IHME Model: Declining into utter bullshit

I’m now checking the IHME Georgia COVID-19 model just for giggles. These dishonest scumbuckets are getting further and further from reality. And I’m not even talking about their projections.

IHME currently claims that Georgia saw 965 confirmed infections on June 4, 2020. Confirmed; not model projection.

Georgia DPH says…

89

So from which stinking orifice did IHME pull an additional imaginary 876 cases? Cases that the state agency that gathers and reports this data doesn’t know about?

Just for scale, while IHME is claiming 965 cases that day, Georgia reports that its peak new cases day was April 20, with 950.

As for deaths, IHME is still holding deaths in reserve to falsely maintain a fake curve. IHME claims 2,084 had died by June 3, while the state says 2,159. While that might seem optimistic on IHME’s part, you have to remember that they’ll maliciously report the extra deaths on a later day to make it look like lots of people are still dying. For instance…

Daily deaths, June 3
IHME: 29
GA DPH (the source of the real data): 9

Georgia has not had 29 daily deaths since May 12. And it’s been dropping since. It was declining then.

The IHME model is fraudulent. If it isn’t deliberate as I think, they need to show the source of their alleged data.

If you found this post useful, please consider dropping something in my tip jar. I could really use the money, what with ISP bills, rabbit feed, and general life expenses.Click here to donate via PayPal.
(More Tip Jar Options)

IHME Model May Not Be Fraud, Exactly, After All

I had a discussion with a reporter who had done a story claiming that there were 32 COVID-19 deaths, and 610 new cases, “in a span of 24 hours in Georgia.”

Since the Georgia COVID-19 Dashboard reported 2 new deaths and 6 new cases (as of 5/13/2020, 1:25:00 PM), I was a little curious where she got her numbers.

It turns out she was just looking at how much the “Confirmed COVID-19 Cases” and “Deaths” total numbers at the top of the page incremented, and apparently assumed all those cases happened in the previous 24 hours. I explained that the increment actually includes cases from different days. Due to delays in testing and reporting by counties and labs, it can be weeks before a case that popped up on 4/27 finally appears on the Dashboard. In fact, I picked 4/27, because yesterday that date did pick up some new cases, making the new peak day (previously the peak day was 4/20).

While older dates are fairly stable now, the numbers for the past two or three weeks can be fluid.

But then it hit me: I’ve called the IHME model fraudulent because they are clearly generating new peaks by lumping days worth of data together and reporting them as occurring on the same day… just like this reporter did.

The reporter is being sloppy, and gives the impression those cases/deaths occurred in that 24 hour period. That’s the kind of reporting that panicked the nation into an unwarranted house arrest. Great for clickbait, bad for informing people.

For IHME — if this is what they’re doing — it’s laziness and incompetence that’s totally irresponsible in an allegedly scientific endeavour.

If you found this post useful, please consider dropping something in my tip jar. I could really use the money, what with ISP bills, rabbit feed, and general life expenses.Click here to donate via PayPal.
(More Tip Jar Options)

This why the IHME COVID-19 model is fraudulent

I mostly avoid the crappy IHME model now, but I happened to see a reference to it today. A quick comparison to reality made it very clear what they are doing.

This is IHME for Georgia, deaths per day; solid red is allegedly the real numbers, and the rest is projected. Note that their total deaths projection is actually fairly close to reality. But also note their curve.

IHME claims Georgia peaked at 100 deaths per day, and continues to have high spikes.

Now this is the actual deaths per day reported by Georgia. Deaths per day gradually increased, accelerated, peaked, and has been declining since mid-April. It peaked at 48.

Why are the numbers so different? Georgia’s graph shows daily deaths. IHME doesn’t; they save up days’ worth of reporting, then graph those days’ worth as occurring on one day. Thus, the total comes close to Georgia’s reality, but they get to generate continuing spikes that, when curve-fitted and projected, give a much higher and extended curve than reality. They “projected” 45 deaths for 4/29 and 4/30, based on that curve. Reality was 7 and 3, respectively.

Falsifying dates of death to generate continuing peaks is fraud. They’ve had plenty of time to look at the real data reported through multiple sources. If it was merely an error, they had plenty of time to correct it. They did not. They continue to use false data. They have no excuse for not noticing that they’re claiming a peak over twice as high as Georgia’s actual peak; 100 vs. 48.

