It's not.
First, chuck the "case" number. It's meaningless except in a really constrained sense--and even then it's not the positive case number that matters, but the ratio. Do 10x the tests, you'll really increase the cases. It's not that the case # actually increases--it's your monitoring of it. It's like counting the words a talk-show host says. Record 30 seconds and count, you get one number. Record 2 minutes and count, you get a different number. It's not that he talked less or more, it's that you counted less or more. Divorce measuring from the measured.
Better--look at COVID hospitalizations and deaths. Esp. with the new revised way of counting dead people. Yes, some won't show up--but they get worked in. If you have access to cumulative hospitalizations, the number's 1 in 4 or 5 hospitalized patients die, and it's been pretty consistent for the two weeks across numerous states. Usually 1-in-5 means "low death toll" and I suspect the doctors have lots of capacity and don't need to be prudent with assigning hospital beds. Or maybe they need the income. Dunno.
To the curve ...
Some places are cresting or peaking. Others are in decline. Some are increasing. That means you're not getting a good / but the results of a lot of little ups and downs and flats all added together. Remember that--you add all the little trajectories to get the big mother of a trajectory.
Think of it terms of the world. Italy's decreasing, the US is increasing, some countries are in that funny random-hits stage before the curve really takes hold, some are post-wave 1. The world's crest is ... Depends on all the projections of the individual countries. It makes no sense to speak of Vietnam and India and the US and Chile all in the same breath.
Sadly, the details matter. Or maybe I've just been looking at too much math with sums and indices while drinking. But I think the details matter.
IMHE had its predictions, and it showed a crest based on the timing of all the little projections. There *is* no overall projection that isn't a composite of lots of little ones. If you spread out the timing of the crests, you start getting a plateau or a gently sloping line. Time them all to occur at once, you get a really sharp peak. (Want fun, learn Fourier analysis. You can use a whole bunch of sine curves to give a flat line. It's a pain. But doable. In this case, it's just for a while.)
In other words, the US isn't plateauing--that's an emergent feature when you add all the little bits together. Like a Lego kit. Don't think of the US as a single entity. For this purpose, it's not. It's also not red/blue. Or just rural/urban. It's complicated. Simplisticate it all you want, but when you're done what you have is simplistic: Simpler than possible.
This isn't to say I'm not bothered by the extension of the curve, but I'm not as much as I would be for two reasons. First, because I know the model was subverted, right at the time that some big states were to have peaked. Gee--instead of declining, we count differently so that the numbers go up again. The IMHE model is based on mapping real data to a modified curved derived from Wuhan/Hubei. The modifications are things like the predicted effect from social distancing and such--and there's feedback, as social distancing produces a wildly different result, they go back and alter their assumptions. Theoreticians hate it--it's not mathematically pure, it's corrupted by data-based feedback. (Oh. The horrors. Quick. Hide. And is that bottle empty?)
When they altered the basis for counting corpses they changed things in a way that I'm sure is a b---h to work in. Meaning that they're going to be tinkering some more. They underestimated the death count on the 4/15 run, and it'll take a while to work out as *more* states update their death totals. If they do. (Assume they will, assume they won't--screwed either way.)
The second reason is that the summed peak for the US was much lower than predicted. Why? Because the curves were splayed out over a longer period. Lower the peak, drag out the lowered peak. It's nice to think that the lower peak meant fewer deaths overall, and IMHE came in over the real value more often than not. But it didn't mean that the lower curve meant a lower overall death toll. It was a possibility. But people jumped the gun in the interpretation (I was hoping, but think I held my keyboard.)
There's also noise--there's always noise. Deaths lump up for whatever reason--they're mixed in with a bit of randomness, and random is lumpy. If you flip a coin and it comes up HTHTHTHTHTHT... you know you're in an impossible universe. Heads/tails comes up lumpy: HTHHTHTHTHHTTTHHTHTHHT. So don't expect the death count to *ever* map really well to any predicted curve unless it's something well studied and the numbers are large. (No, 800 isn't large. At least not large enough for this.) And the reported deaths are when they're reported: Sundays are slow death days in some states not because death's observing a Sunday work-slowdown but because they don't get reported until Monday. So you have to have at least a 3 or 4 day moving average to get rid of the random and systemic error. You can do the math or you can squint and hold an index card against your screen.
Here's the rub and where all the studies on how progressives/liberals are so much better at handling uncertainty than conservatives get tested. IMHE and other models have uncertainty indicated. For IMHE it's a laughable uncertainty, to be honest. There were 45 deaths today, and tomorrow they predict 39 deaths, +500/-45. Ooh. Hard to be wrong with that. So just accept uncertainty--yeah, lives and livelihoods depend on it. But if you can't accept uncertainty when you're under the gun, it's like a rich person's faith in god to provide his/her daily bread. Not exactly a big reach. What I'm saying is a lot of the projections are going to be wrong because random shit happens. Assume that rate of infection is a constant, then suddenly the virus infects a large nursing home. Oops: Predicted 50 dead, you get 70. Sounds callous, but math modeling has never oozed emotion. You look at a lot of state's curves and you see nothing but random events completely swamping any actual curve.
Today NY is down. MA is up. Good news, bad news. I'd like worldometers to have not just a "yesterday" but -1, -2, -3 days so you could have multiple columns visible at once, see something about peaks and crests over time. Don't feel like spreadsheeting the data. Not today. Not this week.
What would worry me isn't the cumulative, totaled-up USA curve, but seeing a couple of states with large death tolls develop two peaks. NY goes from yesterday's ~500 to 450 to 400 then 450, 500, 575, 650, 700. *That* would worry me, esp. if nothing else changed--no significant social distancing or lockdown changes, for instance. Similarly, I'd worry if some states that look like they're following the IMHE curve to within 10% or so suddenly deviate up--10, 12, 15 (as predicted), 20, 35, 60, 100, 160 (instead of the predicted 20, 17, 15, 13 ...). If a state like NE develops two peaks, that could just easily be random--the peak is low, and a few oddly placed random fluctuations could produce that.
And, one more time, don't stress out over the positive test reports. I heard a newscast where the reporter said this, and pointed out that the US's postive/total test ratio is about 20%, and that needs to come way down. That ignores that New York's ratio was close to 40 and other places are way under 10. But that tells you something about how much of the population is infected in an area. If you have limited tests, you test those with symptoms. If 40% come back positive, that means 40% of those with symptoms at hospitals/clinics have COVID. If 10%, only 10% are infected with this coronavirus. You want that number *low*. Texas is about 10%--if those with symptoms are only 10% suffering from COVID, how prevalent could it be in the wider population? You'd *expect* a low death toll--and you get one. But the math is fuzzy--when they ramped up testing, that ratio didn't change, so while it's under 10% for Texas it might be 9% or it might be 2%. Can't tell.
= new reply since forum marked as read
