It’s always the models

I’m getting more and more convinced about the centrality of our animal physicality, in everything from benefiting from nature to worshiping God “in spirit and truth.” A rather saddening recent essay suggests that the greatest harm to children from “online education” in lockdown is going to come from “derealisation,” whereby they become increasingly undistinguishing between reality and the virtual world. Worse still they become less able to see that the difference matters.

There seems to be a similar problem in the sciences.

The release of modelling data last night (as reported in the Daily Mail) shows that the UK government’s ultra cautious release from lockdown – stretching from now until July – is based on SAGE modelling that showed a best-case scenario of 30,000 more COVID deaths whatever happens, and a worst-case of 90,000, if restrictions were lifted too quickly.

By contrast, the data in the real world shows that the massive drop in “cases” started a fortnight before the latest lockdown, and that it matches the kind of New Year plummet in seasonal infections seen in many previous years; that infection and serious infection rates are decreasing significantly faster amongst the vaccinated vulnerable groups, in a programme on-target to protect the susceptible in a month or two more; that the “R number” is at its lowest for ages (0.6-0.9, and dropping a few percentage points daily), and that in any case the exponential increases in cases predicted by the models if “R” goes over 1 have only been seen at the very toe of the Gompertz curves for any country; that the models’ pessimistic estimates of population immunity seem to continue to ignore not only the known natural and T-cell immunity, but the whole vaccination programme; and that even the WHO advises against lockdowns except to avert health-system failures because of the massive economic and health costs.

The government also seems to have forgotten that their original criteria for this lockdown were to reduce pressure on the NHS and to roll out vaccination, both of which have been achieved. Isn’t it fascinating how international that mission-creep has proved, as my American readers will recognise?

The fact is that the predictions of all the models have been thoroughly disproven by events. But that makes no difference to SAGE, because it is accounted for by the supernatural success of lockdowns, even though events correlate abysmally with the lockdowns. “Controls” such as Sweden and Florida, which also falsify the models, are explained away or ignored. The bottom line seems to be that the models are the facts, and the data are negotiable – a classic example of derealisation, only without the child’s excuse of being forced by diktat to live life entirely through a computer screen.

But this is not the sole example of such religious belief in complex models. In 2015 the Climate Change Secretary Amber Rudd convened a meeting between the pro-anthropogenic climate change Royal Society, and the skeptical Global Warming Policy Foundation, to see if they could resolve their differences to the benefit of policy. A spokesman for the latter said that they got their opponents at the RS to admit that there had been a 17 year hiatus in warming, which they explained away in various ways. But when asked how long a pause it would take for them to be persuaded that the warming models were in error, the reply was “fifty years” – and that not from 1997, when the hiatus began, but from 2015.

Now, bear in mind that global warming only became a scientific prediction in the 1980s, immediately after a reversal of the global cooling period that had for over a decade been used to predict a new ice age with great confidence. Many of you were there, and remember that this was not a minority controversial view, as is now claimed. In the 1980s, though, the consensus could turn on a sixpence when the weather changed – now, the modelling is so secure that contradictory data would be ignored for 67 years. A British meteorological scientist, John Mitchell, is attributed with saying: “People underestimate the power of models. Observational evidence is not very useful.”

But as I have discussed on a number of occasions, models become more liable to error the more complex they are, and if they are not validated by their predictive power for reality (as opposed to their ability to be fine-tuned to predict the past) they are very dangerous as tools of major policy. They are best described as computerised hypotheses, and they do in fact embody the hypotheses and interpretive choices of their designers, and stand or fall thereby. The more complex the model, the more hypotheticals are involved. And this excludes the greater programming challenges, as was shown by the inability of the Ferguson Imperial College COVID model, even after a makeover by Microsoft engineers, to predict the same results using the same inputs. (Imperial College treated that as unimportant, and simply took the average of runs of computationally invalid results as the truth).

I don’t know how widespread this confusion between models and reality is in science overall, but given the rarity of its being called out by scientists in other fields, I suspect it may be all too common nowadays. My first introduction to it was in relation to the confidence placed in macro-evolutionary statistical modelling, which is in the nature of things untestable in the real world. Yet that crucial limitation does not seem to instill any caution about their reliability. Specialists and lay people alike will completely discard palaeontological evidence in favour of phylogenetic models on the basis, it seems, that computers are modern technology, whereas spades are not.

In the world of computer modelling the future never surprises human expectations. One suspects that this is what makes it so attractive, because in the real world “The best laid schemes o’ mice an’ men gang aft a-gley.” You’re safe in the world of models, especially when even their internal validation trumps the unpredicatble world out there.

Strangely enough, it usually seems to be the case that the real disasters are entirely unpredicted, and that our besetting fears (aka hypotheses) tend not to be realised.

Meanwhile, though, this February still feels depressingly like last March.

About Jon Garvey

Training in medicine (which was my career), social psychology and theology. Interests in most things, but especially the science-faith interface. The rest of my time, though, is spent writing, playing and recording music.
This entry was posted in Medicine, Politics and sociology, Science. Bookmark the permalink.

Leave a Reply