Edit. So it does include all 50.

But it would still be an amazing feat for all fifty states to error balance according to relative population.

And as youve noted, select few states dominate the totals. Thus, at the very least, for those states, all errors must have been perfectly uncorrelated and perfectly balanced each and every day.

Heck:

https://www.worldometers.info/coronavirus/country/us/


They spell it out for you State by State. The only issue I have with the sight is that they don't show you how it's growing over time. Otherwise I wouldn't be putting any energy into tracking it myself.

He's using all 50 states, he's only plotting the big ones but all 50 are included in the US total.

But check the distribution and the std deviation or something. That would show it.

If I showed all 50 states, you wouldn't be able to see anything. It'd be a jumble of mess. I figured the top 5 plus Texas would give us the most useful information.

magically only spreads to 60 MM?

over what time frame?

And the longer the "perfect" trend line goes, the less I trust it. It seemed reasonable towards the beginning, but the longer and more perfect it is, literally screams that the data is being "massaged", or more likely, the published data is coming from a model and not real counts. The lack of variance is astonishing.

If an actual statistician wants to chime in, I'm all ears. But that looks like material testing reports coming out of China where 100% of all materials meet specification! It's amazing, but when even small samples are tested here upon delivery, large numbers fail to meet spec.

Doesn't matter the time frame unless you want to start talking overrunning the hospital beds and ventilators.

60,000,000 is the number infected from H1N in 2012. I think we know this thing to be more contagious than H1N1 at this point so it's at best a safe assumption.

Staff, can we get a Statistician Badge so we know whether we should trust someone's charts and graphs or not?

And here's the death growth.



I really don't think it's worth arguing that there is some unseen body massaging these numbers.

The curve will start to bend soon, as the isolation measures hit. They'll hit as a delay. There wasn't anything being done before, so you'd expect the cases to grow geometrically. Just like you expect something to fall and increase perfectly with time until drag begins to slow down the acceleration.

What would stop the progression before now?

Here's some states broken out:


Here's USA clustered with other countries.


Looks fine to me.

Notice you didn't show China. I do believe those numbers are horse *****

swimmerbabe11 said:

---

Staff, can we get a Statistician Badge so we know whether we should trust someone's charts and graphs or not?

---

I think I am 50% in favor of this but then again I may be a standard deviant.

Here's an interesting site that UW has put together which includes estimated infections, deaths, and associated hospital resources:

https://covid19.healthdata.org/projections

Thank you, that's a valuable resource.

I've been surprised we haven't seen more bend in the data than we have. I know at some point herd immunity has to come into play. I'm really no expert at the biological aspects of this thing, but do we have data from other countries that says at what % of the population the bend will occur?

For something with R0 of >2 herd immunity isn't a factor until a pretty high number.

The math for a closed system that's well mixed (random movement, no blockage) is log (s) = R0 (s-1) where s is the % left uninfected at the end. For an R0 of 2.2 that works out to only 15% or so.

But happily humans are not well mixed, so it won't get that high. But unless there's an absolute metric crapton of asymptomatics (like, millions and millions) we aren't going to hit herd immunity for this for a long time.

You miss my point entirely.

Even when it starts bending, we should see deviations from the trendline. Its just statistical certainty. The fact that no localized "peaks or valleys" appear around the trendline just screams "manipulation" to me. Or rather, as stated before, this isn't raw data, but estimated data from models.

I just find it nearly unbelievable that on any day NY reported a higher occurrence of cases (vs trend), Washington or California happened to report a lower occurrence of cases. There just aren't that many states impacting the total number of cases, hence, any "randomness" associated with counting, travel patterns, weather (number of people not obeying stay at home directives), and the virus itself, should behave, well, randomly. And over just six or seven states, the odds that we flipped 3/7 and 4/7 every single time we counted such that the errors canceled out is unlikely. Not one day where 5/7, or 6/7, or 7/7 all reported high? Or all reported low? Any such occurrence would have broken trend (but still been within a good and statistically significant curve fit). And over all 20+ days, that's never happened once?

