I think it would be valuable to compile a list of estimates of basic epidemiological parameters of the coronavirus, such as incubation rate, doubling times, probability of symptomatic infections, delay from disease onset to death, probability of death among symptomatics, and so on. I find that my inability to model various scenarios accurately is often due to uncertainty about one or more of these parameters (uncertainty relative to what I suspect current expert knowledge to be, which is of course also uncertain to a considerable degree).
Did you end up finding one besides the MIDAS network, or develop your own? I’m assembling a parameter doc for inputs to a rough model that accounts for ventilator & hospital bed capacity, since it seems like we’re lacking that.
I encourage folks to add parameters w/ citations to the doc, I’ll be active on it for the next few days.
If anyone knows of models that incorporate actual healthcare capacity, please share!
I stopped looking after Bucky supplied the link to the MIDAS network list, since it seemed so comprehensive.
For models that incorporate actual healthcare capacity, see this thread. One limitation of the models I’ve seen is that they fail to account for growth in such capacity. China responded to the realization that they didn’t have enough hospitals by quickly building more hospitals. Maybe Western countries are less competent than China and it will take them longer to build the needed capacity. But it seems implausible that they will be so incompetent that capacity-building efforts will not make a significant difference.
Having a list of of values of interest, with estimates and citations for each, would be great.
But in addition, I gotta say, this seems like just about a perfect use-case for prediction markets: we have a bunch of individual, well operationalized scalars, for which accurate estimates that incorporate all of the existing information are of high value.
Is anyone in a position to set up, or to subsidize, a market on these values?
And to start compiling a list of common problems with the parameter estimates being used. Eg I am seeing some models extrapolate naively from the fact that most cases are coming from the least controlled places with the widest uncertainty bars.
South China Morning Post had a story line a day or so back where Chinese experts were suggesting a 10 fold increase every 19 days. Interestingly the rate seems to be about double that if you look at the last 19 days.
I did not look past the totals but suspect that is highly dominated by South Korea (seems to be slowing), Italy and Iran (these two do not seem to be slowing).
Might also be interesting to put a latitude metric in as well—while I have a “sense” that more equatorial areas have a lower incident (and may be spread rate) I’ve not seen that data plotted anywhere.
I think it would be valuable to compile a list of estimates of basic epidemiological parameters of the coronavirus, such as incubation rate, doubling times, probability of symptomatic infections, delay from disease onset to death, probability of death among symptomatics, and so on. I find that my inability to model various scenarios accurately is often due to uncertainty about one or more of these parameters (uncertainty relative to what I suspect current expert knowledge to be, which is of course also uncertain to a considerable degree).
For current expert knowledge, this list of values from lots of different papers might be helpful.
Wonderful, thank you so much.
Did you end up finding one besides the MIDAS network, or develop your own? I’m assembling a parameter doc for inputs to a rough model that accounts for ventilator & hospital bed capacity, since it seems like we’re lacking that.
I encourage folks to add parameters w/ citations to the doc, I’ll be active on it for the next few days.
If anyone knows of models that incorporate actual healthcare capacity, please share!
Thanks for putting this list together.
I stopped looking after Bucky supplied the link to the MIDAS network list, since it seemed so comprehensive.
For models that incorporate actual healthcare capacity, see this thread. One limitation of the models I’ve seen is that they fail to account for growth in such capacity. China responded to the realization that they didn’t have enough hospitals by quickly building more hospitals. Maybe Western countries are less competent than China and it will take them longer to build the needed capacity. But it seems implausible that they will be so incompetent that capacity-building efforts will not make a significant difference.
Having a list of of values of interest, with estimates and citations for each, would be great.
But in addition, I gotta say, this seems like just about a perfect use-case for prediction markets: we have a bunch of individual, well operationalized scalars, for which accurate estimates that incorporate all of the existing information are of high value.
Is anyone in a position to set up, or to subsidize, a market on these values?
I second this.
And to start compiling a list of common problems with the parameter estimates being used. Eg I am seeing some models extrapolate naively from the fact that most cases are coming from the least controlled places with the widest uncertainty bars.
I’m also very interested in this. Here are some numbers I’ve been using:
Ratio of confirmed to unconfirmed cases (USA):
34 (50%), or 5 (5%) to 94 (95%)
This is based on https://twitter.com/trvrb/status/1234589598652784642 , which estimated the true number of coronavirus cases in Seattle (as of 2020-03-01). I divided that by the number of confirmed cases in Seattle at that time.
Doubling time (USA):
4 ish (which I’m treating as 2 (5%) to 7 (95%). https://en.wikipedia.org/wiki/Template:2019–20_coronavirus_outbreak_data/WHO_situation_reports is how I’m getting 4ish. There are papers that estimate higher: https://www.nejm.org/doi/full/10.1056/NEJMoa2001316 gives 7, for example, but that appears to be in Wuhan post-containment.
South China Morning Post had a story line a day or so back where Chinese experts were suggesting a 10 fold increase every 19 days. Interestingly the rate seems to be about double that if you look at the last 19 days.
I did not look past the totals but suspect that is highly dominated by South Korea (seems to be slowing), Italy and Iran (these two do not seem to be slowing).
Might also be interesting to put a latitude metric in as well—while I have a “sense” that more equatorial areas have a lower incident (and may be spread rate) I’ve not seen that data plotted anywhere.