Not Your Demographic

Assertions so counter intuitive and nonsensical - they must be true!

Cogent insights by Benjamin Scherrey

Prev Next


LACK OF TESTS HAS MADE INCIDENTS COUNTS NOT JUST WORTHLESS - BUT DANGEROUSLY MISLEADING! It's scary when your projections come in waaay under the margin of error so that it appears you are psychic. Inexperienced stat people declare, "SUCCESS!" The rest of us go... "oh crap - we failed." Well people - we have failed in a major major way. And it's 100% due to the lack of testing capacity, that's due to the FDA, and the one person who could have fixed it but didn't is President Trump. This is one I lay at his feet. (Watch the left get confused as to whether to love or condemn me now.)

Last night I was engaged in a debate on Nicholas Palevsky's post about when the US would surpass 1 million official cases. Lots of people were saying definitely this week. A friend, Adam Nyback, said the 28th (ten days - he just drew out the linear line), and I projected not until May. Then I looked at my numbers again and realized I made a simple perspective error and my declining rate of growth for new incidents was inappropriate for this projection because that was a day on day figure which means it could be completely influenced by the test positive % and the number of tests performed each day. To check this I ran a new correlation between these numbers and bingo - it was perfect. What does this mean? It means our number of infections (incidents) number is entirely a function of the test positive rate (which has been steadily 20%) and the number of tests conducted each day. That means if I want there to be 1 million positive tests, I just need to get 300k more positive tests (we're at 700k now) which means I need to perform 1.5M tests - at 150k tests/day that means 10 days. If my test capacity was 300k tests/day I'd hit the number in 5 days.

This can only happen only if there are WAAAY more positive cases out there in the population than we have the ability to count. IT HAS NOTHING TO DO WITH HOW MANY POSITIVE CASES ARE OUT THERE - ONLY HOW MANY TESTS WE CONDUCT! THE NUMBER IS A LIE. Until our rate of testing outpaces the rate of infection, WE HAVE NO IDEA HOW MANY CASES THERE ARE. Any statistic that uses confirmed cases as part of it's model is therefore completely wrong and misleading. Guess what. ALL of our statistics depend on this number except for counted dead. And, of course, the CDC has now opened us up to fudging these numbers too. It's maddening! What we do know now is that the number of actual infections is WAY higher than we've detected and this makes the Standford result of 20x - 50x higher completely believable and likely. Punch line - the ONLY way to know the truth is to do what I've been screaming for since March 19th (feel free to go back and look).

This phenomenon also completely explains why our death % stopped declining right after our rogs peaked on March 19. The number of real (undetected) cases grew faster than our ability to run tests. We should have gone below 1% chance of death on March 28th but instead the number started climbing up. Didn't make any sense to me and lots of people thought it was the lag time between detection of the infection and the time til death. But most people who died from it were only confirmed cases a few days before dying, not 14-20. Looks like the truth is that we just stopped being able to count the new cases and so the denominator of our chance of death slowed down as the number of deaths kept rising. The reality is that your chance of death if you get Wuhan-virus exposure is now way below 1%. We just don't know how much lower it is. Can you see why I'm mad now?

Sooo.... here's the numbers (which are wrong but we know why):

As we completely predicted (I use the word projected when the data can be trusted - it no longer can), both rates of growth (rog) for new incidents and deaths are significantly lower and will continue to drop until Tuesday. (yawn) Given a total of only 140,058 total tests completed in America, our new total for incidents is 722,182 (+27,662 ), down to an increase of 3.98% over the previous day (-0.73% from yesterday). Deaths have totaled 34,139 (+1,774) which is an increase of 5.48% (-1.35%) over the previous day. Since the peak in Mar 19, rates of growth for incidents has fallen by 47.65% and for deaths has fallen by 37.38%.

Now here's some REAL interesting data. You've no doubt hear about this Rt Covid-19 model which has been what people have (wrongly) been using to create their scary exponential models that caused our economy to be destroyed. I have talked many times about how this model is fundamentally flawed as a national metric. Well - someone has actually done it right and applied it on a per state basis. And rather than try to project a nonsensical exponential cases/deaths count - they've created a really good metric to determine whether the incidents of Wuhan-virus will go up or go down in each state (a per-county model would be even better). They're using the same data source as me which is promising as well. Here's the graphs:

Basically, if your state is below that pink line representing 1 - then cases will be in decline in your state and so long as it remains there the virus is under control and will disappear. If you're above that line then you can expect the virus to continue to grow at some level related to how high above the line you are. Guess what - about 40% of states are already below 1. Even better, if you scroll down to look at the trend for each state, 2/3rds will be below 1 in a few days.

Wonderful news, right??? Well, not so fast. This value depends on the rate of new incidents and total confirmed cases - which we now know is completely bogus. The question is, and I haven't had the time to sit down and go through the model to find out, does LOTS MORE cases make these numbers look better or worse?

Tune in tomorrow, same bat channel, same bat time (or thereabouts) - for the exciting conclusion to this tale! POW!

Prev Next