Why Covid-19 predictions won’t always add up

A doom-laden Medium article went viral last week but even the best modelling on the spread of coronavirus is likely to be flawed

17th March, 2020
Why Covid-19 predictions won’t always add up
Shoppers in Shangai have their temperatures checked. At the start of the outbreak in China, the doubling rate was about four to five days. As of today, it has been six weeks since the number of cases doubled. Picture: Getty

Covid-19 has taken over the world. At the time of writing, there are about 180,000 cases in 146 countries, and thousands already dead. It’s a scary time.

One of the scariest things about the new coronavirus is how quickly it spreads. Because of the way that epidemics grow, it can seem like there are virtually no cases one day then all of a sudden you are seeing tens of thousands of people sick.

This is in many ways a reflection of how disease is spread; what happens is that each person on average infects a few more. At the start, one person infecting five more means just six cases in total. If you have already got 100 people sick, and every person infects another five, suddenly you are up to 600 overnight.

Which is where predictions come in. We want to know how many people are going to be sick at a given time, not just because we all want to know if we personally are going to get sick, but because it’s important for planning health services and the like.

The thing is, models like this are incredibly tricky. I’m sure everyone has seen the predictions on the internet, saying that anywhere between ten thousand and a hundred billion people are going to be infected over the next few months. Neither of these numbers is the slightest bit plausible, but the issue is that they sound convincing when you see them referenced in a blog post or tweet.

A piece on Medium by Tomas Pueyo was widley shared last week. It purported to show, among other things, that there were many hundreds or thousands of times more coronavirus cases than the official figures.

The thing to remember about all modelling is that it is just a guess, but with numbers. Some of those guesses are really good – people spend their entire career learning how to do it well – but even those models are still prey to the things you assume at the start.

Let’s look at two different scenarios as an example, using the doubling rate of Covid-19 as a crude model. The doubling rate is the number of days that it takes for the number of cases in an epidemic to double – say from 100 cases to 200.

As of March 15, Ireland had 129 confirmed Covid-19 cases. If we assume that the doubling rate is four days, and it remains constant, this means that by May 1, we would expect there to be about 550,000 cases. If we instead assume that the doubling is six days, we would expect there to be 35,000 cases. While we are only slowing down the rate by half – increasing the doubling time by 50 per cent – it makes a huge difference, because this is an exponential curve.

Wrong assumptions

This is the problem with models in general. Smart mathematicians can and do make amazing, detailed models that predict the growth of diseases, but even these can be ruined by incorrect assumptions. The problem with Covid-19 is that we really don’t know all that much yet, which means that most of our assumptions will be wrong.

In turn, that means that most of the predictions will be wrong too.

It’s also worth noting that things are changing so quickly that we really can’t predict what’s going to be happening in a week’s time. Assuming anything about what people are going to be doing even a short time in the future is a risky business.

If everyone stayed home 100 per cent of the time when feeling sick, we would probably see the rate of new cases drop precipitously. We might even be able to stop the epidemic in its tracks. Take China, for example. At the start of the outbreak, the doubling rate was about four to five days. As of today, it has been six weeks since the number of cases doubled, and it is likely that it will take even longer in the future if the current trends hold.

There’s an old saying in statistics: “All models are wrong, some models are useful.” The point is not that we can’t predict anything; it’s that we can’t be certain that any of our predictions will be even remotely close to the truth. Good epidemic modellers will produce hundreds of different potential futures, because they know that even the smallest differences can completely change the outlook for the future.

Ultimately, this is actually good news for us all. Rather than being terrified of the viral article condemning everyone you know to an untimely demise, realise that things you do can have a big impact on the course of the disease. Start practising social distancing. Wash your hands and try not to touch your face. For everyone’s sake, if you feel sick stay at home.

These are scary times, but we are all in this together.

Gideon Meyerowitz-Katz is an epidemiologist working in chronic disease

Share this post

Related Stories

AIB financial targets will be met by closures and 1,500 job cuts

Analysts Ian Guider 4 months ago

Brian Keegan: Businesses deserve a reward after such a tough year

Analysts Brian Keegan 4 months ago

Flutter playing Fastball as stakes rise in the US

Analysts Peter O'Dwyer 4 months ago

Credit guarantee: The money’s there, but where is it going?

Analysts Ian Guider 5 months ago