We was write
https://twitter.com/chrislhayes/status/1249835313486352385?s=21
3 Likes
So I found out why the IMHE model sucks. It uses a version of the logistic function to map its curves which forces the curve to be symmetrical about the inflection point. That is, if there is a precipitous rise in deaths in some region, it is inherent in their model that there be an equally precipitous drop. You can see this if you look at any of their graphs, for instance here is their current Italy projection:
Spain, same thing:
Steep rise must equal steep drop. Compare with say Denmark:
Slow rise equals slow, drawn out peak with long tail. Denmark is forecast to continue having cases long after Italy and Spain. Itās hopeless. Trump could draw out these model results himself with a Sharpie.
4 Likes
Itās just a shit model. Itās a transparently shit model. Two days more of data caused it to reduce its expectation for the UK from 66k to 37k. Its previous estimate (made two days earlier remember) now lies outside of its upper 95% confidence interval. Meanwhile Italy is already above its 10th April mean projection after just 4 days.
I could do better extending a line with my finger.
5 Likes
Effective rate of transmission. Have to look at it closer but I think 1.0 means constant case rate while 1.1 means 100 cases becomes 110. (Though unsure of this is an arithmetic or long term).
The real world is more complicated in that for a particular infection there is a range of days it can be transmitted and there can easily be changes of policy and behavior during the time window. So to compensate he is using a weighting calculation and some noise reduction or curve smoothing to give recent days more weight.
I worry there is a strong weekday/weekend bias but appears his maths at least smooth that- not seeing a lot of rapid up/down and you look at each state individually.
Also state to state measurement biases.
1 Like
Per the author:
In any epidemic, Rt is the measure known as the effective reproduction number. Itās the number of people who become infected per infectious person at time t.
I take āat time tā to mean at the time under the current conditions. So basically under the current conditions in NY state, on average each infected person is infecting roughly 1 other person if his estimates are correct.
Ponied by one post, the cruellest kind of ponying.
3 Likes
Isnāt what youāve described that the R0 though? Thatās what Iām not gettingā¦
Or is the R0 ātime at patient zeroā?
@Danspartan as well
I think Iām good with nationalizing airlines.
They currently provide an absolute joke of a product that makes the whole experience miserable by design so you pay through the nose to enjoy something slightly less miserable. When they make money it all goes to executives and shareholders, then taxpayers have to pay when entirely predictable (eventually) events wipe them out. Of course they donāt have any fucking cash, it strengthens their bailout pleas. Fuck em.
5 Likes
Per the author:
The most well-known version of this number is the basic reproduction number: R0 when t = 0. However, RO is a single measure that does not adapt with changes in behavior and restrictions.
R0 for this virus is somewhere between 2.5 and 5 depending on who you want to believe.
1 Like
Yeah man, I read all that, I still donāt understand what it means for his plot.
Say his graph says is saying Rt in New York is 1.1. So that means that 1 person who gets infected today will infect 1.1 people on average, right? So whatās the difference between that and R0? Does that mean that his Rt will also be out of date tomorrow? How is his measure adapting with time? The only way I can think is that ātā in this case is time since patient zero, otherwise I donāt understand what it represents.
Edit: Maybe a better question would be, in the example above, what would R1 and R2 be?
R0 is something between 2.5 and 5 for this virus. Itās the reproductive rate without any changes in conditions such as social distancing or wearing masks. The point of the social distancing and wearing masks is to take the natural reproductive rate (R0) and drive it down to slow and eventually eliminate infections.
Sorry, just saw your edit and donāt fully understand your question. R0 can also just be called R.
R = Basic reproductive rate of a virus
R = 2.5 to 5 for SARS-CoV-2 virus
Oh ok, so the R0 is a constant per infectious agent, itās not dependent on other factors? I think I get it then, thatās what was confusing me!
Yes, thatās it. Assuming of course Iām understanding things correctly!
I took ātā to be a subscript denoting time so I assumed R0 would be at ātime zeroā and then there would be an R1, R2 etc. for whatever units of time. But I see now thatās not how it works.
The impression I got was that theyāre making R a function of t, R(t), so itās not constant over time.
Bit of a derail but meh: in general Iām skeptical of market forces in industries in which companies provide an identical product and have limited or no scope to expand the sector. Since they canāt compete in productive ways, they resort to some version of playing tricks on the consumer. This can come in many forms: frequently not actually providing the service promised, hidden charges, sinking money into advertising rather than product improvement. You can see all of these at work in the airline industry. Compare and contrast to something like cars or washing machines where capitalist competition has driven vast improvement in the products over the decades.
In the electricity industry in Australia, this was the result of widespread privatization of the retail electricity market:
Huge growth in sales staff and managers, who do not produce electricity but instead play the zero sum game of attempting to steal market share from competitors. As a result, retail electricity costs tended to increase in areas where the market was privatized.
What many of these companies tend to do is get you signed up and then quietly increase the rates. This happens in car insurance too. I have to spend my life shopping around between companies providing the exact same product so that I donāt get done over by predatory pricing. It fucking sucks.
4 Likes
Right itās changing over time based on conditions. Itās the effective R at the time in question. If you go look at the per state graphs in the article, Rt is generally coming down over time at a different rate in each state based on the conditions in that state changing over the time period.
And the graph I copied into my post is what itās estimated to be for each state as of when he published it.
Your interpretation would seem to jibe with the chart label, which is āMost recent r(t) by stateā, but then, (t) has to be different for each state, right? It would have to be time since patient zero, no? Say t is in days, New York might be showing R(30) whereas Alabama is showing R(5) thenā¦ [making those numbers up]
Iām not trolling, Iāve read his article and the wikipedia article on this and I still donāt get exactly what the chart is showing! Obviously Iām being super dense somewhere along the line.
Just saw this, pretty sure that answers my previous post - thanks.
Once again this thread days ahead of the media.
9 Likes