Comment by throwaway_pdp09 | Hacker News Reader

Comment by throwaway_pdp09 | Hacker News Reader

Parent Full thread

throwaway_pdp09

And I am having serious misgivings about posting it, nonetheless, many people have expressed deep concerned, and I'm doing the same, only with actual data.

I made it clear the limits of my knowledge - which may still not excuse me, however I have to ask if you are an epidemiologist and if so, please tell me how to model the data better. And I don't mean randomly google up some papers to dump here and pretend you are an expert (unless you are).

How far can I extrapolate e^x before it becomes useless? What are the confidence limits I can use in the extrapolation? How much can I approximate the logistic function with an exponential, when as now, the cases are currently a small fraction of the population? Why is this site http://nrg.cs.ucl.ac.uk/mjh/covid19/ done by experts using log axis any different from mine using curve fitting with exponentials? Why does http://nrg.cs.ucl.ac.uk/mjh/covid19/#w say the USA is 11.5 days behind Italy whereas my amateur attemp suggest ~9 days? (and yes, I wish I'd seen these graphs before I posted) What else have I missed? Criticise freely, with reason, don't just accuse me of misinformation because I posted my fears, which others have done too, only I did the maths.

Now please excuse me while I look at the logistics curve and compare its early stages with exponentiation.

>> And I am having serious misgivings about posting it, nonetheless, many people have expressed deep concerned, and I'm doing the same, only with actual data.

There is no "actual data" in your comment above. There is a claim that you make that you have graphed some data, but no graphs and no data.

Also, it doesn't take an epidemiologist to know that a non-epidemiologist should not play at epidemiology. The other poster doesn't have any responsibility to tell you "how to model the data better".

> There is no "actual data" in your comment above

There is in http://nrg.cs.ucl.ac.uk/mjh/covid19/#w which I said matches what I've found rather well.

> The other poster doesn't have any responsibility to tell you "how to model the data better".

Yes he does,

> There is in http://nrg.cs.ucl.ac.uk/mjh/covid19/#w which I said matches what I've found rather well

There may be data there, but if I understand you correctly you did not have it when you made your first post.

It is therefore inadmissible evidence as to whether your first post was a reasonable thing to write.

> Yes he does,

If you have decided you want to be an epidemiologist, or at least to learn their skillset, whose responsibility is it to fulfill that desire?

I don't think there is a reasonable answer to that other than "mine."

I don't need to be an aeronautic engineer to know that a person with no such background should not throw hunches about designing planes.

People spend years studying epidemiology. No one can write a comment here that will teach you how to do it properly.

But I'll try to help. You're looking at a graph of confirmed cases, in other words, positive test results.

That may be due to an increase in the number of infected people; or because the number of tests being performed is increasing, and therefore the percentage of infected people reported as confirmed cases is increasing; or because some positive test results (even with confirmation) are false positives. And there may be other causes that I don't know, because I'm not an epidemiologist either, just a statistics grad student.

Figuring out the the contribution of each cause is a job for the professionals.

You are absolutely right that the early phase of logistic growth is practically indistinguishable from exponential growth. As to what extent this observation is useful when applied to the current situation, I don't know – I am a mathematician, not an epidemiologist. But as a rough guide to aid understanding, I think it probably has some merit.

Thanks, it's nice to have an expert around! I was going to plot the divergence between the two but got seriously despondent and hit the bottle. Trying to sober up now.

May I ask, what area do you work in, numerical, discrete, logic, other?

The "behind Italy" number must refer to the number of days after Italy the area reached 100 confirmed cases. Because those numbers obviously don't reflect an extrapolation of the exponential function even 10 days into the future, even corrected for a reduction in the coefficient when numbers grow above 10.000.

Reply on news.ycombinator.com