An Unemployment System Built to Discriminate

Country/area: United States

Organisation: Bloomberg

Organisation size: Big

Publication date: 19/11/2021

Credit: Authors: Shawn Donnan, Reade Pickert and Madeline Campbell Graphics: Rachael Dottle Editors: Robert Friedman, David Ingold, Ana Monteiro and Mira Rojanasakul


The journalists responsible for this story are reporters, editors and data journalists at Bloomberg News. Shawn Donnan is a senior economics writer; Reade Pickert covers the U.S. economy; Madeline Campbell is a New York-based data journalist, and Rachael Dottle is a graphics journalist in New York. They were assisted by data journalist David Ingold. The project was edited by senior investigations editor Robert Friedman with assistance from U.S. economy editor Ana Monteiro and senior graphics editor Mira Rojanasakul. 

Project description:

An analysis of more than 2 million unemployment claims obtained through an open-records request to Georgia’s labor department found Black workers in the state who lost jobs during the pandemic were more likely than White ones to be denied unemployment benefits and suffered disproportionately when the state withdrew early from a temporary federal program.

Impact reached:

Bloomberg’s findings prompted state legislators to call for an investigation by the U.S. Department of Justice. Competitors at the New York Times and Washington Post called it “important work.” It was widely cited also by experts on the U.S. unemployment system, some of whom have asked for access to Bloomberg’s data. While many news organizations documented problems during the pandemic with overloaded state unemployment systems, the work by Bloomberg was the first to identify racial disparities in the distribution of benefits. What happened in Georgia was emblematic not just of that state’s polarized politics but of a system born in the New Deal that to this day often discriminates against low-income workers. 


Techniques/technologies used:

The project was the product of old-fashioned shoe-leather reporting, beat expertise, and sophisticated data analysis and visualization. It started as a nationwide effort, in which Bloomberg reporters reached out to all 50 states and D.C., filing open-records requests when initial queries were ignored or rejected. The data used in the story was obtained through a request that Georgia officials were initially inclined to reject, as other states had. But Bloomberg reporters convinced state officials to agree to share anonymized data not usually released publicly. After many updates to correct errors and inconsistencies, the team analyzed millions of rows of data using the Python programming language. Our analysis – which identified disparities in denial rates and denial reasons across ethnic groups – was performed on both regular unemployment insurance and pandemic unemployment assistance data. These denial rates were mapped at the ZIP code level alongside Census and Bureau of Labor Statistics data on population and workforce demographics to confirm findings and identify reporting targets. No unique identifiers were available to link applicants to regular state unemployment with their application to the Pandemic Unemployment Assistance program. The team devised logic which was abstracted into a continuous flow which approximated the volume of applicants in GA who ultimately received no benefits from any program. The team created a custom scrolling graphic using code to visualize and explain that flow as well as static graphics made with R and Adobe Illustrator. Maps made with code and Mapbox of rejection rates and the Atlanta Black population show the association between race and unemployment rejection. 

What was the hardest part of this project?

The most challenging part of the project was obtaining data not usually shared publicly by state labor agencies on the demographics of people who are rejected for unemployment benefits. The Bloomberg team initially set out to collect that data from all 50 states. Almost all of them rejected that request outright. Some shared high-level aggregate data. Others, like New York, to this day say they are continuing to process the request. It was only through persistent efforts that Bloomberg was able to secure the data from the Georgia labor department. Even after that data was obtained, many rounds of quality checks required going back to the Georgia data team, which conceded there was an error in their original query and produced a corrected data set. 

What can others learn from this project?

Even in a world rich in data the best stories often lie in the data sets public agencies are not sharing with their citizens. As journalists we have a responsibility to push local, state and federal government agencies to release more data and to fight for greater transparency.

Project links: