Comparison of data sources for COVID-19

UT-Austin COVID-19 Modeling Consortium

We use the New York Times data of confirmed COVID-19 deaths to make daily updates to our COVID-19 mortality forecasts for each of the 50 states + DC.

Please see the technical review (latest version or preprint version) for details of our model.

We have noticed some inconsistencies for mortality numbers between various sources. Because various modeling groups are using different data sources for their forecasts models, we have started this repo to compare numbers across data sources.

Summary of data sources

(as I understand them)

• New York Times (NYT)
  • US data only
  • Reports daily cases and deaths
  • These numbers come from NYT reporters
  • These numbers are confirmed cases and deaths, and are revised as needed
• Johns Hopkins University (JHU)
  • US and international data
  • Posts daily updates of cumulative deaths
  • Data are collected from government sites
  • Not entirely clear if these are entirely confirmed cases, or confirmed + probable

Note the big spike in daily deaths for the JHU data for New York State for April 16-17, which I believe is due to New York City revising their COVID-19 death numbers around April 14 to reflect suspected / probable deaths (see New York Times article), whereas the NYT appears to have strictly confirmed deaths.

Contents

• nyt-jhu-comparison.R compares daily incident deaths and cumulative deaths from NYT and JHU, aggregated at the state level. I restrict comparisons to dates after which a state has exceeded a per-capita mortality rate of 3 per 10 million residents. Note that, due to format differences in reported data, I have to convert from daily deaths to cumulative deaths for NYT, and vice versa for JHU.
• figures/ contains figures comparing numbers of daily deaths and cumulative deaths between data sources for several states. Code for these figures can be found in nyt-jhu-comparison.R.
• data/ contains misc. data for states and other geographic aggregations.

Contact

Please feel free to make a pull request or contact me directly to correct possible errors I've made.

Spencer Woody
The University of Texas at Austin
[email protected]