The COVID Tracking Project

» see covidtracking.com as it lived, on the wayback machine

~ the obituary ~

In early March 2020 two journalists at The Atlantic, Robinson Meyer and Alexis Madrigal, went looking for a single official source of United States COVID-19 testing data for a story they were reporting. There wasn't one. So they started building a spreadsheet. At roughly the same moment the data scientist Jeff Hammerbacher was assembling his own version of the same thing, and on 7 March 2020 the two efforts were merged and thrown open to the public.

What grew out of that spreadsheet was the closest thing America had to a national pandemic dataset for most of a year. The method never got clever. Every day, people read state health department webpages and emailed the officials behind them when the pages disagreed with themselves, then typed what they found into a spreadsheet. The project published counts of cases, tests, hospitalizations and deaths for all fifty states, and for the District of Columbia and five territories besides. Racial and ethnic breakdowns came later, along with a separate tracker for long-term-care facilities.

It ended up with about thirty paid staff and somewhere between 250 and 300 active volunteers, with Madrigal leading it and Erin Kissane as managing editor. An API went up so other people could build on the numbers. The data turned up in more than 80,000 news reports and around a thousand academic papers, and it was cited by federal agencies and by two successive administrations. In June 2020 the CDC published a report noting that the volunteers' race and ethnicity data might be more complete than the agency's own, which is an unusual thing for a government body to put in writing.

On 1 February 2021 the project announced it would stop. The reason given was that the official data had improved enough that a hand-typed volunteer dataset was no longer the best source available, which had been the point from the beginning. The final daily update went out on 7 March 2021, exactly one year after the merge that created it.

~ the case file ~

~ typed by hand ~

The thing worth remembering about the COVID Tracking Project is how unsophisticated it was. There was no scraper farm and no federal feed. There were people, reading government webpages that changed format without warning, arguing about whether a state's Tuesday number included Monday's backlog, and entering the result in a spreadsheet. Some states published a PDF. Some published a dashboard that overwrote yesterday's figure with today's, so if nobody wrote it down at the right hour it was gone.

That fragility is why the project also kept a data log and a set of state-by-state reporting assessments, essentially grading each state on how legible its numbers were. Those pages read now like a survey of how badly a rich country can publish a number it already has.

~ one table ~

This is a problem Turkey did not have in the same shape. The Sağlık Bakanlığı published one daily table for the whole country from the start, so there was never a gap for volunteers to fill by reconciling eighty-one provinces against each other. Whatever anyone thought of the figures in that table, and plenty of people argued about them, the argument was about a single published source rather than about the absence of one. The American problem was federal by design: fifty-six jurisdictions each deciding on their own what counted as a test.

So the COVID Tracking Project is a thing I read about rather than used, and what struck me reading it was that the missing piece was not analysis or modeling. It was somebody agreeing to write the numbers down in the same place every day.

~ the archive that survived ~

covidtracking.com is still up. It serves the whole dataset and every analysis post, with the charts and the API documentation still in place. A banner explains that collection stopped. On 29 July 2021 the University of California, San Francisco and The Atlantic announced that the project's archives would enter the university library's permanent collection, which means the thing has a curator now. Most of the sites on rip.so died because somebody stopped paying a bill. This one was switched off on schedule, by the people who built it, on the day they said they would.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »