do not aggregate upward into city-wide gains or reductions in city-level health inequalities. This is a meaningful constraint on how we should think about gentrification as any kind of public health policy lever. If the effects are real but geographically contained and concentrated in neighborhoods that were already beginning to attract investment, then treating gentrification as a city-wide equity remedy misreads the evidence. There are limitations worth naming directly. The five hundred Cities data are cross-sectional, so causal inference is constrained. Suburban areas outside the five hundred largest cities aren't covered. The health outcome is a crude rate without demographic adjustment. And the gentrification measure captures residential dynamics — housing values and educational attainment — but not commercial change or the more granular displacement dynamics that matter enormously for residents who leave before any health benefits materialize. Still, the study's contributions are substantial: national scope, small-area tract-level estimates, a stage-sensitive gentrification measure spanning twenty-five years, and a multilevel design that explicitly tests whether neighborhood effects translate to city effects. That design is what allows the clean answer — yes at the tract level, no at the city level — and why the divergence is credible rather than just a null result. The neighborhood and city stories aren't contradictory. They're telling you that place-level health advantages in gentrifying areas are real and durable after controls, but geographically bounded in ways that prevent them from reshaping the health landscape of the cities they sit within.

Let's start with the core tension this research is built around. Gentrification is often framed in public discourse as either neighborhood renewal or neighborhood destruction, but what does it actually do to health? And more specifically, does whatever it does at the block level ever add up to something meaningful at the scale of the whole city?

That is the precise causal question that Gibbons, Barton, and Brault set out to answer, and the way they decompose it across two analytical scales is what makes the study worth your attention.

The background they're working against is well-established. Neighborhood context shapes self-rated health. Poverty, environmental hazards, and social disorder tend to worsen health outcomes.

Gentrification could plausibly improve those conditions by bringing new resources and amenities into previously disinvested areas. But displacement and selective migration could offset or reverse any gains. Despite a long theoretical discussion of these competing mechanisms, the authors note there's no conclusive nationwide evidence linking gentrification to neighborhood health outcomes, and essentially no clear story at the city level.

So they set out to test three hypotheses. First, that gentrification is negatively associated with poor self-rated physical health at the neighborhood level. Second, that the proportion of gentrifying tracts in a city is not positively associated with city-level poor health rates, meaning no city-wide benefit or penalty. Third, that these patterns hold regardless of city size.

That third hypothesis matters more than it might seem at first. Large cities, defined here as those at or above the seventy-fifth percentile of the five hundred city sample, think New York on the high end, and Honolulu on the low end of that threshold, may have gentrification concentrated in a handful of high-profile neighborhoods. So if neighborhood effects are real but geographically contained, you'd expect them to wash out when you zoom to the city level. Testing that explicitly is a meaningful design choice.

Now, the data and measurement. The health outcome comes from the centers for disease control and prevention's five hundred Cities project, which provides tract and city-level estimates for the 2014 wave of the Behavioral Risk Factor Surveillance System, or BRFSS. These small-area estimates were generated through a multilevel method linking geocoded county BRFSS data to block-level demographics from the 2010 Census.

The centers for disease control and prevention validated these estimates by comparing county-level reconstructions against raw BRFSS county data in Missouri and Massachusetts, and the measures tracked closely. The dependent variable is the percentage of adults eighteen and over reporting fourteen or more days of poor physical health in the past thirty days, a crude rate, not post-stratified by age, race, or income, which the authors flag as a limitation. The tract-level analytic sample covers twenty-six thousand six hundred twenty observations drawn from five hundred cities, with a parallel city-level ordinary least squares analysis running at N equals five hundred.

The gentrification measure follows the threshold approach developed by Ding, Hwang, and Divringi, itself influenced by Freeman. Tracts are first classified as gentrifiable, meaning their median household income started below the city median in 1990 or 2000. A gentrifiable tract is coded as gentrifying if it subsequently sees increases in gross rent or median home value above the city median and also gains in college-educated residents above the city median.

What distinguishes this study is its commitment to stage. Following Brown-Saracino's argument that stage matters, the authors use four mutually exclusive categories: old gentrification, meaning the tract gentrified in the 1990s but was no longer gentrifiable by 2000; recent gentrification, meaning it gentrified in the 2000s but not the 1990s; continued gentrification, meaning it gentrified in both decades; and never gentrified, meaning gentrifiable throughout but never meeting the criteria. These stages cover the arc from 1990 through 2014 and are the primary neighborhood-level predictors in the models.

Covariates at the tract level include racial and ethnic composition, percentage Black, percentage Asian, percentage Hispanic, along with residential stability, homeownership, unemployment, and median age. At the city level, the models add Theil's H for multigroup segregation across White, Black, Hispanic, and Asian populations, the Gini coefficient for economic inequality, region indicators, and the large city indicator. All continuous predictors are grand-mean centered, and tract-level models are estimated as multilevel regressions with tracts nested in cities, proceeding through four models of increasing adjustment.

Now to the results, and this is where the two-scale design pays off.

At the neighborhood level, the findings are clear and consistent. Across all four progressively adjusted models, every gentrification category shows a statistically significant negative association with poor self-rated health. In Model 1, unadjusted, the magnitudes are large: Continued Gentrification comes in at negative 3.477, Non-Gentrifiable at negative 5.195, Old Gentrification at negative 2.408, and Recent Gentrification at negative 1.268, all relative to the never gentrified reference group.

As tract and city controls are added, these coefficients attenuate substantially, which is expected. But they don't disappear. In the fully adjusted Model 4, Continued Gentrification is negative 1.400, Non-Gentrifiable is negative 2.161, Old Gentrification is negative 0.800, and Recent Gentrification is negative 0.680, all significant at a p-value below 0.001.

A gradient is visible within the gentrifying categories. Continued Gentrification, sustained gentrification across both the 1990s and 2000s, shows the strongest health advantage among the gentrifying tracts in the full model. Old and Recent Gentrification are smaller in magnitude but still significant.

The Non-Gentrifiable category, which encompasses tracts that started above the city median and were never at risk of gentrification, shows the largest health advantage overall, which makes intuitive sense — these are the wealthiest, most stable tracts. The control variables align with prior literature: Black and Hispanic shares predict higher rates of poor health; residential stability and homeownership predict lower rates; unemployment and median tract age push the outcome upward. Theil's H and Gini at the city level both associate with higher poor health in more segregated or unequal cities.

Critically, the large city indicator in Model 4 is not significant at the tract level. That supports Hypothesis C — the neighborhood-level health associations with gentrification don't vary by city size. Whether you're looking at a major metro or a mid-sized city, gentrifying tracts show lower rates of poor self-rated health.

But here is where the city-level story diverges sharply from the neighborhood story, and this divergence is the paper's central empirical finding. When Gibbons, Barton, and Brault run the city-level ordinary least squares models, asking whether the proportion of gentrifying tracts in a city predicts city-level poor health rates, they find nothing. The four gentrification proportion coefficients are small and statistically non-significant across the board: Proportion Old Gentrifying is 0.123, Proportion Continued Gentrifying is negative 0.030, Proportion Recent Gentrifying is negative 0.080, and Proportion Non-Gentrifiable is negative 0.031. None of them move the needle on citywide health.

The city-level model is not a null model overall; it explains about thirty-six percent of the variance in city-level poor health rates. The covariates that do matter at that scale tell a coherent story about structural inequality. Theil's H is positive and significant at 1.083, meaning more racially and ethnically segregated cities have worse health outcomes.

Percentage Hispanic is positive at 0.742, percentage Asian is negative at negative 0.412. Residential stability and homeownership are both negative, unemployment is positive. The large city indicator is negative and significant at negative 0.309 — larger cities tend to have lower reported rates of poor health at the city level, even controlling for everything else. But gentrification? It doesn't register.

The authors interpret this cleanly: gentrification's health benefits, to the extent they exist, are highly localized. They appear within gentrifying tracts. They do not aggregate upward into city-wide gains or reductions in city-level health inequalities.

This is a meaningful constraint on how we should think about gentrification as any kind of public health policy lever. If the effects are real but geographically contained and concentrated in neighborhoods that were already beginning to attract investment, then treating gentrification as a city-wide equity remedy misreads the evidence.

There are limitations worth naming directly. The five hundred Cities data are cross-sectional, so causal inference is constrained. Suburban areas outside the five hundred largest cities aren't covered.

The health outcome is a crude rate without demographic adjustment. And the gentrification measure captures residential dynamics — housing values and educational attainment — but not commercial change or the more granular displacement dynamics that matter enormously for residents who leave before any health benefits materialize.

Still, the study's contributions are substantial: national scope, small-area tract-level estimates, a stage-sensitive gentrification measure spanning twenty-five years, and a multilevel design that explicitly tests whether neighborhood effects translate to city effects. That design is what allows the clean answer — yes at the tract level, no at the city level — and why the divergence is credible rather than just a null result. The neighborhood and city stories aren't contradictory.

They're telling you that place-level health advantages in gentrifying areas are real and durable after controls, but geographically bounded in ways that prevent them from reshaping the health landscape of the cities they sit within.