Much is known about US inequality at the national level. But what about inequality at the level of individual cities, towns and neighbourhoods? Joel Suss, Tom Kemeny and Dylan S Connor introduce the first comprehensive dataset that shows trends in wealth inequality at the local level – opening up new avenues of research in this area.

Wealth inequality – defined as disparities in household assets net of debts – has been sharply rising in the United States, and across many other high-income countries. Various studies suggest that the richest 1 percent of families in the United States now possess almost 35 percent of total national wealth, a share that has grown by more than 50 percent since 1980. These trends are concerning because of the adverse effects of wealth inequality on social, economic, political, and health outcomes, disproportionately affecting non-white households.

While there is clear evidence of country-level trends and outcomes, there is surprisingly very little we knowledge about wealth inequality at finer geographical scales. We simply do not have a clear understanding of where in the United States wealth inequality has increased the most or which communities have been left behind.

Yet understanding how wealth varies within countries is important because wealthier cities and towns have greater resources to invest in schools, health care, transportation and other infrastructure; we know that differences in these public goods confer benefits that can span generations and improve life chances for children born into poverty. Even in places where wealth exists, if it is distributed unequally, it can result in underinvestment in essential local public goods. A lack of wealth or extreme wealth inequality can also fuel resentment, leading to support for anti-system politics. Moreover, changes in wealth, especially housing wealth, shape consumer behavior, which can have major effects on the availability and quality of local jobs.

Why do we know so little about the geography of wealth? Few data sources report on personal assets and debts, and those that do are highly restrictive due to confidentiality concerns. Researchers have access to data on housing markets, but home ownership is only one channel among several by which wealth can vary across locations. In practice, home values and net wealth are only moderately correlated across American households, since the latter also includes households’ financial assets like stocks and bonds, as well as various forms of debt – credit cards, student loans, and so on.

In our recent article in Scientific Data, we introduce the first comprehensive dataset detailing the changing geography of wealth inequality from 1960 to 2020. This data compendium, referred to as the Spatial Wealth Inequality Database (“GEOWEALTH-US”), utilises a novel machine-learning-based method that can estimate distributions of wealth at various scales, from households to regions.

The new dataset relies on “ensemble” learning techniques to generate predictive models of household wealth, using rich survey information from the Federal Reserve’s Survey of Consumer Finances (SCF). We then use these models to impute wealth among households in Census population surveys that include geographical identifiers. The end result is a dataset that permits description of the variation in average wealth between places (“geography of wealth”), as well as how wealth is distributed within cities and regions (“local wealth inequality”). Our paper reports on the extensive validation we have undertaken against a wide range of published data sources, providing confidence in the reliability of our underlying approach.

Importantly, our estimates at the country-level track other published estimates. Figure 1 compares our estimates of the share of national wealth held by top wealth holders against state-of-the-art estimates produced by other researchers, including: Piketty, Saez and Zucman (2018); Saez and Zucman (2020); and Smith, Zidar and Zwick (2022). In this figure, validation comes from the strong alignment between our approach (visualised in orange) and these and other leading methods.

Figure 1: Comparing our national top wealth share estimates against other (state-of-the-art) estimates

Note: For six different measures, each panel tracks the share of total national wealth held by a specific top percentile of the wealth distribution. Across the panels, the different series are: ‘Ensemble’, in orange, which is the estimate generated using our ensemble model; ‘SCF’, estimated from the raw Survey of Consumer Finances; ‘SCF w/Forbes’ which adds the Forbes 400 to the raw SCF; ‘PSZ (2018)’ which makes use of Piketty, Saez and Zucman’s estimates using the distributional national accounts framework; ‘SZ (2020)’ which is an updated distributional macroeconomic accounts series generated by Saez and Zucman; and ‘SZZ (2023)’ which uses income-capitalised estimates based on tax data from Smith, Zidar and Zwick.

Of course, unlike these other estimates which provide a snapshot of the national situation, the main virtue of our approach is that it enables us to track wealth and wealth inequality at relatively fine-grained spatial scales. These include: public-use microdata areas (PUMAs), which are small statistical areas containing roughly 100,000 residents; commuting zones (CZs), which denote labour market areas such as, for example, greater New York City; and states.

Our initial exploration of subnational geographical patterns in the data reveals three key facts.

First, the distribution of wealth between regions has become meaningfully more unequal since 1960. Figure 2 plots the inequality of average commuting zone wealth over time (using a standard measure, the Gini Coefficient). In other words, US wealth holdings have become increasingly concentrated in a smaller set of regions.

Figure 2 also shows that inter-regional wealth disparities have grown much larger than inter-regional income disparities, with income gaps growing between 1960 and 2020 by 20 percent and wealth by 42 percent. This supports the intuition that spatial wealth inequalities require investigation over and above the study of income inequality. The sharp exacerbation of wealth inequality over this period makes this a particularly urgent topic for further research.

Figure 2: Inter-regional disparities in average wealth and average income (across US commuting zones from 1960–2020)

Note: For the period 1960–2020, Gini coefficients describe trends in inter-regional inequality in terms of average household income and wealth across US commuting zones (defined using 1990-vintage commuting flow data). Wealth estimates come from the GEOWEALTH-US dataset. Household income series is from Kemeny & Storper. This figure shows that wealth gaps between places have grown much more sharply than income gaps. 

The new dataset also allows us to analyse patterns of local wealth inequality – in other words, inter-household wealth inequality within local areas. Comparing the maps in Figure 3 below, we can see a mix of both changes and continuity in local wealth inequality over time. In 1960, intra-regional wealth was high throughout the South, low in the Midwest and Northern Plains regions, and more mixed along the coasts. The main change to this pattern up to 2020 has, however, been the dramatic rise in inequality in the Midwest and Plains regions. One example is the Minneapolis labour market area, which saw the share of wealth held by the top 0.1% of households more than double from 1960 to 2020, going from roughly 8% to over 16%. This convergence in inequality between the South and the Midwest is consistent with findings from studies of regional income inequality and social mobility, suggesting interdependence and common underlying sources affecting different facets of spatial inequality.

Figure 3: Changes in wealth inequality within US commuting zones, 1960 to 2020

Note: Gini coefficients for wealth distribution within commuting zones, 1960 & 2020. For 1990-vintage commuting zones (CZs) as in Tolbert and Sizer, this figure maps Gini coefficients Z-scores tracking within-CZ wealth inequality in 1960 and 2020.

To return to where we started at the outset of this post, wealth inequalities continue to assert themselves as one of the greatest societal challenges of the 21st century. Investigating the local and regional facets of this problem is of critical importance. The data infrastructure and initial findings presented here represent the first step toward understanding the changing dynamics of wealth inequality across the neighbourhoods, cities and regions of the United States.

Note: the dataset is openly available at: https://doi.org/10.3886/E192306. Replication code is found at https://github.com/jhsuss/wealth-inequality.

A version of this article was originally published at Nature Communities.

All articles posted on this blog give the views of the author(s). They do not represent the position of LSE Inequalities, nor of the London School of Economics and Political Science. 

Image credits: Marco Scisetti via Shutterstock.