Spatial Disparities in Housing Values in the United States During the Great Depression: A Place-Based Sustainability Perspective

Abstract

Spatial disparities in housing values during the Great Depression reflect not only regional housing market conditions but also deeper inequalities in economic opportunity, social infrastructure, and environmental resilience that are central to place-based sustainability. Despite extensive research on housing inequality during this period, spatial disparities in housing values—particularly in relation to race beyond the neighborhood level—remain underexplored. This study examines county-level spatial disparities in housing values in the United States between 1930 and 1940, framing housing values as an indicator of place-based sustainability. Using spatial visualization, global and local spatial econometric models, and Multi-Scale Geographically Weighted Regression (MGWR), we analyze how economic shocks, environmental stressors, and socioeconomic and demographic factors jointly shaped uneven housing outcomes across space. Our findings reveal distinct regional trends: higher housing values were concentrated in the Northeast, Midwest, and West Coast, while lower values prevailed in the Mountain and Southern regions. Housing values declined from 1930 to 1940, with the Dust Bowl intensifying losses in affected areas. Socioeconomic factors, such as higher illiteracy and unemployment rates, were associated with lower housing values, whereas higher retail sales per capita, a proxy for income, were linked to higher values. Housing values also varied significantly by racial and nativity composition, with persistent disparities disadvantaging Black and other minority populations relative to native White populations within the same regions. By quantifying spatial inequality and identifying uneven regional vulnerability and resilience during a major historical crisis, this study contributes a place-based sustainability perspective on long-term housing inequality and its structural roots.

1. Introduction

The spatial distribution of housing values provides critical insight into the sustainability of places, as housing values reflect long-term access to economic opportunity, public services, environmental quality, and wealth-building potential. Persistent spatial disparities in housing values signal uneven development trajectories and unequal resilience to economic and environmental shocks—core concerns of place-based sustainability. Understanding how such disparities emerge and evolve during periods of crisis is therefore essential for advancing sustainability research focused on equity, resilience, and long-term regional development. The complex interplay between population dynamics, economic conditions, and the housing market is critical to understanding and promoting sustainable development while addressing entrenched social and economic disparities. Spatial variations in median housing values reflect more than just differences in local real estate markets—they signify unequal access to the social, economic, and environmental benefits tied to place. As scholars have noted [1,2], housing value is closely associated with access to quality education, public services, neighborhood safety, wealth accumulation, and long-term life opportunities. When racial or nativity groups are disproportionately concentrated in areas with persistently lower housing values, this pattern reflects not only residential sorting but also deeper, historically rooted structural inequalities. In this context, spatial patterns of housing value serve as a meaningful proxy for broader place-based inequities and sustainability. By examining the relationships among housing values, demographic composition, and economic factors during the 1930s in the United States, this study seeks to uncover the patterns and drivers of spatial disparities in housing values during the Great Depression. We conceptualize housing values as an indicator of place-based sustainability and investigate how economic shocks, environmental stressors, and demographic composition jointly shaped uneven housing outcomes. Using GIS-based visualization, global and local spatial econometric models, and multi-scale geographically weighted regression, we analyze both national patterns and localized variations in the determinants of housing values. By doing so, this research contributes to sustainability scholarship by quantifying spatial inequality, identifying uneven regional resilience, and providing historical context for contemporary challenges in sustainable and equitable housing development.

During the 1920s, the U.S. experienced a surge in building construction and housing prices, leading to a boom in the housing market. However, this prosperity was short-lived, as the housing market became one of the most severely affected sectors during the Great Depression [3]. The economic downturn that persisted throughout the 1930s led to high unemployment rates, diminished incomes, and financial instability. Consequently, many Americans faced increased issues with housing affordability and other fundamental necessities. This instability led to decreased demand for housing, causing real estate prices to fall [4]. The decline in property values made selling or refinancing homes increasingly difficult for Americans, further exacerbating the housing crisis and disparity.

During this period, housing opportunities and resources were distributed unequally across different geographic areas due to issues of race, class, discriminatory practices, and economic factors [5]. High unemployment rates, wage reductions, and a decline in housing construction led to housing shortages and exacerbated existing housing inequalities, disproportionately affecting low-income and marginalized communities. Discriminatory practices such as racial covenants, which were clauses in property deeds that prohibited the sale or rental of homes to individuals of specific racial or ethnic backgrounds, were widely used in the U.S. during this period. These practices reinforced segregated communities and contributed to housing disparities. Furthermore, different regions experienced varying degrees of economic decline. Rapid urbanization, industrialization, and the influx of immigrants created significant spatial disparities in housing availability, quality, and affordability.

Many studies have documented the severity of the housing crisis during the Great Depression, highlighting issues such as overcrowding, poor living conditions, and homelessness [6,7,8,9,10,11,12,13,14]. Research also underscores the role of government programs and policies—including the Home Owners’ Loan Corporation, the Federal Housing Administration, redlining, racial segregation, and New Deal initiatives—in perpetuating housing insecurity during this period [15,16,17,18,19,20,21,22,23,24,25,26,27]. Additionally, several books have explored housing inequality during this era [28,29,30]. Discriminatory practices such as redlining and restrictive covenants limited minorities’ access to credit and confined them to segregated neighborhoods [31]. Moreover, unemployment rates surged across all demographic groups, but minority populations were often the first to lose their jobs and the last to regain employment [32]. This economic instability made it even more difficult for them to afford or maintain housing, exacerbating pre-existing disparities.

While there has been extensive research on the issues of the housing crisis and inequality during this period, only a few studies explored spatial disparity issues of housing values during this period [33,34]. Most existing research focused on the overall economic impact of the Great Depression on housing markets, examining foreclosure rates, mortgage defaults, and the general decline in housing construction and home ownership. Although these studies have provided valuable insights into the widespread nature of the housing crisis and its devastating effects on American families, few studies delved into the geographic distribution of these housing issues across the country, specifically how housing values varied across different counties and regions. In addition, while a growing body of research has examined the impact of race on housing disparity during the 1930s [21,31,35,36,37,38,39,40,41,42], there remains a critical gap in the literature regarding the intersection of race and space beyond the neighborhood scale. Most existing studies focus narrowly on redlining and localized patterns of segregation, yet racialized housing disparity was also shaped by broader spatial dynamics operating at the county or regional level. Discriminatory policies and investment patterns—often driven by federal and interjurisdictional decisions—systematically excluded minority communities from housing credit and infrastructure across entire counties and metropolitan regions. These practices produced stark regional disparities in property values, homeownership, and long-term economic opportunity. Without explicit attention to these macro-spatial patterns, our understanding of how racial composition influenced structural disparity across geographies remains incomplete. Studying race and space at higher spatial scales is thus essential to uncovering the full extent of racialized housing exclusion and its enduring regional consequences.

To address these research gaps in the literature, we use spatial visualization and analysis methods to uncover spatial disparity patterns of housing values during the 1930s and examine the interplay between housing values and racial/nativity composition in a spatial context. We aim to address three research questions: First, how did housing values vary across different counties during the 1930s? Second, how were the Depression-era, Dust Bowl, education, unemployment, and income factors associated with this spatial inequality in housing values? Third, how did racial and nativity group differences in housing values manifest across counties and within shared geographic contexts, such as within the same county or region? By doing so, we may gain a more comprehensive understanding of the regional patterns of housing disparity during the Great Depression.

2. Data

Data on the housing market and control variables—including housing, population, education, race, employment, agricultural, and economic factors—were sourced from the 1930 and 1940 census data available on the IPUMS National Historical Geographic Information System (NHGIS) website [43]. The control variables were selected based on a comprehensive review of theoretical and empirical literature on the housing market [44,45,46,47,48,49,50], along with data availability.

Table A1. Summary statistics of dependent and independent variables.

Variable	Obs	Mean	Std. Dev.	Min	Max	Category	Variable Description
MdnValHous	6200	2044.2	1297.512	0	20,000	Dependent Variable	Median value of owner-occupied dwelling units
pctWhite	6200	87.33	18.76	0	100	Race	Percentage of White population
pctBlack	6200	10.94	18.08	0	85.83	Race	Percentage of Black population
pctOther	6200	1.67	6.99	0	86.02	Race	Percentage of ‘Other’ population
pctNatW	6200	83.24	19.46	0	696.76	Nativity	Percentage of native-born White population
pctForW	6200	1.89	3.93	0	90.63	Nativity	Percentage of foreign-born White population
After	6200	0.50	0.50	0	1	Year	1 for 1940; 0 for 1930
DustBowl	6200	0.22	0.41	0	1	DustBowl counties	1 for DustBowl counties; 0 for other counties
After*DustBowl	6200	0.11	0.31	0	1	Year*DustBowl	After*Treated_DustBowl
TotlDwell	6200	10,085	33,826	0	1,170,557	Housing	Total dwelling units
pctOwNofarmDwel	6200	20.44	11.16	0	79.78	Housing	Percentage of owner-occupied nonfarm dwelling units
pctWhiteDwel	6200	87.65	18.44	0	100	Housing	Percentage of dwelling units by White occupant
pctBlackDwel	6200	10.90	17.93	0	87.35	Housing	Percentage of dwelling units by Black occupant
pctRadio	6200	47.53	28.79	0	100	Housing	Percentage of occupied dwelling units with radio
TotalPop	6200	41,036.72	139,534	0	4,063,342	Population	Total population
pctUrban	6200	22.28	25.62	0	100	Population	Percentage of urban population
pctUrban25k	6200	5.98	19.34	0	100	Population	Percentage of population in cities of 25,000 and over
PopDens	6200	190.81	1961.06	0	85,906	Population	Population per square mile
pctIllit	6200	4.68	5.27	0	58.52	Education	Percentage of illiterate population
pctUnemp	6200	6.46	4.52	0	34.4	Employment	Percentage of unemployed population
pctLaborForce	6200	38.65	4.74	0	78.00	Employment	Percentage of population in civilian labor force
SalRetailPerCap	6200	249.15	134.88	0	2112.98	Economy	Net sales of retail establishments per capita
SalWholPerCap	6200	88,979.77	246,212.10	0	6,048,932	Economy	Net sales of wholesale establishments per capita
ManuValPerCap	6200	244.51	430.39	0	10,265.46	Economy	Value of manufactured products plus value added by manufacture per capita
pctEmpRetail	6200	6.76	3.41	0	79.04	Economy	Percentage of employees in retail business
pctEmpWhole	6200	1.20	1.59	0	31.85	Economy	Percentage of employees in wholesale business
ValueCropPerCap	6200	111.41	109.81	0	1708.64	Agriculture	Value of crops per capita

(see Appendix A) summarizes the statistics for both dependent and independent variables.

County boundary GIS (Geographic Information System) data for 1930 and 1940 were also obtained from the NHGIS website. In 1930, the contiguous U.S. comprised 3102 counties, which slightly decreased to 3100 counties by 1940 due to the merger of Campbell and Milton counties into Fulton County, Georgia. All maps in this study used the 1940 GIS boundaries, reflecting the updated configuration of Fulton County.

The Dust Bowl data are based on wind erosion maps archived by the National Archives. The Dust Bowl of the 1930s ranks among the most devastating environmental disasters in twentieth-century North American history, with its epicenter in the Southern Plains of the United States. Its convergence with the Great Depression marked a critical turning point that deeply disrupted agricultural systems, displaced populations, and reshaped housing markets across the region [51,52]. While public narratives often focus on the dramatic dust storms in specific Southern Plains states, the broader ecological crisis—characterized by severe drought, land degradation, and heat-induced agricultural stress—extended across the entire Great Plains, as noted by Gregory [53]. Academic studies of the Dust Bowl frequently rely on wind erosion maps archived by the National Archives, which delineate 843 affected counties spanning 12 Great Plains states [54,55]. For this study, we also adapted these historical Dust Bowl county designations for use in our regression analysis. Figure 1 presents the mapped distribution of these Dust Bowl counties. The core years of Dust Bowl impact—1935, 1938, and 1940 in Figure 1—are based on historical data compiled from ESRI ArcGIS Pro 3.2.0 Online sources [56,57]. Please note that this study emphasizes spatial patterns and heterogeneity rather than causal identification of the Dust Bowl’s effects.

For the regression analysis, the 1930 statistics of the merged Fulton County were used to construct the panel data. To address skewed data and make the distributions more symmetric and suitable for regression analysis, which assumes normality, natural logarithms of dependent variables (median housing value) and certain control variables (for example, total number of housing units, population density) were employed in the regression models instead of their original values, with the exception of those representing percentages.

Due to missing data for certain control variables, observations with incomplete information were excluded from the regression analysis. We did not impute or interpolate the missing data in this study because spatial imputation or interpolation methods could introduce bias or artificial precision into the estimates, potentially misleading the results of the spatial models.

For clarity and to avoid confusion, we use the U.S. Census Bureau–designated regions and divisions for defining interstate regions (https://www2.census.gov/geo/pdfs/maps-data/maps/reference/us_regdiv.pdf, accessed on 30 July 2025). The U.S. Census Bureau defines four statistical regions, each comprising multiple divisions. These are the Northeast (New England and Middle Atlantic), Midwest (East North Central and West North Central), South (South Atlantic, East South Central, and West South Central), and West (Mountain and Pacific). This region definition is the most commonly used classification system. Since our study only covers the contiguous U.S., the Pacific division actually refers to the West Coast. Additionally, on a map of the contiguous U.S., the Midwest region (East North Central and West North Central) is equivalent to the North Central region; the East South Central and West South Central divisions together can be referred to as South Central; and the East South Central and South Atlantic divisions together can be referred to as Southeast.

3. Methods

3.1. GIS Visualization Methods

The natural breaks (Jenks) classification method in ArcGIS Pro 3.2.0 is used to visualize the geographic patterns of housing values. This method creates class breaks that group similar values together and maximize the differences between classes [58]. It divides features into classes whose boundaries are set where there are relatively large differences in the data values. This widely used approach for determining class intervals in choropleth mapping identifies inherent patterns in the data and is particularly useful when dealing with skewed data. The resulting map produced using this method is generally easier to interpret because it emphasizes the most significant differences in the data. Additionally, because this method is purely data-driven and uses an iterative process to find optimal breakpoints, it ensures a more objective representation of the data.

To improve data visualization and identify significant spatial patterns of housing values, Hot Spot Analysis is conducted using the Getis-Ord’s $G_i^{*}$ statistic. The Hot Spot Analysis can reveal patterns that may not be apparent through simple visualization [47]. By identifying areas with unusually high or low values, we may better understand the spatial distribution of housing disparity in the U.S. during the Great Depression and identify significant trends. The Getis-Ord’s $G_i^{*}$ statistic is calculated by:

$$G_i^{*}={\displaystyle \frac{{\sum }_{j=1}^n{w}_{i,j}{x}_j-\overline{X}{\sum }_{j=1}^n{w}_{i,j}}{S\sqrt{\frac{[n{\sum }_{j=1}^n{w}_{i,j}^2-({\sum }_{j=1}^n{w}_{i,j}{)}^2]}{n-1}}}}$$

(1)

where $x_j$ is the housing value for spatial feature (county) $j,w_{i,j}$ is the spatial weight between spatial features (counties) $i$ and $j$, $n$ is equal to the total number of spatial features, and:

$$\overline{X}={\displaystyle \frac{{\sum }_{j=1}^n{x}_j}{n}}$$

(2)

$$S=\sqrt{{\displaystyle \frac{{\sum }_{j=1}^n{x}_j^2}{n}}-(\overline{X}{)}^2}$$

(3)

The $G_i^{*}$ statistic provided for each spatial feature (such as a county) in housing value data is a z score, eliminating the need for further calculations. A statistically significant positive z score indicates that higher values are clustered together (hot spots), with larger scores showing more intense clustering. Conversely, a statistically significant negative z score points to clustering of lower values (cold spots), with smaller scores indicating more intense clustering.

Bivariate color mapping, an advanced cartographic technique, is used to visualize and analyze the relationship between median housing value and racial/nativity composition in a spatial context. This mapping technique allows for a more complex and detailed interpretation of spatial data by showing how two variables interact. It can reveal patterns and correlations that might not be evident when visualizing each variable separately. It can also identify areas where there is a strong relationship between the two variables. For example, it can highlight regions where high values of one variable correspond with low values of another, providing insights into potential causal relationships or areas of concern. The quantile classification method is employed for bivariate color mapping. The quantile method categorizes data into classes that contain an equal number of features, ensuring that each class has the same number of data points. This method is particularly useful for comparing and visualizing two related variables.

