Predictions Pertaining to the Spread of CoViD-19

Foreword

I am fully aware that: (a) the case counts do not represent the actual prevalence of the disease; (b) they misrepresent it by a different amount in each different territory; and (c) that includes downright lies from regimes that live off lies. I am factoring these factors into my predictions - so I'm predicting the reports rather than the underlying facts those reports are based on, expecting each reporting authority to remain constant with their reporting policy. Yes, that expectation is now known to be false.

The predictions themselves

Skip to the unresolved predictions

100,000 cases worldwide

Prediction

100,000 cases worldwide by March 20th ±4 days, made on Feb 20th

Reasoning

Based just on Chinese data, seeing the effect that lockdowns had on Hubei, and presuming that Europe/US would act quickly with restriction of travel from China.

Reality/Accuracy

March 6th, 2 weeks early, 15 days rather than 29

Excuse

Europe, Iran, and the US did nothing sensible to constrain disease spread into and through their territory whilst I was furiously facepalming. Acting 2 weeks earlier, and reducing the infection rate by a factor of 2 (3 day doubling rather than 6 day) would have been bang on, but that would have required governments to have brains.

200,000 cases worldwide

Prediction

200,000 cases worldwide by March 20th ±1 day, made on March 6th

Reasoning

On March 6th, the number of cases had just reached 100,000. Seeing almost no travel restrictions to, and nothing within, the new hotspots, it was clearly going to grow there at the same exponential rate that it had in China. China was not perceptibly increasing its case counts by that stage, and so was irrelevant to the growth towards the total.

Reality/Accuracy

March 18th, 2 days early, 13 days rather than 15

Excuse

Everything pretty much played out as predicted, I'm good.

1,000,000 cases worldwide

Prediction

1,000,000 cases worldwide by April 4th ±1 day, made on March 18th

Reasoning

On March 18th, the worldwide case count had reached 200,000, and several patterns had started to appear. I presumed the slow trickle of data from China would continue. The other new hotspot countries were seeing no signs of addressing the spread, so I presumed R would continue at the original R₀, and fitted a pure exponential to the rest of the world's data.

Reality/Accuracy

April 3rd, 1 day early, 16 days rather than 17

Excuse

Damn, I'm good, I really should get paid for this.

100,000 cases in the US

Prediction

100,000 cases in the US by March 28th ±1 day, made on March 20th

Reasoning

The US had begun a clear exponential, and showed no signs of adopting or imposing any NPIs, so I did a straight line extrapolation on the log scale.

Reality/Accuracy

March 27th, 1 day early, 7 days rather than 8

Excuse

My informal prediction while chatting with friends did actually include "and becomes the #1 country at the same time", which was a little bit out, as Italy had begun to restrict movement, but basically this was a good example of how utterly predictable unconstrained disease spread can be in an unprepared population.

1,000,000 cases in the US

Prediction

1,000,000 cases in the US by April 25th ±2 days, made on March 27th

Reasoning

Split the US into 3 regions, depending on how advanced the spread was and how much pressure was already on the healthcare systems and fitted separate sigmoids to each region, as the linear stage had started to become clear, then summed. The "it's everywhere, you're fucked" region didn't get a sigmoid, there's not space here to explain why. I presumed that lockdowns wouldn't work particularly well, as people are dumb panicky dangerous animals, and it only takes a small fraction of idiots to ruin behaviour-based policies.

Reality/Accuracy

April 27th, 2 days late, 31 days rather than 29

Excuse

None needed, but I was lucky - the NPIs should have kept things down, but they failed. (The humans failed, the interventions are sound on paper. The politicians applied them outrageously late, and in an overreaction applied restrictions which were guaranteed to cause people to want to ignore them, which is counterproductive.)

10,000,000 cases worldwide

Prediction

10,000,000 cases worldwide by early July ±2 weeks, made on April 27th

Reasoning

With 3,000,000 cases now recorded, and no slow-down in sight, it's clear that 10,000,000 is a real target that will be reached before too long. Many factors at play here, this is a tricky one. We'll start seeing Russia join the big boys, and South America and Africa will start coming into play too. Data gathering is not so reliable there, so the margin of error will be huge. America might see an upturn as movement restrictions are reduced before the spread is under control. Not sure about Europe, but we're more stoic than the USA, so there won't be as much of a second burst to the first wave. Expecting June to hit 2.5-3M in both EU & N.Am. 2M+ elsewhere.

