After several battleground states reinstated the right to vote for people formerly incarcerated for felonies, we undertook a complex data investigation that revealed that no more than 1-in-4 of them registered to vote in time for the 2020 election. We used text messaging to directly engage with people who were newly able to vote in Nevada, Kentucky, Iowa and New Jersey. Many of the people we spoke with didn’t know they were eligible to vote; our investigation showed that states do little to notify them of their restored rights.
This reporting opened the door to a legislative resolution in Nebraska to study voting among formerly incarcerated people. As the resolution states, “data and system errors have impermissibly disenfranchised eligible voters from participation in the election process.”
The resolution exists in large part because of Nicole Lewis’s nuanced reporting in the story, and Andrew R. Calderón’s detailed description of how to unearth and contextualize such data that demonstrated to civil society groups like the ACLU that such a study is feasible and explained how to conduct it efficiently.
The nature of the system means impact will be slow to materialize, but as the pending legislation in Nebraska shows, we believe that over time the exposure of the ways voting participation is still stymied can affect tens of thousands of people.
Our story was co-published by the Louisville Courier-Journal and USA Today Network, appearing in the Des Moines Register, the Reno Gazette-Journal and several Gannett newspapers in New Jersey. That helped bring the issue to the forefront nationally and in the states we analyzed.
The project also raised awareness of the issue through media appearances on CBS News, NBC/Peacock on Zerlina Maxwell’s show along with two more NBC Now appearances, and NPR and WBUR’s Here And Now. It was also covered by Politico and Talking Points Memo.
- Entity matching: We used Python to join the datasets. Early iterations of the project used natural language processing tools like Dedupe, but we were happily able to simplify the problem to the point where we could use more straight-forward techniques.
- Surveying: We used Typeform to design and build a custom survey to embed on our site.
- SMS (text messaging): We used the Twilio API to run our direct survey of formerly incarcerated people in Kentucky.
- Cloud computing: We tracked and logged our SMS survey using Docker containers running on Amazon Elastic Cluster Service (ECS). We used Terraform to manage the stack, allowing us to quickly deploy serious computational resources but only pay for what we actually used and avoid long-term maintenance debt.
- Observable notebooks for sharing results with the editorial team. These included editorial analysis, but also helped show the status and results of the SMS survey for internal use.
- D3.js to visualize the percentage of potential voters who registered.
What was the hardest part of this project?
Our nuanced approach to understanding the data behind this project made this project unusual for a newsroom. Despite the fact that nearly every state purges people from the voter rolls once they go to prison, few states keep track of how many formerly incarcerated people re-register once they are released. This makes assessing the success of the re-enfranchisement laws incredibly difficult. We had to figure out how to do our analysis with imperfect and ambiguous data, using only publically accessible records.
We developed a methodology for joining voting records with release records that was conservative but accurate (“at least one in four”). By using complex logic based on release date, age at release, and name, we were able to set an accurate but fair minimum value.
Another challenge was building relationships with formerly incarcerated people who had been re-enfranchised. When we did our survey, we got negative responses from people who it seemed didn’t want to talk about their experience with the criminal justice system. There is a stigma associated with incarceration . Upon release, people often want to focus on re-entry rather than their time behind bars. Fortunately, we came in contact with a source through our SMS campaign in Kentucky who was willing to share her story, and her anger that the Kentucky government had not done more to inform her that she had the right to vote in the 2020 election.
What can others learn from this project?
Generally speaking, this project demonstrates that entity resolution problems in datasets from multiple sources can sometimes be solved with simple techniques available to many newsrooms. It also demonstrates how journalists can answer a question that the government has not answered using the government’s own data from multiple agencies. And it shows how journalists can use release records and data about prisoners to answer questions about a topic like voting, and similarly how voter rolls are a powerful source of useful information for many kinds of journalistic inquiry.
More specifically, journalists can learn from Andrew R. Calderón’s open methodology on how to measure voting patterns of people released from prison. For practical reasons, we did not consider all states that re-enfranchised people with felony records. In addition, the analytical techniques we used are not limited to states that recently gave people with felony records the right to vote – these techniques could be applied to other questions about the voting patterns of formerly incarcerated people.