Although bivariate color mapping is a powerful tool for visually identifying areas where, for example, high shares of Black or foreign-born residents coincide with particularly low or high housing values, such maps alone do not indicate whether these patterns are statistically significant or could occur by chance. To formally test these spatial associations, we supplement the visualizations with a bivariate Local Indicator of Spatial Association (LISA) analysis. Unlike univariate LISA, which assesses spatial clustering of a single variable, bivariate LISA evaluates whether the value of one variable at a given location (e.g., median housing value) is significantly associated with the spatial lag of a second variable (e.g., the change in foreign-born population share in neighboring counties). We compute the bivariate Moran’s I statistic for each county and use Monte Carlo permutation tests (999 replications) to derive p-values. Statistically significant clusters—categorized as “High–High,” “Low–Low,” “High–Low,” or “Low–High”—are then overlaid on the choropleth map using a 5% significance threshold. This approach allows us to reinforce the visual patterns identified in the color maps with rigorous spatial inference, ensuring that the observed relationships reflect meaningful geographic processes rather than random variation.

The bivariate local Moran’s $I_i^{XY}$ statistic is calculated as:

$$I_i^{XY}=(x_i-\overline{x}){\sum }_{j=1}^n{w}_{ij}(y_j-\overline{y})$$

(4)

where $x_i$ is the value of variable X (e.g., housing value in this study) at location i; $y_j$ is the value of variable Y (e.g., change in racial/nativity composition in this study) at neighboring location j; $\overline{x}$ and $\overline{y}$ are the means of variables X and Y, respectively; $w_{i,j}$ is the spatial weight between spatial features (counties) $i$ and $j$; n is equal to the total number of spatial features.

3.2. Global Regression Analysis

To ensure a robust examination of the drivers of spatial disparity in median housing values, we apply four complementary global regression frameworks. First, Ordinary Least Squares (OLS) serves as our baseline: under the assumption of independent, identically distributed errors, it provides straightforward, unbiased estimates of average relationships across all counties and both years. However, OLS cannot accommodate the fact that nearby counties often influence one another. To address this, the Spatial Lag Model (SLM) explicitly incorporates a spatially lagged dependent variable—that is, it tests whether high (or low) housing values in one county are systematically echoed in its neighbors. In contrast, the Spatial Error Model (SEM) allows spatial correlation in the error term, adjusting inference when unmeasured factors (for example, regional policies or environmental shocks) cluster geographically. Finally, our Fixed-Effects specification includes county fixed effects, which control for all unobserved, time-invariant characteristics specific to each county—capturing, for instance, enduring geographic or institutional features that might bias cross-sectional comparisons.

By comparing OLS, SLM, SEM, and Fixed-Effects models, we benchmark our results under classical assumptions and assess the roles of spatial spillovers and unobserved, time-invariant regional characteristics. Consistent patterns across these models bolster confidence that our findings reflect genuine economic and environmental forces shaping housing disparity, rather than artifacts of model choice or spatial dependence.

Under classical assumptions, OLS provides the best linear unbiased estimates. It is straightforward to implement and interpret. The mathematical formula of OLS is as follows:

$$Y_{it}={\beta }_0+\beta X_{it}+\gamma Z_{it}+{ϵ}_{it}$$

(5)

where $Y_{it}$ the natural logarithm of dependent variable (median housing values) for county i at time t; $X_{it}$ is a vector of interested variables for county i at time t, including the housing, demographic, social, and economic variables listed in Table A1; $Z_{it}$ is a vector of control variables for county i at time t, including the housing, demography, social, and economic variables listed in Table A1; ${\beta }_0$ is the intercept; $\beta$ is the vector of coefficients for the interested variables; $\gamma$ represents the vector of coefficients for the control variables; ${ϵ}_{it}$ is the error term.

The SLM considers the influence of neighboring observations, which is crucial if spatial autocorrelation is present. By modeling spatial dependence, SLM may provide more efficient estimates. The mathematical formula of SLM is as follows:

$$Y_{it}={\rho W{Y}_{it}+\beta }_0+\beta X_{it}+\gamma Z_{it}+{ϵ}_{it}$$

(6)

where W is the spatial weights matrix, defining the spatial relationship (neighborhood structure) between observations; the term $\rho W{Y}_{it}$ captures the spatially lagged dependent variable, reflecting the idea that the value of Y in one location is influenced by the values of Y in neighboring locations. Other variables are defined the same way as in Equation (5).

The SEM accounts for spatial dependence in the error terms, which can lead to more accurate estimates of standard errors and test statistics. The SEM helps in obtaining unbiased estimates even in the presence of spatially autocorrelated errors. The mathematical formula of SEM is as follows:

$$Y_{it}={\beta }_0+\beta X_{it}+\gamma Z_{it}+\lambda W{ϵ}_{it}+{\nu }_{it}$$

(7)

where $\lambda$ is the vector of the coefficients for the spatially autocorrelated error term, capturing the extent to which the error in one observation is related to the errors in neighboring observations; $W$ is the spatial weights matrix; ${ϵ}_{it}$ is the error term from the regression model, and $\lambda W{ϵ}_{it}$ represents the spatially autocorrelated error component; ${\nu }_{it}$ is the independent and identically distributed (i.i.d.) error term. In the SEM, the term $\lambda W{ϵ}_{it}$ models the spatial dependence in the error terms, indicating that the error associated with one observation may be similar to the error associated with its spatial neighbors. Other variables are defined the same way as in Equation (5).

The Fixed-Effects model controls for unobserved, time-invariant characteristics of each county. It is useful when the primary interest is in understanding changes within counties over time. The mathematical formula of Fixed-Effects model is as follows:

$$Y_{it}={\beta }_0+\beta X_{it}+\gamma Z_{it}+{\sigma }_i+{ϵ}_{it}$$

(8)

where ${\sigma }_i$ represents the fixed-effects for each county, capturing unobserved county-specific characteristics that are constant over time. Other variables are defined the same way as in Equation (5). The Fixed-Effects model controls for all county-specific, time-invariant unobserved factors—such as geography, historical land use patterns, and long-standing institutional characteristics—that could bias the estimated effects of time-varying covariates. It is particularly useful when key omitted variables differ across counties but remain constant over time. By leveraging within-county variation, the model allows us to examine how changes in explanatory variables are associated with changes in housing values over time, thereby offering stronger causal inference than cross-sectional approaches such as OLS, SLM, or SEM.

In combination with OLS, SLM, and SEM, the Fixed-Effects approach completes our suite of specifications: it not only benchmarks against the classical, cross-sectional average effects (OLS) and probes spatial clustering of outcomes (SLM/SEM), but also hones in on within-county variation over time—providing a rigorous check that our results are driven by genuine local changes rather than by either spatial spillovers or static county traits.

3.3. Local Regression Analysis

Housing values can be influenced by factors that have localized effects, such as local economic conditions and environmental features. To capture local spatial heterogeneity that OLS, SLM, SEM, and Fixed-Effects models cannot, the Multi-Scale Geographically Weighted Regression (MGWR) model [59], a local regression model, is also employed to study the factors influencing spatial disparity in median housing values.

The use of the MGWR model is essential in this study due to the varying spatial scales at which key explanatory variables operate. The Dust Bowl data capture environmental exposure at a regional scale, reflecting broad climatic and ecological disruptions that extend beyond individual county boundaries (Figure 1). In contrast, most social and economic control variables—such as population demographics, housing characteristics, and labor market conditions—are available at the county level, representing a much finer spatial resolution. Additionally, the study includes a national-scale binary variable for the Depression era, which captures temporal shifts in housing values and associated factors before and after 1930 across the entire country. Traditional global regression models assume constant relationships across space and uniform spatial scales, which can obscure critical geographic variation and lead to biased or oversimplified conclusions when analyzing phenomena that operate across multiple levels. MGWR overcomes this limitation by allowing each covariate to operate at its own spatial scale, thereby capturing localized patterns of association. This is particularly important in a context where a regionally bounded environmental shock—such as the Dust Bowl—interacts with nationwide economic shifts and county-level heterogeneity in vulnerability, adaptation, and housing market dynamics. By accounting for these spatially heterogeneous relationships, MGWR offers a more flexible and accurate framework for understanding how different-scale environmental and economic forces jointly shape localized housing outcomes.