Since they also know governments were using their model to make plans to deal with the outbreak, and knowingly provided — are still providing — a falsified model, I think a criminal investigation is warranted.

Added:For the slow, here’s a simplified example of how counting incidents on the wrong day skews the model. The vertical axis is Cases. The horizontal axis is Days.

example-graph

In this example, we had ten total cases, with 1 case occurring per day. That is graphed in black. The curve is flat.

But using IHME’s methodology…

Day 1: They report 1 case.
Day 2: skipped
Day 3: They added days 2 and 3, and reported both on day 3.
Day 4: skipped
Day 5: skipped
Day 6: Added days 4, 5 and 6 together, and reported all on day 6.
Day 7: skipped
Day 8: skipped
Day 9: skipped
Day 10: Added days 7, 8, 9, and 10 together, and reported all on day 10.

A much different curve, showing cases increasing now. But the total number of cases is the same.

Fraud.

If you found this post useful, please consider dropping something in my tip jar. I could really use the money, what with ISP bills, rabbit feed, and general life expenses.Click here to donate via PayPal.
(More Tip Jar Options)

 

IHME Model: Stephen King has been fired

The IHME COVID-19 model, notorious for “projecting” disease stats that look more like an apocalypse model than reality, has been adjusted once again. This time, though, instead of “fixing” it to make hindcasts even worse, they actually revised it such that numbers are now into the merely exaggerated.

I first noticed the model’s issues when examining March 31 “projections.” I chose that date because it had become the present, and thus a useful test of the model’s accuracy.

It failed. It’s death projections weren’t too bad, but the hospital beds needed numbers were insane.

Then:

3/31 projected beds: 95,581

3/31 actual beds: 38,743

For giggles, earlier this month (April) they revised the model and the 3/31 prediction (a hindcast, mind you) became…

107,638

With this latest, post-Stephen King, revision, 3/31 now hindcasts 41,070 beds needed vs. the actual 38,743 that I estimated. That’s much more reasonable.

The total projected deaths looks more like what I expect, too: 60,415.

Don’t get me wrong. For those that do get COVID-19 and go sour, it’s bad. Very bad. But that’s also true of the seasonal flu (which has already killed more people than COVID-19 is projected to kill). And the people who are particularly vulnerable to SARS-CoV-2 are the same people particularly vulnerable to the flu.

COVID-19 looks less like what most people see as a pandemic, and more like an extended “flu” season (yes, I know influenza and corona viruses aren’t the same thing).

If you found this post useful, please consider dropping something in my tip jar. I could really use the money, what with ISP bills, and general life expenses. And the rabbits need feed. Click here to donate via PayPal.
(More Tip Jar Options)

IHME Model Revisited

So how’s that IHME COVID-19 model doing today? When last we looked, it appeared to be run by Stephen King

For April 5, it projected:

  • Beds: 179,267 (126,649-225,921)
  • Deaths per Day: 1,529 (1,228-1,790)
  • Total Deaths: 9,893 (9,023-10,646)

Reality says:

  • Beds: 37,589*
  • Deaths per Day: 1,030
  • Total Deaths: 9,536

* I’ve estimated that using Georgia’s 19.3% of cases needing hospitalization. I applied that figure to US total cases (333,173), then subtracted the 17,177 recovered cases and 9,536 dead, neither of which need hospitalization for COVID-19 anymore.

DPD comes in almost 200 fewer than the low end of IHME’s range, but nearly 500 below the mean, which they’ve inflated by about 50%. Total deaths predicted is still high, but the range includes reality.

But beds are pure science fiction (there’s nothing scientific about that). They projected 4.77 times as many beds as appear to be in use. On this planet/spacetime continuum, at any rate.

Amusingly, they tweaked the model since my last reality-check. They had “predicted” that 95,581 beds would be needed on March 31. Now it claims we needed 107,638 (90,119-122,430) beds that day. With hindcasting getting worse, I have zero confidence in forecasting.

Holy shit, guys. If you’re going to change your model, change it to get closer to reality, not farther.

If you found this post useful, please consider dropping something in my tip jar. I could really use the money, what with ISP bills, and general life expenses. And the rabbits need feed. Click here to donate via PayPal.
(More Tip Jar Options)