Great Data Tool from U Dub

[Warriors4ever]

I've been following this, too, and I agree that the data posted today for Illinois makes no sense. Up until today it seemed pretty good.

[Frenns Liquor Depot]

Bill Gates just promoted this tool on LinkedIn

[forgetful]

https://www.foxnews.com/politics/coronavirus-model-estimates-us-deaths-down

Still concerned about this model. They updated it yesterday, and predict on the low end around 34,000 deaths. As of yesterday we already had 37,000. So either they think zombies are going to rise from the dead, or they aren't paying very close attention to the data.

Not to mention we currently have 2000 dying each day, so any model that would have less than 40-45k deaths is absurdly inaccurate.

Are they being pressured to have numbers that are revised down? I just don't believe their model would be so obviously incorrect. And it is suspicious that it would be inaccurately updated, and immediately be pushed by right-wing media as proof that the lockdowns are overly aggressive and we can reopen.

[Frenns Liquor Depot]

Am I missing something.  The tool predicts 60k deaths. 

[forgetful]

Am I missing something.  The tool predicts 60k deaths.

Apologies for not being clear. After updating yesterday the probability range for deaths is now:

34,063-140,381

The lower end of the range is 3,000 deaths below what we already had yesterday. It is massively incorrect, and implausible.

That error alone is likely more than enough to cause the overall model to lower the total death prediction from 68,841 to 60,308.

Clear and obvious error, so it is either intentional or extremely careless in terms of updating.

[Frenns Liquor Depot]

I think they take a day or two to put in actual results.  So it's behind what happened.  There also has been revisions as states start counting differently with the cdc definition change.  My guess is it's those items.   

I've been watching it for ct and it's been directionally accurate.

[forgetful]

I think they take a day or two to put in actual results.  So it's behind what happened.  There also has been revisions as states start counting differently with the cdc definition change.  My guess is it's those items.   

I've been watching it for ct and it's been directionally accurate.

I thought about that, and it generally makes sense. But then put a caveat on the page, instead of saying updated April 17. Similarly, don't go out and do interviews and report the numbers, which are no longer accurate.

The going out and doing interviews with obvious inaccurate numbers, is at least irresponsible. I'd say the same for the reporting. The way it is being reported is also knowingly inaccurate. It all gives these modelers a bad name, when they are already facing trust issues.

[lostpassword]

Not making any judgement on the source or those quoted but I've been seeing this model criticized with increasing regularity the past week.

https://www.statnews.com/2020/04/17/influential-covid-19-model-uses-flawed-methods-shouldnt-guide-policies-critics-say/

Others experts, including some colleagues of the model-makers, are even harsher. "That the IHME model keeps changing is evidence of its lack of reliability as a predictive tool," said epidemiologist Ruth Etzioni of the Fred Hutchinson Cancer Center, who has served on a search committee for IHME. "That it is being used for policy decisions and its results interpreted wrongly is a travesty unfolding before our eyes."

[forgetful]

Not making any judgement on the source or those quoted but I've been seeing this model criticized with increasing regularity the past week.

https://www.statnews.com/2020/04/17/influential-covid-19-model-uses-flawed-methods-shouldnt-guide-policies-critics-say/

Lots of criticism for this model, and as time has evolved, it is easy to see why.

The authors provide no fundamental methodology. So we have no idea how their model is constructed.

The plots look like they simply have a Gaussian distribution fit to current trends. That is going to lead to significant errors after the peak. Data from all nations shows (and this is what is normal for social distancing) that the disease curve, and death curve, more accurate follow a stretched exponential function. An exponential rise to the peak, then a long tailing off.

Unless the US is magically different, that means there will be a lot of mortality on the long tail. Leading to a substantially higher death count than the model predicts. That will lead to a lot of patting themselves on the back right now, looking pretty foolish.

[rocky_warrior]

Yeah, this could have been a great data tool - but I don't trust it at all now.  Not updated enough, and misleading figures when it is updated.

[Frenns Liquor Depot]

Yeah, this could have been a great data tool - but I don't trust it at all now.  Not updated enough, and misleading figures when it is updated.