From a sustainability perspective, MGWR is particularly well suited for this analysis because it captures place-specific vulnerability and resilience by allowing relationships between housing values and socioeconomic or environmental factors to vary across space and scale. This flexibility is essential for sustainability research, as the impacts of economic crises and environmental shocks are inherently uneven across regions and communities.

The MGWR model is an extension of the standard Geographically Weighted Regression model that allows for each predictor variable to have its own bandwidth, enabling the model to capture spatial heterogeneity at different spatial scales. The mathematical formulation for the MGWR model is as follows:

$$Y_{it}={\beta }_0(u_i,v_i)+{\sum }_{k=1}^p{\beta }_k(u_i,v_i)X_{ikt}+\gamma Z_{it}+{ϵ}_{it}$$

(9)

where ${\beta }_0(u_i,v_i)$ is the intercept term that varies spatially with coordinates $\left(u_i,v_i\right)$; ${\beta }_k(u_i,v_i)$ are the spatially varying coefficients for each predictor $X_{ikt}$ at county $i$ and time $t$; $X_{ikt}$ are the predictor variables at county $i$ and time $t$. The subscript $k$ indexes the different predictor variables included in the model. If there are $p$ predictors, $k$ will range from 1 to $p$. Other variables are defined the same way as in Equation (5).

In this MGWR model, each predictor variable $X_{ikt}$ has its own bandwidth $b_k$, which determines the spatial scale at which the variable influences the dependent variable. The weight assigned to the observation county $j$ for the predictor $k$ at county $i$, typically determined by a kernel function that depends on the distance between counties $i$ and $j$ and the bandwidth $b_k$, is given by:

$$w_{ijk}=\mathrm{e}\mathrm{x}\mathrm{p}(-{\displaystyle \frac{d_{ij}^2}{2b_k^2}})$$

(10)

where $d_{ij}$ is the distance between counties $i$ and $j$.

The local parameter estimates ${\beta }_k(u_i,v_i)$ for each county $i$ are obtained by solving the weighted least squares problem:

$$\widehat{\beta }\left(u_i,v_i\right)={\left(X^T{W}_{i}X\right)}^{-1}{X}^T{W}_{i}y$$

(11)

where $X$ is the design matrix of predictor variables; $W_i$ is the diagonal weight matrix for county $i$, incorporating the individual bandwidths for each predictor; $y$ is the vector of observed dependent variable values.

MGWR can adapt to varying densities and distributions of spatial data, making it particularly useful in contexts where observations are irregularly spaced. This flexibility is an advantage over fixed spatial weights matrices used in SLM and SEM.

4. Results

4.1. How Did Housing Values Vary Across Different Counties During the 1930s?

Figure 2 shows the spatial disparity patterns of the median value of owner-occupied housing across different counties in 1930 and 1940. Figure 2a depicts the spatial disparity pattern in 1930 using the natural breaks (Jenks) classification method, and Figure 2b presents the spatial disparity pattern in 1940 using the same classification method. Figure 2c shows the percentage change in median value from 1930 to 1940.

Based on the maps in Figure 2a, it can be observed that, in 1930, the eastern part of the U.S., particularly the Northeast and the eastern part of the Midwest (that is, East North Central), displayed higher median housing values. The West Coast, especially California, also showed relatively higher housing values, similar to the Northeast. This was due to the concentration of population, economic activity, and better infrastructure in these areas. The West North Central and Mountain regions of the country, including states like Nebraska, Kansas, and the Dakotas, generally showed lower housing values. The South exhibited a mix of low to moderate values, with a few pockets of higher values in states like Texas and Florida.

As shown in Figure 2b, which uses the same legend to depict the 1930 values, one can see that 1940 had a noticeable decrease in median housing values across much of the country. The Northeast and East North Central regions continued to show higher values relative to other regions but with a noticeable decrease in intensity, suggesting a decline in housing values. The West Coast maintained relatively higher values but also showed a decrease compared to 1930. While still showing relatively lower values compared to the coasts, the Mountain and South regions also reflected the general decline in housing values. Figure 2c indicates that most of the country experienced a decrease in housing values. The Northeast, East North Central, and southern West Coast showed significant negative changes.

Comparing Figure 2a,b, it can be seen that the spatial disparity patterns of housing values are similar between 1930 and 1940. In 1930, there was a clear spatial disparity with higher housing values concentrated in the Northeast and the West Coast, while the Mountain and South regions lagged behind. By 1940, the disparity remained, but the overall decrease in housing values was evident. The Northeast and West Coast continued to lead in relative terms, but their absolute values had decreased. These patterns reflect the economic conditions of the time, including the impact of the Great Depression and the Dust Bowl, which led to a general decline in housing values across the country.

To identify significant spatial disparity patterns of housing values, a hot spot analysis was conducted. Figure 3 shows the hot spot analysis results of the median value of owner-occupied housing in 1930 and 1940. As shown in Figure 3a, in 1930, the Northeast, particularly around New York, and the West Coast, especially in California, show significant hot spots. These areas have high median housing values with high confidence levels (90%, 95%, and 99%). Other notable hot spots are observed in parts of the Midwest, such as around Chicago. Cold spots with high confidence levels are prominent in part areas of the South, Mountain and West North Central, particularly in states like New Mexico, Texas, Louisiana, Montana, South Dakota, and Nebraska.

Figure 3b shows hot spots of the median value of owner-occupied housing in 1940. The spatial disparity patterns are similar to those in 1930. The Northeast continues to show significant hot spots, although there is a slight reduction in intensity compared to 1930. The West Coast, mainly California, still exhibits hot spots, but with a more dispersed pattern. Cold spots remain prominent in some areas of the South and Mountain regions. Figure 3c displays the comparison results of the hot spots in 1930 and 1940. Regions that remained hot spots from 1930 to 1940 include parts of the Northeast, East North Central, and California. Regions that remained cold spots are prevalent in the South and parts of the Mountain region. The transitions from hot spots to cold spots or vice versa always occurred near existing hot or cold regions. Figure 3d shows hot spots of change in the median value of owner-occupied housing from 1930 to 1940. Areas with significant negative changes are noticeable in parts of the Northeast, Midwest, and West Coast (mainly southern California). Areas with significant positive changes occur mainly in the Mountain and the South (mainly South Atlantic and East South Central) regions. The Northeast and Midwest regions, which were industrial powerhouses at the time, suffered heavily from unemployment, business failures, and population loss, leading to declines in housing demand and property values. Parts of the Northeast were affected by outmigration as people moved westward and southward in search of new opportunities. Additionally, the farming crisis in the Midwest exacerbated economic challenges in rural areas. The Mountain region experienced less severe economic effects compared to the industrial heartlands. The relatively lower population density and slower development allowed for more stability in housing prices. Some parts of the South, particularly in the South Atlantic and East South Central regions, experienced some positive changes. This could be due to government intervention, such as New Deal programs that were designed to boost infrastructure and economic development in these historically underdeveloped regions. The introduction of labor-intensive industries in these areas also helped spur housing demand and raise property values.

4.2. How Were the Depression-Era, Dust Bowl, Education, Unemployment, and Income Factors Associated with This Spatial Inequality in Housing Values?

4.2.1. Global Effects

To assess the effects of the Depression-era, Dust Bowl, education, unemployment, and income factors on housing values across the entire country, four types of global regression models (OLS, SLM, SEM, Fixed-Effects models) are used to quantify the strength and direction of the relationships between housing values and these explanatory variables. Multiple other variables (listed in Table A1) are included as potential confounding control variables in these regressions. Due to multicollinearity issues, several control variables in Table A1 such as “pctWhite,” “TotalPop,” “pctWhiteDwel,” and “pctBlackDwel” are removed from the final models based on the calculated Variance Inflation Factor (VIF < 5.5 in this study) in the final models.

Table 1. Regression model results for interested variables.

