Friday, April 03, 2020

Lombardy Lockdown is working?!?

A few small tweaks to the model I was using yesterday and a bit more data have resulted in quite exciting results. Firstly, I improved the death calculation which I had implemented a bit naively. I didn't think it mattered, and perhaps it doesn't much, but every little helps. Now also using the Gamma infection-to-death distribution in the IC work.

I've also changed the prior a bit, I reckon with hindsight that the mean of Rt = 1.5 was a bit high. I'm really interested in whether the data point to a value lower than or higher than 1 with highest likelihood, and if the prior mean is 1.5, a posterior result of 1.3 (say) doesn't really tell you. So I have changed to N(1,1) for the prior of Rt.

And we have two more days of data, since yesterday's file wasn't quite up to date. And the result is....

Looks pretty impressive to me. The lockdown seems to be working! I wonder how believable it is? With deaths starting to trend slightly down, it does seem hard to imagine that the disease is growing.

Uncertainties in all this are a bit dodgy and open to question. But it certainly seems more likely than not that the lockdown in Lombardy has worked.


William M. Connolley said...

Assuming 1 (+/-) for Rt seems a touch dodgy. Can you plot the posterior against the prior?

James Annan said...

I redid the calc with a prior of 1.5±1 and it still looked rather good. I think the tweak to the death calculation has helped a bit, as well as the two extra data points. It gets the turn point a bit better.

Hard to imagine it's going up really just looking at the data set. Is the model even needed? I guess it helps define the relevant time scale and delay for fitting a trend.

Phil said...

...lockdown in Lombardy has worked.

...lockdown in Lombardy is working.

Sorry to be Captain Obvious, but the lock down needs to be maintained until testing and targeted isolation can start the dance.

Phil said...

James Annan said...

Interesting article. i agree with your previous post as well of course, there is no obvious exit strategy until the outbreak is small enough that other control measures are feasible. And even then, it's not obvious it will work. A really challenging situation. China may give some clues, along with SK, but their approaches may not translate well.