Features Australia

Virtual reality check

It’s the modelling, stupid

19 September 2020

9:00 AM

19 September 2020

9:00 AM

There is an old joke that a geologist, a physicist and an economist are marooned on a desert island with just a can of soup but nothing to open it with. The geologist says ‘let’s use a rock to smash it open’ and the physicist suggests ‘let’s use the sun and heat it up till it blows open’. Then the economist steps forward and says ‘your suggestions will ruin the soup, let’s just assume we have a can opener.’

Assumptions are the mother of all stuff-ups. And perhaps the mother of all those has been the Victorian government’s coronavirus response. Dan Andrews has claimed that his disastrous lockdown policies have been decided by a ‘supercomputer’. Alarm  bells would be ringing for anyone who has experience with actual models.

I have had such experience, at least with economic models. I was sceptical of this ‘supercomputer’ claim because as complex as models can become they normally do not require enormous amounts of computing power.

Sure enough in about an hour, I had downloaded the model, the required software and run the scenarios on my three-year-old, trusty, Dell XPS 15 laptop. I did not realise I owned a supercomputer. This should not have been a surprise to those advising Dan Andrews. On their website the builders of Dan’s pandemic model say that ‘…it was actually originally built for use by anyone on a laptop.’

For those unfamiliar with the complexity of any topic, modelling gives a reassuring level of precision. The outputs are clean and precise. Graphs can demonstrate an unerring prediction of the future and all can be reassured about the correctness of their policies.

These outputs provide a false precision that hides the making of the sausage and the, often, hundreds of assumptions that must be made to generate a precise estimate. What you get out of the model is only as good as what you put in and there is often little scrutiny of what gets put in.

The modelling that Dan Andrews relies upon makes 26 different assumptions about a variety of inputs to the model. One of the crucial assumptions is that an ‘unmitigated epidemic’ would result ‘in approximately 60 per cent infection across the population.

In effect, the modellers assume that if we don’t take any action to control the coronavirus’s spread then 60 per cent of us will get it.

The first thing to say about that is the modellers have effectively assumed failure for a no policy option. In reality, even if the government did not impose severe lockdown measures, individuals themselves will take preventative action in the middle of a pandemic. That is not to say that government’s shouldn’t do anything to protect public health; just that it is naive to assume in the absence of government action, all people would go about their lives as normal.

The more severe criticism is that this 60 per cent ‘herd immunity’ rate is out of date. The most recent evidence indicates that herd immunity could be achieved at much lower rates of spread than that figure thanks to an innate immunity that would appear to be present in some people. Studies published by the Karolinska Institute in Sweden and the Tubingen University Hospital in Germany indicate that some patients have reactive T-cells that ward off the coronavirus. A study in Science showed that this immunity could exist in 20 to 50 per cent of people.

These results mean that the Covid-19 herd immunity threshold could be as low as 10 to 20 per cent according to Oxford University’s Centre for Tropical Medicine and Global Health.

Changing these assumptions completely changes the policies that would be required to get coronavirus cases below a certain number. An over reliance on models ends up hiding the key determinants of a decision’s outcome rather than illuminating them.

Remember we were told at the start of the pandemic that 50,000 to 150,000 Australians could die. These predictions have proven to be widely inaccurate. At the time they were made in good faith but the assumptions do not appear to have been updated based on the information we now have on the virus.

We have seen the same saga occur over other contentious policy issues, such as climate change. Climate change modelling regularly forecasts armageddon-like outcomes if there is no action taken to try to change the temperature of the globe. These models of global climate have to make gigantic assumptions over weather patterns, the role of water vapour and the formation of clouds. It is no surprise then that the models often generate incorrect forecasts.

Exposing the inadequacy of climate change models takes decades, however, given their long forecast periods. Pandemics run over the course of years not decades and so the inadequacy of epidemiological models has been exposed much quicker.

When I was last on Q&A, the first question I was asked was: given we have listened to the experts on coronavirus, why shouldn’t we listen to the experts on climate change? I answered that the two questions are very different because acting on climate change requires global coordination which has not been forthcoming. We can shut our borders and lockdown our economy in response to a pandemic on our own.

Now we can add another response to this question. Why should we destroy our economies based on climate change models when we have just destroyed our economy based on pandemic models that have proven to be so off the mark?

There is always one ingredient than any model lacks: common sense. And the common sense is that you need a can opener to open a can.

Got something to add? Join the discussion and comment below.

You might disagree with half of it, but you’ll enjoy reading all of it. Try your first 10 weeks for just $10

Show comments