So, it’s rant day, here at Skellywords. Maybe it will become a regular-ish Friday feature. Today’s rant is scientific predictions. I do these predictions for a living - if in a totally different field. Through painful experience, it became clear that understanding data quality was a required part of understanding the uncertainty of that prediction. As we learn more and more about the Covid data, it’s clear that the only prediction that we can make with any certainty is that if everybody acts like they did last year at this time, lots of people would get sick and lots of people would die. I’ll put that at ninety nine percent likely. It doesn’t rate higher, mostly cus I’m not totally positive the world really exists and this isn’t a simulation or dream or something. That’s the other percent.
The modelling kind of indicates-ish what’s going on, and confirms that going back to how we lived last year at this time would be a disaster…a bigger disaster The real useful purpose of numerical models in prediction is they give you a sense of what variables change outputs very much. Sensitivity analysis. It isn’t about them being right, it’s about them being useful in understanding how the variables interact. They can’t predict anything other than what variables we can change, like isolating ourselves, that will move the needle on cases and deaths. That is what they have been used for, and we shouldn’t expect them to predict the future or tell you how many people are going to die from this (beyond ‘lots’) - especially when the data quality is this poor.
The reported number of cases misses the asymptotic maybe fifty percent-ish of cases (or twenty percent or eighty percent or whatev number is popular today…see my point?). The tests have a 30%-ish false negative rate, in any case, so those numbers aren’t great to start with. I tried to cheer my data driven mind by focusing on the dead. I figured the number of dead people would be well counted. My data driven mind doesn’t relate the numbers to people if I try really hard, so I can deal with it. Mostly.
The recent news has a number of examples of people dying of Covid-19 and not being counted in the Covid data set, because underlying condition were blamed. Someone who has a heart attack brought on by Covid might look like they had a heart attack unrelated to Covid and only a full autopsy will figure that out. It means the death data is unreliable and not an ‘unreliable’ that makes reality look any better. The NYT had a story about some epidemiologists looking at the death rates in March in New York this year compared to March in previous years and it was depressingly high enough I stopped reading. That makes the numbers of deaths closer to guesses than I thought.
The degree of miscount in everything will be different in different countries, so the numbers don’t compare well from one country to the next that well, either. The John Hopkins numbers have a 9-11 a day going on in America, at 3300 deaths, while WHO has ‘only’ 2200...a day. Every day. And the number started increasing today as other outbreaks catch up to the Eastern Seaboard. The criteria they use to include numbers is different, so comparing American trends to trends in other countries is difficult. That means figuring out where this is going, from places it happened earlier, has a lot of unconstrained variables. Unconstrained variables make modeling really hard.
Lots of the politicians and pundits are talking about what the shape of the curve will be, and how we can respond in whatever way, and they are just blowing wind. The politicians want to see the numbers they are making decisions on, and the models they come from, and think they are representing more than the possible shape of things to come, They don’t (and shouldn’t be expected to) understand the context without more time. It leaves them telling the public things that aren’t true - but that the politicians and talking heads think it’s true. It’s not a lie if you believe it (Credit to Seinfeld), but it isn’t helping the public dialog to represent loose ‘it’s going to act this way-ish, provided people act this other way’ numbers as they are real ‘counting things’ kinds of numbers. It’s about moving the curves and not overwhelming medical systems - not preventing two hundred and twenty three point seven deaths - exactly. The numbers in the models aren’t about ‘exactly’ anything.
America (and Canada) have very incomplete mixing, so there is no reasonable expectation that they will act the way England or France or Spain or whatever has or will. Note those integrated economies in Europe, where movement between small countries and small areas is trivial, and was totally trivial into the start of this, have peaks at different times, and differently shaped peaks because of incomplete mixing. Those different peaks will be merging in different ways in America and Canada during this period, because of barriers to mixing (distance) have been made stronger by shutting down flights, and locking people away in their houses. Flights are the stir sticks of North American geographical populations mixing, because the regions are so big. Just because the Eastern Seaboard has got past a peak by locking everyone away doesn’t mean Georgia is over the worst and can reopen. And Florida! WWE! WTF! It is not an essential service, please…just…please. If they are going full-bore into bread and circuses to keep the masses distracted, they need better circuses and the ‘bread’ part is the more critical piece. Get that one covered, people.
There is only one set of data that follows through to the burn out of the virus (where it can’t find anyone to infect, so dies off), and that’s China. From that, we know we can burn the virus out with a complete lockdown for a month-ish, and yes, we can’t be sure of the Chinese data quality that leads to any conclusion, but it’s almost certainly good enough for that particular conclusion. We don’t know if it is still lurking in corners and ready to launch new attacks, and we do know that cases are being reintroduced from out of the country. The only sure thing the China data tells us is that we can mitigate the problem by sheltering in place. There are now several populations, along with China in the early period, that show us that not mitigating the virus gets really ugly really fast, and that gets us to where we are now.