Update (May 22th)

No changes in prediction, just a "progress report" after 25 days, and 2,200,000 more cases tallied. There's no reason to expect the worldwide rate to slow, as South America's kicking off exactly as predicted, and the US is likely to see some surge as it comes out of lockdown. Over the remaining ~45 days to my target a rate of about 100,000+ per day would bring it in exactly on time.

Update (June 11th)

No changes in prediction, just a "progress report" after 45 days, and 4,500,000 more cases tallied. The growth remained below 100,000 per day for a while, I was fearing this would be a serious undershoot, but South America has finally started to ride an exponential wave, and that has more than made up for the earlier shortfall. 3 weeks away from July, I'm pretty sure my prediction will be very accurate, to within days, not weeks, so let's focus on just the very start of July, plus or minus a couple of days.

Reality/Accuracy

28th June, 3 days early, 62 days rather than 65, but well within my original error bands.

Excuse

Even though projecting that far in advance is like weather forecasting, and largely guesswork after a certain range, everything panned out as predicted. The USA's ramping up did surprise me a little, which explains why the target was reached early.

Terminal US deaths

Prediction

80,000-90,000 (55,000-135,000 with high confidence) deaths in the US to end of 1st wave, made March 29th

Reasoning

Bendavid et al. had just revised their predictions down from 2,000,000 to 20,000 on March 24th, having every bit of data right at their fingertips. This sounded absurd. Not having all the data available, I gave myself an extra week to fit a model to the summarised data to which I had access. I used the same sigmoid fits as the 1,000,000 case prediction. No model was capable of giving a number lower than 35,000, so Bendavid's numbers were beyond impossible. He should perhaps try Hungry Hippos instead, he's out of his depth in this game. Was able to squeeze a 200,000-240,000 upper bound out of the data if I presumed the worst at every stage, but I consider that very unlikely.

Update (April 27th)

Sigmoids are likely to be underestimates, as the gradient is left-skewed (related to the "fucked" region above), so really I should revise this. One thing that's clear is that there are different reporting standards in different authorities, and in particular rumours of a pro-CoViD incentive (the 2 doctors' controversy) which would skew the figures upward.

Update (May 7th)

The US's curves are behaving like no other country in the world's, they've stayed linear for far too long, which implies a mathematically unstable equilibrium, which can happen if someone's curating the numbers, for example. There are now corroborated reports of financial incentivisation to over-count CoViD deaths since April (New York was mentioned by name by the 2 censored doctors, but it appears not limited to just there now), and thus the numbers will be higher than initial projections indicated. For example, 50% overreporting would mean that 1/3 of the deaths (yes - since April, in tainted states, but that's a tiny correction) wouldn't previously have been counted.

Update (May 31st)

Some of the idiocy of the lockdowns will pay back in spades as they are lifted, and several of the more freedum-oriented states will see significant upticks as people utterly ignore all the sensible behavioural advice that places like Taiwan have been following since January (and surviving almost entirely unscathed because of that), so this will be a small underestimate, 20-30%, say? Scratch that!!! What the fuck, US police - you're supposed "To Protect and Serve", not "Murder in Cold Blood". OK, if shit's gonna kick off, so will the case counts as large numbers of people exercise their apparent right to idiocy. The real numbers will probably be 50-100% higher now.

Reality/Accuracy

Who knows, but by April 27th I was already in my "high confidence" band, and by May 9th I'd arrived at my target with the daily increase starting to taper. Of course, then idiot mode happened and my May 31 120,000-180,000 looks reasonable…

Woh - I remember this prediction, let's analyse it. Looking at the US curves, the first wave of deaths could have died down by not long after July 5th, but thanks to stupidity had a second lease of life ending Oct 17th. On those dates, the fatalities had reached 134,000 and 228,000. The former is just at the top end of my "high confidence" prediction, the latter is 69% higher, so satisfies the 50-100% revision.

