• Welcome to OGBoards 10.0, keep in mind that we will be making LOTS of changes to smooth out the experience here and make it as close as possible functionally to the old software, but feel free to drop suggestions or requests in the Tech Support subforum!

COVID Thread 2: Operation Ludicrous Speed ! (Super Political!!!)

birdman said:
What were the confidence intervals of prediction interval on the 255000 prediction and the 219000 prediction? Predicting 61 days ahead that there will be 255000 deaths and actually having 223000 is pretty good.
Click to expand...

IHME no longer posts their old projections in an easy to reference manner, and I stopped downloading them quite a while ago so I'm not sure

Keep in mind as of September 1 we knew there were 176,000+ deaths to date in the US; therefore the prediction was ~79,000 additional (IHME) or ~43,000 additional (YYG), and actual was ~46,000

so, IHME, after 5 months of practice, was off by more than 70% in its estimate of deaths over those two months

YYG noted that IHME's model for daily deaths had fallen out of the confidence interval after 20 days:

 
Last edited:
thatguy2016 said:
IHME no longer posts their old projections in an easy to reference manner, and I stopped downloading them quite a while ago so I'm not sure

Keep in mind as of September 1 we knew there were 176,000+ deaths to date in the US; therefore the prediction was ~79,000 additional (IHME) or ~43,000 additional (YYG), and actual was ~46,000

so, IHME, after 5 months of practice, was off by more than 70% in its estimate of deaths over those two months

YYG noted that IHME's model for daily deaths had fallen out of the confidence interval after 20 days:

Click to expand...

I agree with all of this. The IHME projections have been inaccurate from the get-go, and Youyang Gu has done a great job of modeling this pandemic in the US.
 
thatguy2016 said:
IHME no longer posts their old projections in an easy to reference manner, and I stopped downloading them quite a while ago so I'm not sure

Keep in mind as of September 1 we knew there were 176,000+ deaths to date in the US; therefore the prediction was ~79,000 additional (IHME) or ~43,000 additional (YYG), and actual was ~46,000

so, IHME, after 5 months of practice, was off by more than 70% in its estimate of deaths over those two months

YYG noted that IHME's model for daily deaths had fallen out of the confidence interval after 20 days:

Click to expand...

I my opinion, how much uncertainty your incorporate into your model and your predictions should be dependent on the data, of course, but also the decisions you are making with the prediction and the risk tolerance of decision makers. It looks like Gu is using a combination of a mechanistic mathematical model that updates model inputs with each bit of new data that emerges through an AI estimate algorithm. I'm not a epidemiologist, but I do work in ecological modeling and environmental decision making, and we are a little wary of AI driven models because they tend to be over precise, that is they don't effectively account for stochasticities and "unknown unknowns" that may arise in the future. The stakes are usually pretty high though, for example if we fuck up our models a species could go extinct, or something. Also ecological systems are a bit harder to model because of the measurement error, stochastic variation and partial controllability. Kind of like how so many 2016 predictions were way over confident of a Clinton because they didn't consider measurement errors and other stochasticities, like Weiner's laptop.

I am not a decision maker, but it seems to me that in the case of modeling a deadly pandemic, incorporating more uncertainties into the simulations would be a smart move because the decision makers should be working under extreme risk aversion. So an AI model, while currently making good predictions, could be limited in it's ability to predict future stochasticities and system perturbations. That is just me though. Ultimately I agree with your assessment that multi-model inference is the best way to approach this.
 
I feel like a lot of you are putting too much faith in the current Covid19 death-toll being accurate.

These things are almost always revised upward.

IIRC, the CDC had already counted 300,000+ excess deaths in the US by mid-October. Only 2/3rds of those were attributed to Covid.

Looking back, with the benefit of time, I suspect the IHME modeling is better than we realize.
 
vtwhat etc said:
I feel like a lot of you are putting too much faith in the current Covid19 death-toll being accurate.

These things are almost always revised upward.

IIRC, the CDC had already counted 300,000+ excess deaths in the US by mid-October. Only 2/3rds of those were attributed to Covid.

Looking back, with the benefit of time, I suspect the IHME modeling is better than we realize.
Click to expand...

That's a decent point. However, the excess death is not all due to COVID. It is due to people not seeking normal care for heart disease/cancer/stroke, suicides, drug overdoses, etc. I don't think retrospective analyses in the future are going to attribute more deaths directly to COVID.
 
vtwhat etc said:
I feel like a lot of you are putting too much faith in the current Covid19 death-toll being accurate.

These things are almost always revised upward.

IIRC, the CDC had already counted 300,000+ excess deaths in the US by mid-October. Only 2/3rds of those were attributed to Covid.

Looking back, with the benefit of time, I suspect the IHME modeling is better than we realize.
Click to expand...

Yeah, this is what I would term a measurement error, we rely on a network of agencies to report data accurately and on time and that may come with errors. I don't really have the time to dig into Gu's Python3 code to see how he treats possible detection bias / reporting error / sampling error, but an AI algorithm would not model that kind of error unless it was coded in as a model parameter.
 
3d791ba0628bbb2d690002b981ec8c45.jpg
 
The Onion continues to deliver.

Got word today of a possible outbreak at the middle school our oldest would be attending. A band kid tested positive and so the other band kids are quarantining for two weeks. Third student reported to test positive. At least one teacher has tested positive.

We had a conference with our 3rd grader's teacher and she said the plan as far as she knows is for all students to go back to brick and mortar in January.
 
Obviously take this with a grain of salt, but the rumor in Ed circles is that we’re all going remote again for a while after Cooper wins the election.
 
PhDeac said:
The Onion continues to deliver.

Got word today of a possible outbreak at the middle school our oldest would be attending. A band kid tested positive and so the other band kids are quarantining for two weeks. Third student reported to test positive. At least one teacher has tested positive.

We had a conference with our 3rd grader's teacher and she said the plan as far as she knows is for all students to go back to brick and mortar in January.
Click to expand...

Band class seems like a bad idea when an aerosolized virus is going around the world. Cant mask up a tuba.
 
So it looks likes it’s not going away on November 4th

 
buddy in Japan pointed out that that country has had 100k cases total

~40% of the population
 
You must log in or register to reply here.
Back
Top