What's your background btw? Obviously, by my spread-sheeting and lack of ability to explain the situation, I'm an engineer. I wouldn't have the foggiest idea how to actually predict this stuff from square one.

No, I understand what you're saying. I get it. You can verify those numbers yourself and prove me wrong if you wish instead of going by your gut.

Also keep in mind, you are looking at a log plot. A lot of noise is hidden just with that fact.

For instance, if you calculate the % daily increase you get:

1-Mar 0.25
2-Mar 0.19
3-Mar 0.22
4-Mar 0.29
5-Mar 0.31
6-Mar 0.27
7-Mar 0.20
8-Mar 0.23
9-Mar 0.29
10-Mar 0.24
11-Mar 0.23
12-Mar 0.24
13-Mar 0.24
14-Mar 0.20
15-Mar 0.21
16-Mar 0.27
17-Mar 0.31
18-Mar 0.33
19-Mar 0.24
20-Mar 0.32
21-Mar 0.23
22-Mar 0.25
23-Mar 0.16
24-Mar 0.17
25-Mar 0.23

So the noise is definitely in there.

Engineer. But a very learn-ed engineer with a masters and an MBA. The MBA courses I found most intriguing were finance and data analysis. I'm also a big fan of stats and have read up about it.

A great book that talks in simple english about many issues with numbers is "How Not to Be Wrong: The Power of Mathematical Thinking" by Jordan Ellenberg.

And it's also quite convincing that the death totals *do deviate from trend to some degree; it is inherently much easier (in the US) to know when someone has passed (as well as much harder to fudge). We should *see similar deviations in the actual data for infections vs the trend, but we don't. This data is almost assuredly being modeled and then backfit. It does not represent actual counts. It simply cannot bc of its level of precision.

*edit

This is a ridiculous and baseless claim. Don't spread FUD. Show your work or stop stating it as fact.

Especially considering these are easily verifiable. Just go do the work and stop with the conspiracy theory BS.

jamaggie06 said:

---

Engineer. But a very learn-ed engineer with a masters and an MBA. The MBA courses I found most intriguing were finance and data analysis. I'm also a big fan of stats and have read up about it.

A great book that talks in simple english about many issues with numbers is "How Not to Be Wrong: The Power of Mathematical Thinking" by Jordan Ellenberg.

And it's also quite convincing that the death totals *do deviate from trend to some degree; it is inherently much easier (in the US) to know when someone has passed (as well as much harder to fudge). We should *see similar deviations in the actual data for infections vs the trend, but we don't. This data is almost assuredly being modeled and then backfit. It does not represent actual counts. It simply cannot bc of its level of precision.

*edit

---

I think the infection data is being skewed by the testing itself. We have limited tests and are only testing people who meet certain symptomatic criteria. Basically, the data suggests that x% of people who meet testing criteria are positive for the virus.

Sponsored by the Bill Gates foundation. Cute. There's that name again.

Yeah, but when you look at the global average of increased infections, we are right in the middle. Globally it's like 33% a day. We're at 29%.

I should also point out that this means we need to ramp up the rate at which we run these tests. As much as NY is testing a day, it soon won't be enough to capture the true growth.

I am showing my work. I use evidence to update my beliefs. And the evidence (or lack therof) that the ata is being fudged grows each time the data fits perfectly with trend. The longer that goes, it becomes equivalent to saying "I flipped a coin 100 times and never got more than three heads in a row". Sure, its possible, and even likely in the beginning, or even for the first ten tosses. Less likely for the first twenty. The longer it goes, the stronger the evidence becomes, ie. the tighter the confidence interval becomes around the fact that we should see some variance.

Unless of course you don't believe that the spread of a virus contains any random components.

For reference, the odds of never getting 5 or more heads in a row in 30 flips is greater than 50%. And a streak of 7 heads in a row is almost 20%. So far, that hasn't happened. The errors have always miraculously been cancelling out.