Excuse

Americans still living in the Wild West…

Terminal Arizona deaths

Prediction

1,400-1,500 deaths (1,200-1,800 high confidence) in Arizona to end of 1st wave, made May 7th.

Reasoning

Firstly, the only reason I'm making this prediction is because the Arizona governor has decided to stop running models, presumably out of embarrassment. With the current case count at 450, healthdata.org predicts 1,043 deaths, which seems to assume a simple sigmoid where the inflection's nearly been reached. However, that's a very naive model (their models are terrible generally, but they're the only site I know which makes projections (apart from what you're reading right now)). From the data I have access to, the daily case rate has not levelled off, and neither has the mortality rate, which we'd now not expect to level off for at least a week or two, as it lags the case rate, so I think there's a lot more running left, even without the policy change that's about to occur. Oh, did I not mention that earlier? Arizona's just about to end lockdown, whilst the curve's still pointing upwards…

Update (May 11th)

I notice that healthdata.org now predict 2,987 deaths, which is quite a change in just 4 days. I'll hold off for a week or so to see if I feel the need to significantly reevaluate.

Update (May 19th)

I don't feel the need to reevaluate. However, healthdata.org, after a momentary dip to about 2,871 deaths, have cranked their prediction up to 6,174 deaths, a factor of 6 change in just 12 days - are their predictions growing at an exponential rate with a doubling time of 5 days?

Update (June 11th)

I still don't feel the need to reevaluate, the death curve is rollin' just peachy. Arizona unsurprisingly didn't have much protest mayhem to give it any extra boost, so the numbers stay put.

Reality/Accuracy

Looking at the curves, AZ really kicked off after July 4th, I can't imagine why. After that, the first wave ended Oct 25th 2020. At that point, the number of deaths was 5874. That's a miss.

Excuse

I wasn't expecting Americans to be idiots in enormous numbers on July 4th. Who's the idiot now, eh? However, this idiot didn't put thousands of additional lives at risk. That's an understatement, risks are just an abstract concept, it was thousands of unnecessary concrete deaths that occured because of socially irresponsible behaviour.

(Provisional) Terminal Brazil Cases

Prediction

4-6 million cases (3-10 million high confidence) in Brazil to end of 1st wave, made June 11th, but gimme a couple more days to check the maths. Hence the wide error bounds presently, this isn't even back-of-a-fag-packet, this is pure eye-balling.

Reasoning

I've not made a prediction for a while, and I seem to be rather good at it. That aside, Brazil's at 750,000 and still ramping up massively, and with a peak still to arrive, and a 1:3 ratio to its daily case skew (similar to UK, better than US, worse than EU), it's only seen 1/8th of the underside of that curve.

Reality/Accuracy

Looking at Brazil's curves, the first wave of new cases ended Nov 5th. At that point, the total number of cases had reached 5,645,000. The bit of the curve visible in mid June was indeed 1/8th of the final area. Nailed it.

Excuse

God-like powers of prediction.

None. Why bother? Human stupidity has again been proven to be boundless.

Conclusion

If anyone says "we could never have seen this coming", they're fooling themselves, and possibly others. And when people in a position of power and responsibility over others fool themselves, and base policy on that delusion, people die.

Note: This is an apolitical comment. Delusion is a property shared across all political parties mostly equally. I'd expect there to be a decent correlation of height of office with level of Dunning-Kruger, so unfortunately the vast majority of people will be able to point their fingers upwards and say "this was badly managed". Replacing a psychopath from one party with a sociopath from another party really wouldn't have helped matters, alas. Top tip: please stop voting for the psychopaths and sociopaths.

P.S.: Why's everybody talking about Sweden and only Sweden when it comes to minimal restrictions? Estonia has handled this at least as well as Sweden. Our restrictions being characterisable by the fact that I have been to a pub or restaurant every single day in the last 4 months. Not all were open, but none were forced closed.

Page last knowingly updated: 2021-11-18