Depend. Var. ln(MdnValHous)	(1) OLS		(2) SLM		(3) SEM		(4) Fixed-Effects
After	−0.304	***	−0.299	***	−0.339	***	−0.320	***
	(0.019)		(0.014)		(0.014)		(0.018)
DustBowl	−0.037	**	0.014		−0.033
	(0.018)		(0.016)		(0.030)
After*DustBowl	−0.145	***	−0.175	***	−0.178	***	−0.105	***
	(0.026)		(0.021)		(0.019)		(0.015)
pctIllit	−0.017	***	−0.009	***	−0.014	***	−0.013	***
	(0.002)		(0.001)		(0.002)		(0.002)
pctUnemp	−0.031	***	−0.023	***	−0.024	***	−0.014	***
	(0.002)		(0.001)		(0.001)		(0.002)
ln(SalRetailPerCap)	0.375	***	0.291	***	0.292	***	0.184	***
	(0.020)		(0.015)		(0.015)		(0.029)
Control Variables	yes		yes		yes		yes
Observations	4838		4838		4838		4838
R2	0.715		0.781		0.812		0.850
Adjusted R2	0.714						0.850

*** p < 0.01, ** p < 0.05.

lists the regression results for the variables of interest. Full model results, including all explanatory variables, are available upon request.

The variable “After” is a binary indicator for the year 1940 (1 if the observation is from 1940, 0 if from 1930), capturing the general effects of the Depression era across all counties. Its negative and statistically significant coefficient across all models indicates that, on average, median housing values declined between 1930 and 1940, consistent with the broader economic decline of the Great Depression. The “DustBowl” variable is a binary indicator identifying counties later affected by the Dust Bowl, based on historical classifications (Figure 1). Since the Dust Bowl emerged primarily after 1930, the coefficient on “DustBowl” reflects baseline differences in housing values in 1930 between Dust Bowl and non–Dust Bowl counties. The negative and significant coefficient in the OLS model suggests that Dust Bowl counties already had somewhat lower housing values prior to the onset of the environmental disaster, which may be due to preexisting economic or geographic disadvantages rather than Dust Bowl impacts per se. This effect, however, is not robust across all model specifications. The interaction term “After*DustBowl” is our primary variable of interest, as it captures the differential change in housing values from 1930 to 1940 in counties exposed to the Dust Bowl compared to those that were not. The negative and statistically significant coefficient across all models indicates that Dust Bowl counties experienced a steeper decline in housing values during the 1930s relative to non–Dust Bowl counties. This supports the interpretation that the Dust Bowl had a substantial, localized impact on housing markets, above and beyond the broader economic decline of the Great Depression.

The percentage of the illiterate population (pctIllit) negatively affected housing values in all models. Higher illiteracy rates were associated with lower housing values. The percentage of the unemployed population (pctUnemp) also negatively affected housing values, with higher unemployment rates leading to lower housing values. Because comprehensive income estimates are not available at the county level, we use retail sales per capita as a proxy for personal income based on the literature [18]. The natural logarithm of sales per capita in retail (ln(SalRetailPerCap)) positively affected housing values, indicating that higher income (retail sales per capita) was associated with higher housing values. The R² values indicate the proportion of the variance in the dependent variable that is predictable from the independent variables. The relatively high R² values (OLS: 0.715, SLM: 0.781, SEM: 0.812, Fixed-Effects: 0.850) suggest that these models are reasonable. The Fixed-Effects model has the highest R² value, suggesting it provides the best fit among the models.

To facilitate the comparison of the relative importance of different variables in the models, standardized regression models are employed in this study.

Table 2. Standardized regression model results for interested variables.

Depend. Var. ln(MdnValHous)	(1) OLS		(2) SLM		(3) SEM		(4) Fix-Effect
After (Scaled)	−0.267	***	−0.262	***	−0.298	***	−0.282	***
	(0.017)		(0.012)		(0.012)		(0.016)
DustBowl (Scaled)	−0.025	**	0.015		−0.022
	(0.012)		(0.010)		(0.020)
After*DustBowl (Scaled)	−0.072	***	−0.088	***	−0.088	***	−0.052	***
	(0.013)		(0.010)		(0.009)		(0.008)
pctIllit (Scaled)	−0.150	***	−0.073	***	−0.125	***	−0.114	***
	(0.017)		(0.012)		(0.015)		(0.019)
pctUnemp (Scaled)	−0.228	***	−0.162	***	−0.175	***	−0.107	***
	(0.014)		(0.010)		(0.010)		(0.012)
ln(SalRetailPerCap) (Scaled)	0.358	***	0.265	***	0.278	***	0.176	***
	(0.019)		(0.014)		(0.015)		(0.028)
Control Variables	yes		yes		yes		yes
Observations	4838		4838		4838		4838
R2	0.715		0.791		0.812		0.850
Adjusted R2	0.714						0.850

*** p < 0.01, ** p < 0.05.

presents the results of standardized regression models for the variables of interest. The results of standardized regression models for all explanatory variables can be provided upon request.

In all of the models (OLS, SLM, SEM, and Fixed-Effects models), “ln(SalRetailPerCap)” had the strongest positive impact on housing values, highlighting the importance of income (retail sales per capita) in determining housing values. The “After” variable consistently had the largest negative effect on housing values, emphasizing the significant decrease in housing values during the Great Depression-era. While “pctUnemp” also had a strong negative effect in the OLS and SLM, its impact was less pronounced compared to “After” in the SEM and Fixed-Effects models. “pctIllit” showed moderate negative effects across all models, and “After*DustBowl” had the smallest but still significant negative impact.

4.2.2. Local Effects

While global regression models provide the overall relationships between housing values and explanatory variables, they assume that these relationships are constant across the entire study area. In reality, however, these relationships often vary across space. To capture the spatial non-stationarity and localized effects of various factors on housing values, MGWR is employed in this study for a more detailed and accurate analysis of how different variables influenced housing values across different regions.

Figure 4 presents the coefficients of key variables in the MGWR model, while the coefficients of other variables are shown in Figure A1 (see Appendix A). Each map illustrates the spatial distribution of coefficients for specific variables, highlighting how their relationships with housing values vary across regions and providing insights into the underlying drivers of these patterns. Figure 4a displays the spatial distribution of the coefficient for the “After” variable. Regions with negative coefficients include the West Coast, West North Central, and West South Central areas. The West Coast, despite being more urbanized, suffered from high unemployment and reduced industrial output during the Depression. Many industries that drove housing demand, such as construction, agriculture, and manufacturing, were severely impacted. These regions lacked the economic resilience to cushion the housing market from the full effects of the economic downturn. The West North Central and West South Central regions were heavily reliant on agriculture, which was devastated by the collapse in crop prices and the Dust Bowl that followed shortly after. This led to rural depopulation, bankruptcies, and foreclosures, further depressing housing markets.

Figure 4b shows the spatial distribution of the “DustBowl” coefficient, revealing a mixed association between Dust Bowl counties and housing values. Some regions show positive associations, while others exhibit negative ones. However, most of these coefficients are statistically insignificant. The mixed association implies that the Dust Bowl did not affect all regions uniformly. Some counties within the Dust Bowl region may have experienced more severe agricultural losses, economic hardship, and outmigration, leading to negative impacts on housing values. In contrast, other counties might have been less affected or may have benefitted from spillover effects such as government aid, infrastructure development, or migration from severely impacted areas, thus showing positive or neutral impacts on housing values. The fact that most coefficients are statistically insignificant suggests that, on the whole, being a Dust Bowl county had no strong association with housing values.

Figure 4c illustrates the spatial distribution of the interaction term “After*DustBowl,” our primary variable of interest that captures the Dust Bowl’s compounded effect. Negative coefficients indicate a greater decline in housing values in Dust Bowl counties after the disaster, with the most significant effects concentrated in the West Coast, Mountain, West South Central, and West North Central regions, particularly in the southern areas. These regions were both directly impacted by the Dust Bowl and vulnerable to broader economic pressures during the Depression, resulting in a compounded negative impact on housing values. The Dust Bowl primarily affected the Great Plains, which includes much of the West North Central and West South Central regions. The ecological disaster devastated agriculture, the primary economic activity in these areas, leading to widespread crop failures, farm closures, and mass outmigration. With agriculture collapsing, there was little demand for housing in these rural areas, causing housing values to plummet. Even areas that were not directly hit by the Dust Bowl, such as parts of the West Coast, experienced negative spillover effects. Additionally, the broader national economic downturn meant that even regions on the periphery of the Dust Bowl were not immune to its effects, as agricultural and economic ties to Dust Bowl counties weakened housing demand more broadly.

Figure 4d maps the coefficient for the percentage of the illiterate population (pctIllit), where negative coefficients indicate that higher illiteracy rates were associated with lower housing values. This effect was widespread, especially notable in the West South Central and Midwest regions. The relationship likely stems from the fact that higher illiteracy rates often coincide with lower educational attainment and limited employment opportunities, which can suppress local incomes and demand for housing, driving down property values in those areas. The West South Central and Midwest regions, during this period, were heavily reliant on agriculture and low-skilled labor industries, which were more vulnerable to economic downturns like the Great Depression. The combination of a high illiterate population and fewer economic opportunities meant these regions lacked the economic dynamism seen in more diversified or urbanized areas, further depressing housing demand and values.

Figure 4e presents the spatial distribution of the coefficient for the percentage of the unemployed population (pctUnemp), with predominantly negative coefficients suggesting that higher unemployment rates were linked to lower housing values. The significant impact in the central regions, including the Midwest and West South Central, can be attributed to the economic hardships these areas faced during the Depression. The Midwest and West South Central regions were heavily reliant on agriculture and manufacturing sectors that were hit especially hard by the Great Depression. As crop prices collapsed and industrial production slowed, unemployment soared in these regions. The resulting economic contraction weakened local economies, reducing investment in housing and decreasing property values. The dependence on these struggling sectors made these regions particularly vulnerable to unemployment-related declines in housing demand. In response to the economic hardship and high unemployment, many people migrated out of rural areas and small towns in search of jobs, particularly in the Midwest and West South Central. This migration contributed to depopulation in already struggling areas, leaving many homes abandoned or devalued due to lack of demand. The loss of population further depressed housing values in these regions.

Finally, Figure 4f presents the spatial distribution of the coefficient for the natural logarithm of sales per capita in retail establishments (ln(SalRetailPerCap)), illustrating the relationship between retail activity and housing values across U.S. counties. Positive coefficients indicate that higher retail sales per capita are associated with higher housing values. This relationship is particularly strong in the western U.S., including the West Coast, Mountain, and West North Central regions. These regions, especially on the West Coast, had more diversified economies compared to the agriculturally dependent Midwest and Southern regions. The economic diversity in the West—encompassing industries such as shipping, manufacturing, and early forms of entertainment in cities like Los Angeles and San Francisco—helped shield these areas from the worst effects of the Great Depression. In contrast, counties in the Midwest and Southern regions, which were more reliant on agriculture and natural resource extraction, experienced more severe economic devastation. This vulnerability likely weakened the connection between retail activity and housing values in these areas. The figure suggests that the resilience of regional economies during the 1930s, driven by more diverse industries in certain areas, supported higher levels of retail activity, which in turn contributed to stronger housing demand and higher housing values. Counties in the West were better positioned to maintain consumer spending, supporting both retail sectors and housing markets during this challenging economic period.

In general, these maps illustrate the spatial heterogeneity in the relationships between these key variables and housing values, highlighting the importance of using MGWR to capture these local variations. MGWR has the highest R² value (0.863), indicating that it explains the most variance in housing values.

4.3. How Did Racial and Nativity Group Differences in Housing Values Manifest Across Counties and Within Shared Geographic Contexts?

To examine the impact of race and nativity on housing disparities across counties during the 1930s—beyond the neighborhood scale—we first employ bivariate color mapping, an advanced cartographic technique, to visualize and analyze the spatial relationship between median housing values and racial/nativity composition. Figure 5 presents the relationships between median housing values and various population percentages using bivariate colors across counties.

Figure 5a–c depicts the relationship between median housing values and Black population percentages. Figure 5a shows that in 1930, the South had significant Black populations paired with low housing values, while the Northeast and Midwest generally had higher housing values with low Black population percentages. The West displayed a mix of high and low housing values with low Black populations. This highlights the economic and demographic disparities across the country in 1930. By 1940, as shown in Figure 5b, these patterns remained largely consistent, suggesting stability in the relationship between these variables over the decade. Figure 5c shows that between 1930 and 1940, many Black individuals moved from the South to the North and West, though these regions still had relatively smaller Black populations by 1940. The Northeast had counties with high housing values and stable or increasing Black populations, while the South and Midwest displayed a mix of housing values and population changes. The Western states saw increases in Black population percentages, but with varying housing values. The Southern Plains and Texas showed significant Black population increases in areas with lower housing values.

Figure 5d–f illustrates the relationship between median housing values and the percentages of the ‘Other’ populations, where ‘Other’ refers to populations other than Black and White. These maps reveal several important spatial patterns. In both 1930 and 1940, the West region consistently shows high percentages of ‘Other’ populations with varying housing values. This indicates that the West was a significant destination for ‘Other’ populations. Many areas in the Northeast and Midwest consistently exhibit high housing values with low percentages of ‘Other’ populations in both 1930 and 1940, suggesting economic prosperity in these areas with limited immigration of ‘Other’ populations. The Southeast (including South Atlantic and East South Central) region shows persistent low percentages of ‘Other’ populations with low housing values, a trend that remains consistent from 1930 to 1940, indicating ongoing economic challenges and limited ‘Other’ foreign immigration in this region. As shown in Figure 5f, although ‘Other’ populations moved from the West to other parts of the country between 1930 and 1940, the highest percentiles of population change occurred in the eastern part of the country with both low and high housing values.

Figure 5g–i illustrates the relationship between median housing values and Native White population percentages. As shown in Figure 5g,h, in both 1930 and 1940, many areas in the Northeast, Midwest, and West had high percentages of Native White populations and high housing values. The Southeast and southern Mountain regions consistently had low percentages of Native White populations with low housing values, indicating economic challenges and a limited Native White presence there. Figure 5i highlights the relationship between median housing values in 1940 and changes in Native White population percentages from 1930 to 1940. The West region saw significant increases in Native White populations, including some areas with lower housing values. Some Northeast and Midwest areas with high housing values also had significant increases in Native White populations, reflecting economic prosperity and demographic growth in these regions. The central and southeast regions, while showing low housing values, experienced limited changes in Native White population percentages, indicating persistent economic challenges.

Figure 5j–l illustrates the relationship between median housing values and Foreign White population percentages. Based on Figure 5j, the Foreign White population in 1930 was concentrated primarily in the western part (including West and West South Central) of the country. In contrast, many areas in the Northeast and Midwest had low percentages of Foreign White population, despite having high median housing values. This distribution suggests that the western regions were a major destination for Foreign White immigrants, while the Northeast and Midwest as well as the Southeast (South Atlantic and East South Central) were not. Based on Figure 5k, by 1940, some areas in the Northeast and Midwest regions saw an increase in Foreign White populations, especially in counties with urban centers and high median housing values. This shift suggests that economically prosperous areas attracted more Foreign White immigrants during this period. The western regions continued to attract a significant proportion of Foreign White populations in 1940. The Southeast shows persistent low housing values and low percentages of Foreign White populations in both 1930 and 1940, indicating continued economic challenges and limited foreign immigration. The similarity between Figure 5k,l indicates that the regions with high percentages of Foreign White populations in 1940 were also the regions where significant increases in the Foreign White populations occurred over the decade from 1930 to 1940. This suggests that these areas were consistent destinations for White immigrants throughout the decade, maintaining or increasing their attractiveness to these immigrants.

To statistically validate and reinforce the spatial patterns shown in Figure 5, we conducted a bivariate LISA analysis. Figure 6 presents the bivariate LISA cluster results for median housing values (in 1930 or 1940) and racial/nativity population percentages across counties, while Figure 7 shows the results for housing values and changes in racial/nativity composition from 1930 to 1940. In both figures, the left column displays the cluster maps, while the right column shows the corresponding maps of statistically significant clusters based on Monte Carlo permutation tests (999 replications). These results provide formal spatial evidence that supports and sharpens the visual associations identified in Figure 5.