Like I said, it's *not definitive proof; But it is evidence (in the statistical sense) that data provides is not coming from actual counts. We should see some degree of variance from trend.

Note, that variance from trend does not make the trend insignificant or wrong; rather, it simply means that there are effects that the trend-line is not capturing.

But believe what you will. I won't change your mind.

And, as stated it is *possible* that the actual data fit that perfectly; it's just unlikely (statistically).

That's a different phenomenon all together and a hypothesis I happen to agree with; we aren't testing random samples. Ideally, I would hope our modeling takes this into account, but, until we get enough real and verified data, its hard to know how far off we are on the estimated parameters.

He's updating the best fit each time he adds a sample.

To see what you're describing he'd have to draw consecutive trends each day. That would show a jumble of lines, exactly as you're looking for.

but the country didn't go into hiding then

I had never heard of social distancing until two months ago

I never considered wearing a mask in public until now

I would never have gotten randomly tested for H1N1

the paranoia matters. people of a certain vulnerability will absolutely not interact with the wider world.

so the proliferation of the disease is somewhat overrated. it won't break down your door. it actually seems less contagious than advertised outside of super spreader events. I know people with direct COVID contact for extended periods that didn't catch the bug. that is possible.

so I think it's definitely as deadly as we think, especially for older and obese Americans. but we're spending 6 trillion to slow it down.

maybe that's too optimistic.

jamaggie06 said:

---

I am showing my work. I use evidence to update my beliefs. And the evidence (or lack therof) that the ata is being fudged grows each time the data fits perfectly with trend. The longer that goes, it becomes equivalent to saying "I flipped a coin 100 times and never got more than three heads in a row". Sure, its possible, and even likely in the beginning, or even for the first ten tosses. Less likely for the first twenty. The longer it goes, the stronger the evidence becomes, ie. the tighter the confidence interval becomes around the fact that we should see some variance.

Unless of course you don't believe that the spread of a virus contains any random components.

For reference, the odds of never getting 5 or more heads in a row in 30 flips is greater than 50%. And a streak of 7 heads in a row is almost 20%. So far, that hasn't happened. The errors have always miraculously been cancelling out.

Like I said, it's *not definitive proof; But it is evidence (in the statistical sense) that data provides is not coming from actual counts. We should see some degree of variance from trend.

Note, that variance from trend does not make the trend insignificant or wrong; rather, it simply means that there are effects that the trend-line is not capturing.

But believe what you will. I won't change your mind.

And, as stated it is *possible* that the actual data fit that perfectly; it's just unlikely (statistically).

---

You understand it's a trendline and not some independent prediction right?

Philip J Fry said:

---

No, I understand what you're saying. I get it. You can verify those numbers yourself and prove me wrong if you wish instead of going by your gut.

Also keep in mind, you are looking at a log plot. A lot of noise is hidden just with that fact.

For instance, if you calculate the % daily increase you get:

1-Mar 0.25
...
23-Mar 0.16
24-Mar 0.17
25-Mar 0.23

So the noise is definitely in there.

---

Either you or I are calculating % daily increase differently/wrongly or you're using some kind of averaging here. I'm pretty confident in my calculations, so just wondering what you're doing?

Take the difference between today and yesterday and divide by today. I suppose I should be dividing by yesterday, but whatever.

I don't actually use it for anything other than spot checking.

Philip J Fry said:

---

Take the difference between today and yesterday and divide by today. I suppose I should be dividing by yesterday, but whatever.

I don't actually use it for anything other than spot checking.

---

Got it. I'm calculating rate of increase (new cases today/new cases yesterday) and you're calculating total daily increase... and yeah divide by yesterday

barnyard1996 said:

---

Patentmike said:

---

barnyard1996 said:

---

If one thing is certain, we are all going to die.

---

But what happens then?

---

Sup bro?

---

Just trying to derail a thread

Out of curiosity, what are you tracking and how are you doing it?