Hacker News Reader

The trouble is that pretty much all of the data is garbage quality, and if you feed garbage into a computer you of course just get garbage out. The figures from Wuhan are particularly bad, probably just a function of their testing ramp-up until well after the peak, and the ones from Italy might not be much better. (The other problem is that China has taken the position that their numbers are an exact description of reality and their approach is a model for the rest of the world to follow, basically for domestic political reasons, and the WHO seems to be eagerly regurgitating this to the many people who'll listen.)

You can model the quality of your data too, with proper tools. No articles in public circulation do that, though.

Is there an open data portal for emergency room visit stats? That would probably paint a better picture of the real case count, by showing us whether there's a spike in visits for acute respiratory illness.

I wouldn't necessarily call the data "garbage" but the numbers are definitely _soft_.

But soft numbers are a common issue with many areas of public health. In particular, analysis of the current pandemic suffers from both a "numerator" and "denominator" problem (http://conflict.lshtm.ac.uk/page_83.htm)

I think many of us are accustomed to getting high-quality stats when studying a process. But that's often not possible with a fast-moving public health issue.

I think you have to acknowledge the limitations of the data, but it can still be helpful.

The thing I _don't_ like about the JHU dashboard is that it doesn't really let you slice the data in interesting ways. For example, I'd like to be able to click on a country or state/province and see temporal data. This doesn't seem like it would be that difficult, but maybe I'm naive.

I've tried doing some stuff with the Italian data but got a similar roadblock. For example, I'd be interested in knowing, among the ICU cases, if these cases were new (people arrived in the ICU with a diagnosis) or a worsening of the already existing patients (perhaps in sub-intensive care earlier).

It would help forming an idea on how the current situation (in addition to the new cases) is going.

That kind of tool will give you accuracy without precision. There are so many unknown assumptions involved that a model like that is going to give at most qualitatively correct results.

People who model the effects of climate change, like Nate Silver and the 538 crew, run thousands of simulations and present a summary. Even then you can get hung up on the "unknown unknowns".

For instance 1970s estimates of nuclear reactor accidents assumed the main risk was that the pressure vessel was going to burst. That's an easy risk to control, so you hardly ever hear of a pressure vessel bursting at a chemical plant, but all the time people blow up or suck in storage tanks that can't handle 1 psi of under or overpressure.

Now really a legacy nuclear power plant melts down 100% of the time if the power goes out and they can't start the emergency generators. That happens a lot more than bursting pressure vessels!

I'm not saying someone should throw a model with 100 parameters and say it's the truth. But you could incrementally build up a model for questions like "should we close schools and when", refining it step by step and getting a range of outcomes, a probability distribution. Sure, the precision is still limited, but it's much, much better than how the discussion looks today: "oh, our government decided to close all schools for the next two weeks, I guess that buys us a couple of days, so we're maybe no longer 3 weeks behind Italy, but 4". Because that's the level of the discussion on the smarter end.

My point is: we have the tooling necessary to talk about this problem better, more accurately, more precisely, in real-time, and to cover more angles simultaneously - which would lead to better decisions. But we're not using it. At least not publicly.

(This is a part of a general issue of mine, that we absolutely suck at having any kind of data-based discussion; best what we do is exchange badly made charts and context-free data points. We should be exchanging and discussing whole dynamic models. But I suppose this wider issue can wait until we're done with the coronavirus mess.)

We’re in a real situation where choices (or the lack thereof) will have major consequences. If the best estimate we can make is a rough one, it’s still sensible to use it as a guide for our actions.

I have heard through word-of-mouth (third hand) that there are models made by experts not being shared publicly (but presumably with governments etc).

This is definitely the case. I don't know if the lack of sharing is purely reflexive or because they want to avoid societal breakdown.

The third option is "because they don't want to be blamed for model error". Governments aren't necessarily competent, but you can try to get them to understand 5%/95% confidence intervals, at least in hindsight. If you publicly release a prediction, and then the real outcome is the 10% confidence line, you're probably going to be yelled at for being wrong regardless of the error bars.

Of course, if the center of the prediction is horrifying, "people don't understand confidence intervals" then becomes a case of avoiding societal breakdown.

Industrial grade is the key word here. They have there customers, so they don't need "cheap" PR. They are specialized in chemical plants, usually professional organizations know there limits (conditions apply). Plus, the WHO and others have these models and tools as well. Plus the domain knowledge needed.

Why wo don't have anything from these bodies is a different story, so.

> Plus, the WHO and others have these models and tools as well.

Assuming they have[0], my core issue is: why won't they share them? Society is not 100% made of illiterate rubes who can only add (but not multiply) and read a line chart (and get confused by a log scale). There's enough people who could make use of more sophisticated information, including politicians and political advisors.

The discussion about e.g. whether a country should stop mass gatherings now or later would be infinitely better if it wasn't just "I think so" and "we have to flatten the curve!!11", but instead if one could say "look, the WHO model shows it won't help much", or "an alternative model proposed by virologist X suggests that it has most benefits if done today; here, look at this chart", especially if people could play with the models themselves and propose corrections.

--

[0] - I'm not convinced; I have the feeling most work in this space is still done with Excel or basic Python and R toolkits.

You are basically asking the same question I am. Why are they not sharing this kind of things? I think it is because the majority of the population is illiterate in that regard. This whole "exponetial growth modelling" going shows this pretty clearly. Not showing just opens a door they are not aware of I think.

also, I don't care about the tools they use. As lonfg as the results are correct.

“Flatten the curve” is a reference to a chart suggesting it has the most benefits if done today. I can imagine objections about the parameters of the chart or the authoritativeness of the source, but I haven’t seen anyone argue against it on those terms; the vast majority of counter arguments I’ve seen are casual dismissals that it could become so bad when things are mostly OK today.

> Society is not 100% made of illiterate rubes

> There's enough people who could make use of more sophisticated information, including politicians and political advisors.

Far be it from me to suggest politicians and political advisors are "illiterate rubes", however I can tell you with confidence that in my experience politicians don't like numbers and models and won't listen to you if you try to tell them about them. Fundamentally this means that even when they're not actually illiterate, they are functionally illiterate (unless of course said model serves their political interest exactly).

>It is surprising to me to. If this was Star Trek world, you'd have computational models built by experts shared and discussed the very first day

The problem with Star Trek (esp. TNG) is that it doesn't reflect reality at all. It gives you the idea that the people in charge really know what they're doing and are competent. In our reality, this simply isn't the case.

I love watching TNG reruns to relax. It lets me fantasize about living in a well-run society run by mostly competent people. That's not the society I live in now, but this is why this form of entertainment is called "escapist".