Thread by @TheLawyerCraig

Yes I was aware you broke out the data separately, as this tweet stated in the same thread to which you replied:

https://twitter.com/TheLawyerCraig/status/1369663456748703750?s=20

https://twitter.com/kissane/status/1370579552330649600

But certainly CTP is and was aware that legions of people used its API/CSV data, and those that did almost invariably used the two columns circled in red. A simple, “watch out, this isn’t a good comparator” doesn't stop the world from using data derived from these columns.

I don't get why cases/tests didn't match up (either include probable & antigen or just confirmed & molecular) in the columns that pretty much everyone used (and which everyone knew that everyone used). I get that cases and tests are separate beasts, but we know people pair them.

This is a lighter version of the "people tested" issue that led to media firestorms at times over why some states were showing 30%+ positive testing percentages when that was clearly not the case. CTP sorted that out well, but it's not like you guys could shut down twitter.

So on these dubious claims raged, echoed over and over, even after you had the great "people" vs. "encounters" vs. "specimens" story.

And by the way, I was talking about this issue and updating my data for better comparisons prior to CTP swapping over. But I'm just one guy.

By the way, I highlighted GA in the picture above of CTP's final state-level CSV (which I'll include again here) to show how different the numbers would be if one adds in the 1.5M+ antigen tests (or removes the almost 200K antigen-positive cases). It changes completely.

CTP saying "% positive doesn't work because states do not report equivalent data" doesn't stop people from using the data to try and glean that information as best they can—and perhaps base policy on it. How many stories/blue-check tweets did we see on % positive in FL? Many!

So my issues when I realized this were twofold: (1) I assume CTP knew that the twitter, chart-making, media world generally used its data, and that the "positive" and "totalTestResults" columns were primarily those used (in fact, CTP used them for it's daily updates), and...

(2) Given the above, why include antigen cases in the case total but not antigen tests in the test total (for the states that broke out that information) in the columns that everyone used, and that CTP used itself for its daily updates?

Of the top 10 most populous states, the states that appeared more negative than they otherwise ought to have were Texas, Florida, Ohio, and Georgia.

Quite a bit of negative ink spilled on many of those states, and some relating to this very issue.

But this wasn't some imputation of malevolence. It was more of an "Oh shit" moment when I realized that the hefty amount of antigen tests in certain states would keep their floor way too artificially high in terms of % positive. I didn't catch it until I wrote that thread.

And those tests became a larger and larger amount of certain states' tests, like Texas. So probable cases kept on being added to the case counts, but not the increasing type of test delivering them. Here is Texas's graph as of today showing tests by type:

To me, it's a bit too much head-in-sand to think that admonitions against calculating state-level % positive ("this is irresponsible") would stop the world from doing it, as was evident if anyone took a scroll through twitter any day in the last 350.

Sorry for the long thread. And sorry if I came off as accusatory. It was supposed to be bewilderment, and at least as much at myself for not realizing it.

CTP was maybe the most important account/project of this past year, and I'm grateful for all your teams' efforts.

*Initial tweet didn't quote the way I intended. Here is the tweet recognizing CTP broke out the data separately, but that the columns used by damn-near all (including CTP) were mismatched, in my opinion.

https://twitter.com/TheLawyerCraig/status/1369663456748703750?s=20