In Figure 6a, which examines median housing values and Black population percentages, the Southern U.S. features prominent Low–High clusters (light blue), indicating counties with low housing values surrounded by counties with high Black population shares. This pattern underscores spatial inequality rooted in racial segregation and economic underdevelopment. Meanwhile, High–Low clusters (light red) appear in parts of the Midwest and Northeast, where counties with high housing values are surrounded by counties with low Black population shares. In Figure 6c, bivariate LISA results for the “Other” racial group reveal High–High clusters (dark red) in the Western U.S., particularly in California, Arizona, New Mexico, and parts of Washington, indicating these areas’ historical role as immigrant destinations for non-Black, non-White populations. In contrast, Low–Low clusters (dark blue) are widespread in the Midwest and also appear in parts of the Northeast and interior South, including areas of Ohio, Indiana, Missouri, Western Pennsylvania, and Upstate New York. These areas had both lower housing values and lower shares of “Other” racial groups, reflecting limited demographic diversification in economically stagnant or rural regions. In Figure 6e, Native White percentages show High–High clustering (dark red) in the Northeast, Midwest, and Mountain West, where both housing values and Native White concentrations are relatively high. Low–Low clusters (dark blue) appear in the Southeast and Deep South, consistent with areas of lower housing values and lower Native White concentration. In Figure 6g, Foreign White percentages produce High–High clusters (dark red) in the Mountain West, Upper Midwest, and parts of the Northeast, supporting their historical role as immigrant destinations. Conversely, Low–Low clusters (dark blue) dominate the Southeast, where housing values and foreign-born White populations are both low.

Turning to Figure 7, which examines changes in racial and nativity composition from 1930 to 1940, the spatial patterns broadly reinforce earlier findings about housing and demographic inequality. Figure 7a displays a mix of High–High (dark red) and Low–High (light blue) clusters across the Midwest, Northeast, and Southwest, indicating that Black populations during the 1930s increasingly migrated from the South into counties where nearby areas were experiencing significant growth in Black population shares. Some of these destination counties had relatively high median housing values (High–High), while others had low values (Low–High), reflecting the varied housing market contexts into which Black migrants settled. These county-level patterns underscore both the regional scale of the Great Migration and the spatial inequalities embedded in access to higher-value housing markets. In Figure 7c, changes in the “Other” population show High–High clusters (dark red) concentrated in the Eastern U.S., particularly the Northeast, Mid-Atlantic, and parts of the Southeast, where counties with high housing values were near areas experiencing growth in non-Black, non-White populations. In contrast, the Western U.S. shows predominantly High–Low clusters (light red), suggesting that high housing value areas were located near counties with relatively limited increases in “Other” population shares. Figure 7e highlights High–High clusters (dark red) for Native White population changes in the Southwest, including Arizona, New Mexico, Texas, and Southern California, where rising housing values coincided with surrounding counties experiencing Native White demographic growth. Low–Low clusters (dark blue) appear in the Southeast, Appalachia, and parts of the Midwest, reflecting regional stagnation in both housing values and demographic change. Finally, Figure 7g displays High–High clusters (dark red) between housing values and changes in Foreign White population percentages in the West and scattered parts of the Midwest and Northeast, suggesting these regions remained attractive to foreign-born White immigrants during the 1930s. Meanwhile, Low–Low clusters (dark blue) are concentrated in the South, reinforcing the broader pattern of demographic exclusion and economic underdevelopment in that region.

Together, the bivariate LISA results in Figure 6 and Figure 7 confirm that the spatial relationships illustrated in Figure 5 are statistically significant and geographically structured. The clustering of racial/nativity groups and their demographic changes with housing values—whether high or low—underscores the enduring intersection of race, immigration, and economic geography in shaping housing disparities during the 1930s.

Please note that although minority populations sometimes lived in counties or regions with relatively high housing values during this period, significant housing disparities persisted between racial and nativity groups within the same counties or regions. While our bivariate spatial analyses focused on county-level associations, Figure 8 and Figure 9 provide more direct evidence of disparities by allowing intra-county and intra-regional comparisons across racial and nativity groups.

Figure 8 offers a comparative analysis of median values of non-farm homes in the Southeast in 1930, disaggregated by race. It includes side-by-side maps and histograms that reveal stark contrasts in both the spatial distribution and statistical characteristics of housing values for Black and Native White households. The maps in Figure 8a,c show that across many of the same counties, the median housing values for Black households were consistently lower than those for Native White households. This disparity is further illustrated by the histograms: Figure 8b shows that the distribution of Black household housing values is highly skewed toward the lower end, with a mean of $1076 and a median of $900. Very few counties had median values exceeding $2675. In contrast, Figure 8d shows that housing values for Native White households were substantially higher, with a mean of $3245 and a median of $3141. The distribution approximates a normal curve, concentrated in mid- to upper-value ranges. These visualizations demonstrate that even within the same counties, Black and White households experienced markedly different housing value conditions.

Figure 9 presents a similar comparison based on nativity, focusing on Foreign White and Native White households in the Northeast, Midwest, and West. The maps reveal that the two groups shared broadly similar spatial distributions, often residing in the same high-value housing areas. However, the histograms in Figure 9b,d show that Native White households consistently had slightly higher median and mean housing values than Foreign White households (e.g., Native White mean: $3636 vs. Foreign White mean: $3285). These differences suggest a nativity-based housing advantage, even among White populations living in the same general locations. Together, Figure 8 and Figure 9 strengthen the argument that racial and nativity-based housing disparities were not only regional but also present within the same local contexts—at the county and sub-county level.

Table 3. Pearson correlations between median housing values and race/nativity variables.

Pearson Correlation r	MdnValHous (1930 & 1940)	MdnValHous (1930)	MdnValHous (1940)
pctWhiteDwel	0.1721	0.2419	0.1548
pctWhite	0.1656	0.2346	0.1479
pctBlackDwel	−0.1409	−0.1811	−0.1276
pctBlack	−0.1389	−0.1794	−0.1178
pctOther	−0.0907	−0.1544	−0.1193
pctNatW	0.0332	0.0749	0.0437
pctForW	0.01	−0.0603	0.4171

presents Pearson correlations between median housing values and various race/nativity variables for the years of 1930 and 1940. It highlights the influence of racial and nativity composition on housing values. White populations generally correlated with higher housing values, whereas Black and other minority populations correlated with lower housing values. Particularly noteworthy is the shift in correlation for foreign white populations, which moves from a notable negative correlation in 1930 to a strong positive correlation in 1940. During the 1930s, many foreign white populations might have initially settled in areas with lower housing values due to economic constraints and recent arrival. Over the decade, as these people integrated into the economy, they might have moved to more prosperous areas, contributing to higher housing values in 1940. In addition, changes in immigration policies during this period might have affected the socio-economic characteristics of foreign white populations. More skilled or economically stable immigrants might have been allowed to enter, leading to higher housing values in areas they settled by 1940.

5. Discussion

This study examined spatial disparities in housing values during the Great Depression through a place-based sustainability lens, highlighting how historical housing markets reflected uneven access to economic opportunity, social infrastructure, and long-term wealth-building potential. Rather than viewing housing values solely as market outcomes, these patterns underscore deeper structural differences in regional resilience and vulnerability. The persistence of high- and low-value clusters between 1930 and 1940 suggests that spatial inequality in housing markets was not merely a short-term response to economic contraction, but reflected enduring regional development trajectories shaped by broader economic and demographic forces.

Large-scale economic and environmental shocks played a critical role in shaping these sustainability outcomes across space. While the Great Depression led to a nationwide decline in housing values, the Dust Bowl imposed additional, localized losses in affected counties, exacerbating pre-existing vulnerabilities. These findings illustrate how environmental stressors can amplify economic distress and produce spatially uneven sustainability outcomes. Moreover, the strong associations between housing values and socioeconomic factors—such as illiteracy, unemployment, and income proxies—highlight the importance of human capital and local economic structure in sustaining housing markets during periods of crisis.

Our study demonstrates the necessity of using Multiscale Geographically Weighted Regression (MGWR) to capture the spatial heterogeneity in the relationships between housing values and explanatory variables during the 1930s. Because our explanatory variables operate at multiple spatial scales—ranging from local socioeconomic indicators (e.g., illiteracy and unemployment rates) to broader economic measures (e.g., regional Dust Bowl wind erosion or national Depression-era variable)—MGWR offers a more appropriate modeling framework than global or single-scale local regressions. While global regression models assume spatially uniform effects, MGWR revealed that the influence of key factors varied substantially across regions. For example, negative housing value trends were concentrated in areas like the West Coast and Great Plains, where economic distress and environmental disasters such as the Dust Bowl significantly undermined housing markets. In contrast, more diversified regional economies, particularly in the western U.S., exhibited positive relationships between retail sales and housing values, reflecting localized economic resilience. Higher illiteracy and unemployment rates, especially in the Midwest and South, were associated with lower housing values, emphasizing the vulnerability of agriculturally dependent regions. The spatial variability uncovered by MGWR confirms that incorporating geographically varying effects—and accounting for the distinct spatial scales at which explanatory variables operate—is crucial for accurately analyzing the impacts of historical economic crises on housing markets.

