Thread by @TheLawyerCraig

Ok--I said that I'd provide some more information on the issue of how I (and others) have reported testing numbers wrong in several states.

This is actually a pretty big deal, and could affect policy decisions with a cursory look at the wrong positive testing percentage.

1/x

People looking at the national snapshot often use @COVID19Tracking's data. I do. Their site provides easy-to-access data.

And they have updated testing data for several states in a new column on their spreadsheet, but many of us continued to use the "legacy" column.

2/x

For 24 states, using that column provides a *very* skewed version of daily testing numbers (some states more skewed than others), creating erroneously high *current* positive testing percentages.

3/x

First, here are the 24 states:

CO, DE, FL, HI, IA, IN, KS, LA, MA, MD, MN, MO, ND, NE, NH, NV, OR, PA, RI, SD, UT, VT, WI & WY.

Many high % states here, but also some low % states that would be even lower (e.g., MA, NH, VT)

4/x

Here's the issue: this data creates a tabulation of daily tests excluding every person who has ever tested previously. At any time. Not the same day--ever.

That's a big problem. Why? You have an ever-shrinking pool of people who "count" as a test as more people get tested.

5/x

I'll use an example: North Dakota.

My data as of Friday showed ND with a 7-day average percent testing positive of 28.12%. Holy hell, that's high, right?

Well, let's go to their state's page:
health.nd.gov/diseases-condi…

6/x

Take a look at the snapshot of their testing. We have:

5661 tests processed (5444 suitable encounters)
1548 first time testers
343 positive cases
144 of those are from repeat testers
199 of from first time testers (343 - 144)

7/x

Most of us who use @COVID19Tracking's data reference the day-over-day change in positive and negative cases, and use the combination thereof to get total tests.

Here is where that information lived on their spreadsheet, which they update daily here: docs.google.com/spreadsheets/u…

8/x

Over time, those numbers have become poor for current test counts in the 24 states that stop counting a person's test after their first one. These states essentially are just tracking how many of their citizens have ever gotten a test.

Back to North Dakota for why...

9/x

Using those two columns, North Dakota would show 343 new cases (the *total* amount of new cases), but only 1548 tests (the amount of *first time ever* testers).

So ND's % for today is 6.30% (343/5444), but I (and others) would report that as 22.16% (343/1548)

10/x

That's a massive difference, but in a pretty small state, so it's fairly hidden nationally. Over 30% of the state has been tested once though, and will never be counted again!

Also, this isn't on @COVID19Tracking--it's on some pretty poor data presentation by the states.

11/x

Fortunately, @COVID19Tracking added a column that fixes this issue for 16 of the 24 states, and that's what I'll be using going forward (and correcting retroactively).

For the other 8, however, I'll manually update 3 of them (FL, WI, KS).

12/x

But 5 states *only* show the change in *first time ever* testers (as far as I can tell): Vermont, Oregon, Louisiana, Delaware, and Iowa.

VT is so low that it won't make much difference in the NE anyway. And at least the other states are all in separate regions of my data.

13/x

Starting with today's thread, I will be using the new data, and you'll notice some changes.

My "Great Plains" region, for example, was absurdly over-inflated, since all seven states had this reporting issue!

14/x

One final note: the overall US trends are don't change *that* much with the shift. 8 of the 10 most populous states were already being reported the more accurate way, including states conducting high levels of testing (CA, NY, TX, IL).

But they do change some...

15/x

For example, with this updated data (even with the 5 lost-cause states hurting), the US percent testing positive yesterday was 4.36%.

And now (as of yesterday's reporting), we showed a week-over-week drop in percent testing positive for 64 of the last 65 days.

16/x

Thanks to all of you who follow me or engage with me for helping me figure this stuff out. Folks poked me about FL, CO, and other states when seeing the discrepancies in positive testing between the state websites and my data.

You've helped me fix this. I appreciate it.

17/x