These two stories used excess deaths analysis to show that the winter storm and power outages that hit Texas in February and the extreme heat wave that hit the Pacific Northwest in June caused hundreds more deaths than we counted for by the affected states. The second story broadened the narrative to show that the failure to accurately count deaths from extreme weather is a general problem that is leaving government officials ill-prepared to respond to climate change and reduce its toll. It also discussed reforms in data collection, analysis, and disaster planning that would improve resilience against the single
The first story story drew widespread coverage from local and national media outlets, and prompted state Democrats to demand an investigation. This was a largely avoidable disaster: After an earlier winter storm in 2011, federal regulators warned Texas that its electricity grid, which is run independently from the two large interconnections that serve most of the rest of the nation, was vulnerable to failure in extreme cold.
Bills to better prepare the Texas grid to handle extreme cold were already in process. They were passed by the Texas legislature shortly after our article appeared and were signed into law by Republican Gov. Greg Abbott in June. However, critics argue that these measures fall short of what is required. The issue has moved to center stage in the 2022 Texas gubernatorial election, with the Democratic challenger, Beto O’Rourke, citing our reporting in his attacks on Abbott’s record.
The CDC maintains counts of weekly deaths, for all and certain selected causes, at the national and state level, for 2014 to 2019, and for 2020 and 2021. To estimate expected deaths for each jurisdiction and week, not including deaths from COVID-19, we trained regression models on the 2015-2019 weekly deaths data, accounting for long-term demographic changes using a linear component and using a smoothing spline to account for seasonal variation. We then used these models to predict the expected number of weekly non-COVID deaths for each state, with 95% confidence intervals.
For each jurisdiction and week, we calculated weekly death anomalies minus deaths for which COVID-19 was given as the underlying cause, again with confidence intervals. We found significant spikes in non-COVID deaths in Texas in the week ending Feb 20, immediately following the winter storm and when the power outages occurred, and in Washington and Oregon in the week ending Jul 3, at the peak of an unprecedented heat wave, that greatly exceeded the official tallies in the affected states. Three independent academic experts in excess deaths analysis reviewed the findings, methodology, and analysis code.
We also looked in more detail and individual causes of death in the CDC data, finding notable spikes for cardiovascular disease and diabetes in Texas for the week ending Feb 20, and for Washington in the week ending Jul 3. The analyses were performed using R and RStudio. Graphics for the stories were made with Datawrapper and ggplot2 R package, showing the time course of the power outages by county in Texas and neighboring states by animating frames using ImageMagick.
What was the hardest part of this project?
Finding family members of those who died in the Texas winter storm to reveal the harrowing human stories of these uncounted deaths. Aldhous looked for specific causes of death listed in the CDC data that were associated with the overall spikes, which flagged cardiovascular disease and diabetes as the causes listed on death certificates that were most likely to be associated with the extreme weather. He then filed public records requests to medical examiners in the largest Texas counties, asking for complete lists and cause of death for those who died after the winter storm and during the power failures.
This allowed Lee and Hirji to triage their reporting by focusing on deaths that were attributed to these causes in the records obtained from medical examiners — turning what would otherwise have been a guessing game into a targeted search for medically vulnerable people whose deaths were triggered by the weather extremes. This allowed their individual stories to be told.
What can others learn from this project?
The GitHub repositories with our methodology, data, and analysis code provide a recipe for other data reporters wanting to apply excess deaths analysis to their own work.
Our analysis of specific recorded causes of death, when combined with medical examiner’s records, also shows how data can be used not just to provide a top-line finding, but also to triage reporting by helping to identify individuals affected by the issues raised, emphasizing the human stories involved.