Data Critique

Opportunity Atlas and Mapping Inequality: Our Thoughts on the Value-Add and Limitations

We have utilized two primary datasets: "Redrawing the Lines: Mapping the Legacy of Redlining in U.S. Cities": the Opportunity Atlas dataset, as well as Home Owners' Loan Corporation (HOLC) redlining maps from the University of Richmond’s Mapping Inequality project. The HOLC redlining maps are digitized scans of city survey files from 1935–1940, kept at the National Archives and available under a CC-BY-NC license; Opportunity Atlas is a joint collaboration between the U.S. Census Bureau and Harvard University. 

Opportunity Atlas includes data from the 2000 and 2010 censuses, in addition to anonymized data from 1990s income tax returns and American Community Surveys (ACS) launched between 2005 and 2019. In the first bucket of data, “Neighborhood Characteristics,” the researchers included key neighborhood economic statistics, as well as demographic information. The second bucket, “Children’s Outcomes in Adulthood,” includes indicators of childhood mobility within neighborhoods, like incarceration rates, employment and college graduation rates, income distributions, and incomes of children staying in the same commuting zone they grew up in, as well as comparisons between immigrants and American-born residents.

Mapping Inequality includes both HOLC georectified maps with color-coded risk grades and textual “area descriptions” recording explicitly racist and classist justifications for disinvestment. These maps categorize specific regions of major American cities on an A to D scale based on “residential security,” or how beneficial it would be for a bank to provide mortgages to particular residents, with A-grade neighborhoods being more likely to receive loans. 

The HOLC data prop up our analysis, motivating in-depth understanding of our research question. In conjunction with the Opportunity Atlas data, Mapping Inequality helps us trace the origins of disparities, matching outcomes to neighborhood characteristics and highlighting patterns of infrastructural neglect. For instance, in Oakland, a quick analysis shows that “D” rated neighborhoods by the HOLC closely align with areas now bifurcated by Interstate 980, the new West Oakland BART plaza, or suffering from housing disinvestment. Similarly, in Detroit, historically redlined zones are associated with higher foreclosures and lending disparities, while in New Orleans, flood zones from Hurricane Katrina demonstrate how environmental risk and overt systemic racism intersect on the map.

A strong attribute of Opportunity Atlas is its utilization of a diverse array of economic indicators. We are introduced to uncommonly seen metrics, such as hours worked per week, percent staying in the same commuting zone as adults, and neighborhood job growth rate, all with time-based evolution over fixed periods. Fundamentally, understanding regional differences at as granular of a level as a census tract paints a rich, compelling narrative about the evolution of inequality. We are granted flexibility in using rarely-seen indicators to craft temporal and geospatial visualizations, dynamically redefining our narrative as contemporary rather than rigidly historical. 

Despite their ability to bridge historical context and quantitative data, Opportunity Atlas and Mapping Inequality have limitations regarding nuance and explanations. The Atlas maps outcomes, not causes. For example, while visualizations show differences in adult wages, they cannot explain why middle-income Black men have a higher incarceration rate than lowest-income Black men in specific neighborhoods. Primary cohorts’ mobility outcomes are studied in 2005 and 2019, for children born in 1978 and 1992, respectively (Chetty et al., 2019). Thus, trends may not be fully representative of populations impacted by redlining. As for the HOLC data, maps only exist for certain cities, and even within those, boundaries do not align perfectly with modern census tracts, necessitating geospatial interpolation that introduces error and uncertainty. The IRS and census data also exclude undocumented immigrants, which impacts representation, as the HOLC maps show 30% of “foreign” populations, which is a sizable portion of Detroit (Nelson et al., 2018). Both datasets reflect biases in their original collection processes: the HOLC maps were shaped by 1930s institutional racism, while non-response bias and misreporting may exist in voluntary responses to the Bureau-run ACS. 

To account for these limitations, we have input modern-day stories into temporal narration: our interactive timeline includes contextual commentary. As Boyd Davis, Vane, and Kräutli remind us, “there is no neutral visualization: the intention [is] based on a shared understanding of the objectives” (Boyd Davis et al., 2021, para. 35), compelling us to foreground the ethical stakes of our work and to question how we (re)present marginalized communities. We adopt the “seven main principles [of][...] Data Feminism” as outlined by Rezai, especially her imperative that we must “elevat[e] emotion and embodiment [...] making labor visible” (Rezai, 2022, para. 2). Ultimately, we aim to create reliable, neutral, and persuasive commentary to illustrate the consequences of redlining, rehabilitating modern stories ex post with thorough, transparent, and accessible documentation.