Lessons From the Processing and Sharing of Public Data Sets for the Study of Structural Racism

TL;DR

This paper discusses challenges in using public data sets to study structural racism and introduces tools to standardize and share these data.

Contribution

The paper introduces standardized metadata and data transformation pipelines to facilitate the study of structural racism using public data.

Findings

Public data sets for studying structural racism face issues like recency bias and missingness.

Standardized metadata and transformation pipelines improve data usability for researchers.

Creating a public repository helps address challenges in open science practices.

Abstract

As part of a larger study of structural racism, we collected 50 publicly available data sets containing geographic measures. These data sets covered over 100 years of history, six geographic units, and nine domains of inquiry (civics, credit/income/wealth, education, employment, environment, healthcare, media/marketing, neighborhoods, and policing). Structured metadata about each data set were compiled and used to standardize data files with shared conventions, allowing analysts to combine data files to study structural racism. To allow researchers to assess the potential value of these data in relation to their areas of inquiry, we created dashboards that summarize key measures in each data set, including the geographic level of measurement, the years covered by the data, and the extent of missingness. This process made clear several problems researchers seeking to use public data files to study structural racism will face, such as recency bias, temporal and geographic mismatch of data sources, and data missingness. We also developed data transformation pipelines to process each data file to follow shared conventions that allow reliable linking of data files across spatial units and time periods. To help outside researchers to use these data in their own projects, we created a public repository of structured metadata about the form and content of data files, accompanied by tools to allow users to procure data files, process them, and analyze them. This effort provided insight into the challenges faced by researchers trying to follow best practices with respect to open science, particularly around data licenses and data custody.