2022
Pro-Trump counties now have far higher COVID death rates. Misinformation is to blame
Country/area: United States
Organisation: NPR
Organisation size: Big
Publication date: 05/12/2021
Credit: Reporting/Analysis: Geoff Brumfiel, Daniel Wood. Graphics: Daniel Wood, Rina Torchinsky. Editing: Brett Neely, Alyson Hurt.
Biography:
Geoff Brumfiel is a senior editor and correspondent on NPR’s Science Desk. Daniel Wood is a graphics reporter with NPR’s News Apps team. Rina Torchinsky is a digital news and graphics intern at NPR and a student at the University of Maryland. Brett Neely is an editor for NPR’s Washington Desk, focusing on disinformation, election security and voting rights. Alyson Hurt is an editor for NPR’s News Apps team.
Project description:
An analysis by NPR shows that since the vaccine rollout, counties that voted heavily for Donald Trump have had more than twice the COVID mortality rates of those that voted for Joe Biden.
Impact reached:
To our knowledge, this is the first story from a major news outlet that both analyzes these correlations and gives readers the ability to find their individual county in the data. While each of these sources (elections, COVID deaths, vaccination rates) is widely available, combining these data and making a coherent methodology represents a step forward in discussions about the virus, vaccines, partisanship and misinformation.
This story was the third most-viewed story on NPR.org for the month of December. It was widely shared on social media and reddit. And the analysis was picked up by Rolling Stone, Salon, Politico, NBC News and others. Further collaborations and analyses are in the works.
Techniques/technologies used:
Data was collected from various sources, then joined and analyzed using Jupyter notebooks, pandas, matplotlib and other Python libraries. Once processed this way, the data was brought into the browser, where it was visualized using d3.js. We wrote custom JavaScript to allow for individual counties to be looked up. Two additional charts were created using d3.js and our team’s homegrown graphics rig.
What was the hardest part of this project?
While the data collating and visualization were difficult for these diverse and large datasets, the biggest challenge was coming up with a cohesive, defensible, and straightforward methodology. The quality of county-level COVID-19 death data is notoriously spotty. Since some counties and entire states have ceased sharing new COVID-19 data altogether, it was difficult to know whether records of counties with no or very few new deaths since May 2021 were accurate. None of the options available to address this were very good. We could try to verify individual counties piecemeal, but that would be very tedious and error prone. We could try to patch specific states when alternative data was found, but we tried to limit this where possible due to the complexity, error and bias it adds to the methodology. Finally, we could simply leave this data as is, and understand that it introduces some error to our findings.
Ultimately, we mostly chose the final option. Our reasoning was that since most of the counties with erroneous zeros were a high percentage of Trump voters, including these zeros as “true” would weaken our analysis, and be a more cautious course of action. Thus, if a weaker analysis is still a stunning correlation, we can err on the side of caution knowing that, while not perfect, our findings are defensible.
What can others learn from this project?
I think this analysis shows that it is impactful to apply numbers to our suspicions, to see if they are true. This analysis may be somewhat unsurprising on one level, because it confirms what many of us have long suspected about vaccine-denialism. But it is not enough to suspect these things. If these things are true it is a disturbing indictment on those sowing disinformation, and important for the public record to have it down in black and white. Similarly, other journalists ought to apply data to their instincts, strengthening arguments where correlation is found and dismissing it where it is not.
