Thread by @TheLawyerCraig

*Updating #COVID19 Data*

I just realized this morning a large error in several states in the way that @COVID19Tracking (and now our team) reported many states testing totals compared to the way it presented cases.

It's another denominator problem, folks.

🧵🧵🧵

When available, CTP (and our team) reported probable cases as the preferred metric, which is defined as follows:

Poking around on the state dashboards, the overwhelming majority of probables come via positive antigen test.

But when it comes to reporting tests, we (and CTP, in its list of test totals that everyone uses) exclude antigen tests!

The result? Fewer tests. Higher % positive.

Take Texas, for example. It's been conducting 20K+ antigen tests daily for months now. We've been counting every single positive case (as a "probable" case) that came from those tests. But we haven't counted a single one of those tests! That's no good. And we're going to fix it.

Just at a glance, there are several other states that conduct a good number of antigen tests that don't get reported (while the positive "probable" cases do): AR, GA, ID are a few that I saw just quickly.

Some states, however, go the other way...

For California, we only use confirmed cases, no probables. But for tests, CTP's last update said this about California's testing number:

So, it's possible that California's numerator is the lowest number while the denominator includes everything. Result: lowest possible positive testing %.

*Note, just because it's unclear doesn't mean CA is using non-molecular tests in its running total.

The result will be more tests than would have otherwise been shown. For states that *do not* report probable cases, we will remove antigen tests from the testing number to the extent the dashboard provides that level of nuanced data (again, we want apples-to-apples).

For the life of me, I don't know why it ever made sense to report probable cases but not the tests that yielded them.

(Yes, CTP broke out the data, but the API/CSV and everyone who used the data used total cases and molecular tests as the preferred numbers)

Starting today, our team will make sure it's apples-to-apples comparisons. I'm not just going to dump every state's antigen tests into the total, since that would give us ridiculous numbers (we'd have millions of extra tests).

But we will make sure numerator & denominator match.

*Update*

I just checked, and this is going to make quite a difference. It affects 17 states to varying degrees. We will see the largest change in the Southeast, where several states have tons of probable cases being counted & antigen tests that weren't. TX & FL will change too.

Again, I'm not backdating the data, so we'll only see the change going forward. But a week from now, we'll have filtered out all the old data. So while comparisons between states will be better, comparisons between time periods won't be as good (better % positive starting now).