Fair warning: Much of this piece is going to draw on a piece originally published by Fraser Nelson in the Spectator, which you can, and absolutely should, read here. And this is the key part, with parts in bold my added emphasis:
Earlier today I had an unexpected chance to ask questions of Graham Medley, the chair of the Sage modelling committee.
He’s a professor at London School of Hygiene & Tropical Medicine (LSHTM) which last weekend published a study on Omicron with very gloomy scenarios and making the case for more restrictions. But JP Morgan had a close look at this study and spotted something big: all the way through, LSHTM assumes that the Omicron variant is just as deadly as Delta. ‘But evidence from South Africa suggests that Omicron infections are milder,’ JP Morgan pointed out in a note to clients. Adjust for this, it found, and the picture changes dramatically:
“Bed occupancy by Covid-19 patients at the end of January would be 33% of the peak seen in January 2021. This would be manageable without further restrictions.
So JP Morgan had shown that, if you tweak one assumption (on severity) then — suddenly — no need for lockdown.
The basic facts of the matter are simple enough: The UK’s models for the Omicron variant only project the overcrowding of hospitals and a requirement for lockdown if you assume, when putting the model together, that Omicron is exactly as severe an illness as the Delta variant. To be clear, what we mean by “severe” is not how easily it will transmit, but how sick it will make you. If Omicron makes you less sick than Delta did, then the models change dramatically. And JP Morgan, when they changed that one assumption just slightly, arrived at an entirely different conclusion to the model used by the UK.
Why, asked Fraser Nelson, could this be? And so, he turned on twitter to one of the scientists behind the pessimistic model to ask for a reason. This was the answer he received, in full, and unedited:
We generally model what we are asked to model. There is a dialogue in which policy teams discuss with the modellers what they need to inform their policy
— Graham Medley (@GrahamMedley) December 18, 2021
“We model what we are asked to model”.
That is not a scandal, before you jump to any conclusions that it is. But it is, however, very important information.
When the public thinks of “the model”, they often think instinctively of the model as a prediction – a projection – based on the expert views of data scientists. But the truth is that if those data scientists are only inputting to the model what they have been asked to, then they have no responsibility if that model ends up being wrong. The rule of models, after all, is this: Garbage in, Garbage out.
The question here is simple: Who is deciding what goes into the model? And why is that data not being widely shared?
For example, any projection about hospitalisations must, de facto, be based on an underlying equation about how many infected people will end up in hospital. If you say – at a very rough level – that five out of every hundred will end up in hospital then you will get a very different hospital number at the end than if you say that two out of every hundred people will end up in hospital. Being accurate about that number is vital to the whole exercise.
As yet, there simply is no evidence that the Omicron variant is as deadly to an individual as the Delta variant. In fact, the preponderance of evidence from South Africa appears to be that it is, in fact, substantially less dangerous at an individual level. The worry has consistently been that it is more transmissible, not that it is more dangerous.
So what assumption, then, do the Irish modellers make about how dangerous the variant is?
We do not know, is the answer.
And who decides what assumptions go into those models? And what are their qualifications?
We do not know that either, is the answer.
What we do know, as a statement of fact, is that Irish modelling, in common with the UK models, has been consistently and persistently pessimistic relative to the ultimate outcome of each wave of the virus.
That suggests that some of the assumptions going into these models are unduly pessimistic. And that they result, consequently, in overly scary scenarios, and overly cautious policy responses.
This is a very important issue. It’s the kind of story that, in a normal country, Fraser Nelson and the Spectator would win awards for. Chances are though that it won’t get so much as a mention in any other Irish publication. Such is life.