We need to shelter in place until we have testing and contact tracing that can control new outbreaks before they get too big to trace, or until a vaccine or treatment that is highly effective is worked out. Technology takes time to get sorted out, and doesn’t often work that well, first thing out of the box. This is going to take a while, unless we get really lucky. People are looking into the hiding spots luck could be hiding, and we can hope they find something, but we shouldn’t expect that to happen. We need to have a robust Plan B that doesn’t make it worse. The more people that get it, the more likely it finds an animal reservoir (or mutates, or other bad things) and becomes a lot harder to mitigate with social distancing less extreme than locking people inside - forever. If you want to know the worst case, that’s it.
We don’t even know people that have been infected and recovered can’t get it again the next day! There is almost certainly a window of immunity to the strain that person had, and multiple strains are less likely in a poorly mixed geographic area, but as for how long the window is, it’s guesswork. The landscape on the other side of that window is fairly blurry, along with most everything else of what’s going on. Real time is hard to work in, which is why you want plans for bad things in case they happen, so less has to be done working through blurry windows in a panic.
aside/
If we’d found a cure to the common cold (a corona virus) instead of focusing on treating it since forever, we’d be a lot closer. Difficult problems are the ones to keep plugging away at, not just giving up because it’s hard. Funding basic research is how we find those things out. The more recent push to only funding research that has a known application that makes money is short sighted. Short-sightedness gets you to this sort of situation unprepared.
end aside/
Peaking at things my brain doesn’t want to face, we want to minimize the number of cases for more reasons than minimizing people directly dying. If this gets into animal reservoirs in city biota it will make tracing breakouts harder, and make breakouts more common. If more people get exposed to it, especially people that have dwindling immunity from previous infections, it makes mutations and new strains more likely. The odds are on it mutating into less aggressive forms in the long term, because killing a host makes transmission harder, but the Spanish Flu mutated into strains that were more deadly, so it happens - and the Keynesian quip ‘in the long run, we’re all dead’ has new and more upsetting connotations at the moment. There are well understood scientific reasons and mechanisms for the stuff brushed past above, but my mind doesn’t want to linger there, so just trust me, or learn about it in your free time and tell me where I’m wrong so I can sleep better.
If we want to inform plans for the future (which I’m fond of), we don’t want to give false hope to people and pull it away when it turns out to be bunk (I’m looking at you Trump, and it is a really unpleasant sight so I’d appreciate it if you didn’t give me reasons to do it). People need hope in times like this, and there are reasons to be hopeful. False hope is destructive because it encourages people to make stupid plans and do stupid things (like go to protests encouraging governments to ‘reopen the whole economy, now!’).
Science is good at these sorts of things, but not always that quick about it. We’ll learn more about the numbers, and how to make better predictions as time passes - but the predictions aren’t very good yet for practical planning like knowing how many ventilators to order. Unfortunately we are in a period that living people get to be guinea pigs because we need a solution quickly and ‘people’ are where the problem started.
We have almost all of the top scientific minds across the world working on the parts of this problem that are relevant to their area of expertise. They are even collaborating across borders and regions and getting on with it in using appropriate urgency. Lets fund them massively, and let them try lots of things that won’t work, so they will find things that will. Climate at least changes slowly enough that there was some time to muddle along - not as much time as we’ve wasted since we figured it out, but that’s a different rant.
In summary. The data quality is presently too poor to make predictions that are more than educated guesses. As more data comes in, we will be able to make better and better predictions and probably make breakthroughs in big data analysis and medical science and epidemiology, and understanding lungs and whatevs - and move the whole bunch of us into a new paradigm of life that is much better in all ways than now. Or not. That’s one of the things about life that pisses me off. It would be so much easier if we could peak into the future sometimes.
Anyway, we will certainly learn important things people weren’t trying very hard to learn before this emergency, and in time, the understanding of the data, and corrections, and reworking, and whatever, will get us through this into a new place. That new place will probably value life more highly and shiny things a little less and be a better place to hang out in for a while. This might even encourage people to make the kind of effort the world needs to mitigate the climate problem - before it leads to an even bigger, if slower moving crisis. We might learn collective problems can be solved collectively and encourage that sort of thinking in our lives! How cool would that be!
Right. That’s enough ranting for now, and I was drifting into optimism toward the end. We can’t have that. We need better data first.