Updated: 07 Mar 2022
Author: Marc Bevand
This repository charts cumulative excess deaths per capita for each US state, during the COVID-19 pandemic. This is the first analysis that we know of that presents excess deaths broken down by age group. We present two sets of charts:
- Overall excess deaths, based on the US CDC's estimate of excess deaths (source code: all_ages.py)
- Excess deaths by age group, based on our own model of expected deaths per age group (source code: by_age_group.py)
Long-term excess deaths are defined as excess deaths after 2020-04-26 (roughly the peak of deaths of the first wave in Spring 2020). Long-term excess deaths are an indirect result of each state's long-term plan and interventions against the pandemic:
The charts below are based on our own analysis that models expected deaths by age group.
Raw numerical output of our model: by_age_group.csv
As this CSV shows, our model produces figures close to the CDC. As of week ending 2022-02-19, the CDC estimates 1,078,226 excess deaths, while our model estimates 1,064,325, a difference of only -1.29%. Our model tends to produce excess death estimates slightly below the CDC's estimates because their model ignores deficits of deaths ("Negative values, where the observed count fell below the thresholds, were set to zero").
We take deaths reported by the US CDC, per state, per week (column "Observed Number" where "Type" is "Predicted (weighted)"). The US CDC weighted these figures to account for reporting delays. This affects only data from the last few weeks. In our experience, weighted vs unweighted has little impact on the charts above.
For each week we calculate the difference between reported deaths and the average expected number of deaths from the US CDC (column "Average Expected Count"). The difference can be positive or negative.
We add up these differences to calculate the final cumulative excess deaths per capita.
We use states population from the US Census Bureau 2019 estimates.
We analyze the weekly count of deaths by state and by age group. This dataset contains data beginning on the week ending 2015-01-10. For each state, for each age group, for each week of the pandemic, we calculate expected deaths by modeling both seasonal variation and year-to-year trends: we gather deaths that occured on that week of the year during the years preceding the pandemic, then we perform a linear regression to project expected deaths in the future.
We adopt the same convention as the CDC: we assume the pandemic starts on year 2020, MMWR week 6 (which starts on 2020-02-02).
The per capita figures need detailed population information by state by age group, which we obtained from the US Census Bureau 2020 estimates, file Single Year of Age and Sex for the Civilian Population