That's fair.  I defended it but the fact it is only actual data through the 16th makes it not useful

[rocky_warrior]

That's fair.  I defended it but the fact it is only actual data through the 16th makes it not useful

Oh yeah - I like where they started.  And I still think they're trying to get it right.  Unfortunately,  pressure is now on them because it was promoted so early and so much.  It's researchers trying to create an accurate model - and now influential people and politicians are using as justification one way or another.  Unfortunately, the model is still developing - and given they're trying to encapsulate a fragmented data set (some staying at home, some not, and everywhere in between in the 50 states) - it's going to be difficult to be accurate!

[forgetful]

Oh yeah - I like where they started.  And I still think they're trying to get it right.  Unfortunately,  pressure is now on them because it was promoted so early and so much.  It's researchers trying to create an accurate model - and now influential people and politicians are using as justification one way or another.  Unfortunately, the model is still developing - and given they're trying to encapsulate a fragmented data set (some staying at home, some not, and everywhere in between in the 50 states) - it's going to be difficult to be accurate!

His own colleagues at U Dub have came out and criticized his methodology and have been quite harsh on its utility.

[Coleman]

We are well over 40,000 deaths, so I see no way this thing ends with only 60,000, which is what the tool is still predicting.

Could we stay under 100,000? Possibly, and hopefully. But if states start reopening sooner vs later that will be tough.

[GooooMarquette]

His own colleagues at U Dub have came out and criticized his methodology and have been quite harsh on its utility.

Yeah. I applaud the guy's intent and effort, but the execution and resulting accuracy have been lacking.

[Jockey]

Yeah. I applaud the guy's intent and effort, but the execution and resulting accuracy have been lacking.

Yesterday they revised and said deaths would be 45,000 - 65,000

I guess they figure today might be the last day that people die. Pretty flawed data.

[Lennys Tap]

Yesterday they revised and said deaths would be 45,000 - 65,000

I guess they figure today might be the last day that people die. Pretty flawed data.

I don't think you're reading it correctly. When I go to May 2nd, their projection is approximately 59,000 deaths. Range is 45,000 - 100,000.

[rocky_warrior]

I don't think you're reading it correctly. When I go to May 2nd, their projection is approximately 59,000 deaths. Range is 45,000 - 100,000.

I think his point is that there were already 45k dead in the US yesterday - and no statistical model should show zero people dying after today.  Though I guess an asteroid could hit the earth killing everyone - and thus zero more Covid-19 deaths!

[ZiggysFryBoy]

I think his point is that there were already 45k dead in the US yesterday - and no statistical model should show zero people dying after today.  Though I guess an asteroid could hit the earth killing everyone - and thus zero more Covid-19 deaths!

C'mon SMOD!

[Jockey]

I don't think you're reading it correctly. When I go to May 2nd, their projection is approximately 59,000 deaths. Range is 45,000 - 100,000.

You could be right - maybe I read it wrong. But that almost makes it worse. Their lower limit is fewer than the deaths as of April 22.

I would also be thrilled if the total is "only" 59,000 by May 2. That would mean we are making great progress at bringing down the daily death totals.

[Lennys Tap]

I think his point is that there were already 45k dead in the US yesterday - and no statistical model should show zero people dying after today.  Though I guess an asteroid could hit the earth killing everyone - and thus zero more Covid-19 deaths!

Actually (if anyone survived) we'd have a boatload more Covid deaths since everyone who die with it we are classifying as dying from it.

[TSmith34, Inc.]

Actually (if anyone survived) we'd have a boatload more Covid deaths since everyone who die with it we are classifying as dying from it.

If you are trying to say what I think you are trying to say, that we are overcounting deaths from COVID, then no--we actually undercounting them by about 50% according to a couple of statistical studies I've previously linked.

[rocky_warrior]

If you are trying to say what I think you are trying to say, that we are overcounting deaths from COVID, then no--we actually undercounting them by about 50% according to a couple of statistical studies I've previously linked.

(shouldn't have to explain...) He's being funny - saying if the asteroid hit, all positive covid cases (2M-ish) would count as covid deaths, not asteroid dealths...

[Lennys Tap]

(shouldn't have to explain...) He's being funny - saying if the asteroid hit, all positive covid cases (2M-ish) would count as covid deaths, not asteroid dealths...

Thank you, Rocky.

[ZiggysFryBoy]

(shouldn't have to explain...) He's being funny - saying if the asteroid hit, all positive covid cases (2M-ish) would count as covid deaths, not asteroid dealths...

Tsmith has a non-existent sense of humor.   He's like a kindergartner, who needs explaines to him in simple words.