Chronology Current Month Current Thread Current Date
[Year List] [Month List (current year)] [Date Index] [Thread Index] [Thread Prev] [Thread Next] [Date Prev] [Date Next]

[Phys-L] modeling is hard

Hi Folks --

Modeling the pandemic is either really easy or really hard.
-- The easy part is this: Any model tells you that the outcome
is exquisitely sensitive to public policy aka interventions
aka suppression efforts aka mitigation strategies.
-- Obtaining any more-quantitative predictions is therefore
next to impossible.
-- Actually it's even worse than that. Even if you knew
what decisions would be made, there would still be multiple
hurdles to overcome, as laid out in this article from our
friends at FiveThirtyEight:

They have refused to put forth a model. Fools rush in where
angels fear to tread.

I have confined myself to cartoonishly simply models. Let's
be clear: The outcome depends on decisions that have yet to
be made.

Consider in particular the IHME projection that has gotten a
lot of attention lately:

I think the lead author on that model was Rose E. Señario.
That is to say, the model you see assumes that starting
tomorrow, the administration will get its act together and
start doing everything right. This scenario produces an
ugly result.

There is of course no basis for such an assumption. So far,
they have done almost nothing right. We know what a successful
pandemic response would look like. Not only do we not have
that, we are not even on a path that will get there any time
soon. The consequences are too ghastly to contemplate.

Here is a detailed takedown of the IHME model:

Note that the IHME guys are not stupid. Inside their model
they have parameters they're not showing you, parameters that
could be used to evaluate not-so-rosy scenarios.

As previously mentioned, a good introduction to what a decent
response might look like is the 5000-word essay by Donald

Another reminder: The media keep throwing around the number
of "confirmed" cases. At the moment, that is meaningless,
because testing is so scandalously lacking in quantity and
quality. It is *not* like an iceberg, where you only see
the tip, but you can use that to infer the size of the whole
thing. It's more like if you are given a cup of tea and
asked to guess the size of the teapot it came from. You
really have no idea. If tomorrows cup is smaller or larger,
you still have no idea.

The number of /deaths/ is more reliable (but still not 100%).
The problem is, it is a lagging indicator. Public policy
interventions instituted now will have zero effect on the
death rate for several weeks.

The number of /hospitalizations/ is somewhat less reliable
(but still incomparably better than "confirmed" cases).
It has some lag but not quite as much. This data is not
easy to come by.

An interesting source of zero-lag data is /fever/ data
brought to you by:
but I haven't entirely figured out how to incorporate that
into a predictive model.

Some *limited* understanding of the importance of policy
decisions can be seen in the graph near the top of this

There is no reason why every country could not do as well
as South Korea. China spent a month trying to gaslight
their population before realizing that wasn't going to
work, and therefore got an order-of-magnitude worse outcome.
The disease spread to the US at the same time as South
Korea. As you can see from the graph, the US curve is
very high and still rapidly climbing, with no end in
sight. *More importantly* keep in mind that this shows
"confirmed" cases, and the US has a uniquely bad testing
program, so the /actual/ US situation is much worse than
the figure suggests.

As y'all know, I tend to be very results-oriented. I
like results! The analysis presented here, sketchy
though it is, leads to some exceedingly obvious policy
imperatives. However, this probably isn't the right
forum to discuss them. Anybody who wants to discuss
policy can contact me off-list.