Our spatially explicit analyses also reveal substantial racial and nativity-based disparities that intersect with regional housing inequality. We examined the impact of racial and nativity composition on housing disparity using bivariate color mapping and bivariate LISA analysis methods. Our findings from bivariate LISA maps (see Figure 6 and Figure 7) largely reinforce the visual patterns seen in the bivariate color maps (Figure 5). These results validate that the observed spatial patterns are not merely visual artifacts, but represent statistically meaningful geographic relationships. We found that housing values varied significantly based on racial and nativity composition. The South region showed persistent significant Black populations with low housing values. The western states (West Coast, Mountain, and North West Central) had high percentages of ‘Other’ populations and Foreign White populations with consistently varying housing values. Many areas in the Northeast, Midwest, and West had high percentages of Native White populations with high housing values. While minority populations resided in counties or regions with high overall housing values during this period, a significant disparity persisted between them and the native White population within the same areas. Black households were concentrated in neighborhoods with substantially lower median housing values compared to those of Native White households. Foreign White households, though generally living in areas with slightly lower housing values, still faced disparities when compared to Native White households in the same counties or regions.

5.1. Implications for Contemporary Housing Inequality, Sustainability, and Policy

Although this study focuses on housing values during the Great Depression, the spatial disparities identified have important implications for understanding contemporary housing inequality and place-based sustainability. A large body of research demonstrates that spatial patterns of economic advantage and disadvantage tend to persist over time through institutional reinforcement and cumulative disadvantage associated with residential segregation [60]. Regions that experienced sustained declines in housing values during major economic and environmental shocks often entered subsequent periods with weakened fiscal capacity, reduced infrastructure investment, and limited opportunities for wealth accumulation, contributing to long-lasting spatial inequality in housing markets [61].

The findings of this study suggest that regions most adversely affected by the Great Depression and the Dust Bowl—particularly parts of the South, Mountain, and Great Plains regions—faced compounded vulnerabilities that likely shaped long-term housing and development trajectories. Historical declines in housing values reduced household wealth and local tax bases, constraining future investments in housing quality, public services, and infrastructure. Prior research shows that such early shocks can generate persistent regional disparities through mechanisms including long-run population redistribution and constrained local economic adjustment [18,62]. These dynamics help explain why spatial inequalities observed during the 1930s may continue to influence regional housing outcomes decades later.

Racial and nativity-based housing disparities documented in this study also have clear contemporary relevance. Discriminatory housing practices and unequal access to credit during the early twentieth century limited wealth accumulation for minority populations and shaped residential patterns that continue to influence housing outcomes today. Even when minority populations resided in regions with relatively high overall housing values, disparities persisted within shared geographic contexts, consistent with structural exclusion in housing markets shaped by historical policies [63].

From a broader sustainability perspective, these historical findings illustrate how economic and environmental shocks can entrench long-term spatial inequality when coupled with structural disadvantage. Understanding these historical dynamics provides valuable context for current efforts to promote sustainable, resilient, and equitable housing systems. From a policy standpoint, contemporary housing policies aimed at affordability, resilience, and sustainability may benefit from explicitly accounting for historically rooted spatial disadvantages rather than relying solely on place-neutral interventions, as literature on place-based economic development suggests that targeted approaches can improve outcomes in disadvantaged areas [64,65]. By situating present-day housing challenges within a historical spatial framework, this study suggests the importance of integrating historical insight into the design of place-based sustainability and housing policies.

5.2. Limitations and Directions for Future Research

Although we provided valuable results about the spatial disparity in housing values during the Great Depression, it is worth mentioning that a few limitations exist in this study due to some objective reasons. First, the study relies on historical census data from 1930 and 1940, which may be incomplete or inaccurate, potentially affecting the robustness of the findings. Using retail sales per capita as a proxy for personal income might not fully capture the true economic conditions of the regions, leading to potential biases in the analysis. Due to data constraints, the study does not account for the impact of specific local, state, or federal policies that might have influenced housing values during the Great Depression, potentially confounding the results.

Second, while the MGWR framework offers important advantages for capturing spatial heterogeneity, its results are sensitive to several modeling choices. In particular, local coefficient estimates may be influenced by bandwidth selection, kernel specification, and spatial weighting schemes. In this study, bandwidths for each covariate were selected using the standard cross-validation procedure implemented in the MGWR framework, which aims to balance model fit and overfitting by optimizing predictive performance. Nevertheless, alternative bandwidth choices or kernel functions could lead to differences in the magnitude or spatial smoothness of local parameter estimates. Similarly, MGWR results depend on assumptions embedded in spatial weights and distance metrics, and different specifications could alter localized patterns, particularly in regions with irregular county geometries. As a result, the MGWR findings should be interpreted as indicative of spatial heterogeneity rather than precise local causal estimates. Despite these limitations, the consistency between MGWR results and the global regression findings provides confidence that the main conclusions regarding spatial disparities in housing values are not driven solely by specific MGWR parameter choices.

Third, while this study documents strong spatial associations between Dust Bowl exposure and housing market outcomes, it is not designed to provide a causal estimate of the Dust Bowl’s impact on housing values. In particular, the analysis does not explicitly control for differences in agricultural production types, baseline crop composition, or pre-existing regional economic structures that may have influenced vulnerability to drought conditions during the 1930s. As a result, some observed spatial patterns may partially reflect underlying regional characteristics rather than the Dust Bowl shock alone. Formal causal identification strategies—such as placebo tests, falsification exercises, or exploiting quasi-exogenous variation in Dust Bowl intensity—are beyond the scope of this paper, which focuses on spatial disparities, regional heterogeneity, and place-based patterns in housing values during the Great Depression. The results should therefore be interpreted as descriptive spatial relationships rather than causal estimates. Causal identification of the Dust Bowl’s effects on housing markets is examined in a companion study by the authors, which employs difference-in-differences designs, alternative treatment definitions, and robustness and placebo tests to isolate the impact of Dust Bowl exposure on housing prices [66]. Taken together, the two studies provide complementary perspectives: the present paper highlights spatial inequality and heterogeneity, while the companion study rigorously evaluates causal effects of the Dust Bowl.

Finally, the observational nature of the study, combined with data and model limitations, makes it difficult to establish causal relationships between the explanatory variables and housing values. In addition, the study focuses on a specific period (1930–1940), limiting the ability to generalize findings to other time periods with different economic conditions and policies. Future research could extend this work by systematically testing the sensitivity of MGWR results to alternative bandwidth selection criteria and spatial weighting schemes, incorporating additional historical policy variables, or linking historical spatial disparities in housing values to contemporary housing inequality.

Despite these limitations, the results from this study may provide a historical perspective to understand how current housing disparity and segregation patterns are rooted in past practices. Wheelock [27] and White [67] noted similarities between the Great Depression and recent economic downturns, suggesting that a review of historical events can provide valuable insights into alternative policies to relieve mortgage distress faced today. We hope our analysis results may provide historical context of housing disparity and inform contemporary efforts to design sustainable and equitable housing policies.

6. Conclusions

This study examined spatial disparities in housing values across U.S. counties during the Great Depression, framing housing values as an indicator of place-based sustainability. Using historical census data, GIS visualization, and spatial econometric methods, we documented persistent regional inequalities and substantial racial and nativity-based disparities in housing values. Higher housing values were concentrated in the Northeast, Midwest, and West Coast, while the South and Mountain regions experienced persistently lower values. Although housing values declined nationwide between 1930 and 1940, spatial disparities remained pronounced, highlighting uneven regional resilience to economic and environmental shocks.

By revealing how large-scale crises such as the Great Depression and the Dust Bowl interacted with local socioeconomic and demographic structures, this study provides insight into the historical roots of housing inequality. The findings underscore that spatial and racial disparities in housing values were not transitory outcomes of short-term shocks, but reflected deeper structural processes that continue to shape housing outcomes today. By situating contemporary housing challenges within a historical spatial framework, this study contributes to sustainability research by emphasizing the importance of place-based perspectives in understanding and addressing long-term housing inequality.