This article is taken from the December-January 2024 issue of The Critic. To get the full magazine why not subscribe? Right now we’re offering five issues for just £10.

---

For want of a nail, the shoe was lost,” the proverb begins. “For want of a plan, the pandemic was lost” should be where Lady Hallett’s Covid inquiry turns its gaze. Deeply enjoyable to the media though it is, with its spectacle of repeated “gotcha” moments as each official, Spad or minister is questioned about obscure points long forgotten is not an effective route to finding out what went wrong in 2020.

To do that, the inquiry is going to need to look further back and do a root cause analysis. Before we enter into that, we should consider just why this is key. The lack of a proper plan for a pandemic is the distal cause of so many related harms that followed. What this tells us is instructive about the failings of the modern administrative state; and also an instructive criticism into how modern inquiries work. On the way, it happens to expose some failures of the academy, too. 

I assumed, partially because I knew it had been “rehearsed”, that in early 2020 a rigorous pandemic plan existed. This also seemed quite reasonable: the SARS virus wasn’t that long ago, and Ebola was sporadically flaring up in Africa. Depending on who you ask, we now know that there either was no plan; or that the plan was so hyper-focused on pandemic influenza (sars being something that clearly happened to other people); or that it was more of an aspiration than a plan. The inquiry may yet work out which of these options is correct. Either way, if there was a plan it failed to survive contact with the enemy.

Dominic Cummings has his failings, but he cannot be faulted for drawing a fateful graph on a whiteboard in early 2020, by which point it is now obvious that the lack of a well-designed plan was causing issues. Namely, that there was no stream of high quality, detailed and reliable data from the nhs to Number 10 — or indeed, to anywhere. A side point in the inquiry last month was when someone (it may have been Cummings) noted that at this stage the nhs was providing “statistics”, a term I will use very lightly, via fax machine.

A robust plan would have had reams of data arriving. I don’t just mean how many infections, or deaths. I mean who and where, and when, and what they did two days prior. The very early parts of a track and trace system, in fact. Several states in East Asia managed just this. Why? They had found themselves without a nail in their horse’s shoe during sars and learned from it. Data is king. 

Cummings did the next best thing. He plotted what little data he had, and extrapolated. He found an exponential curve of death and was rightly spooked. A plan would have then considered how to analyse it with more detailed data. We have not yet got to modelling; all was doomed even before then. 

The other issue was that without a proper, prearranged workflow everything becomes reactive

The other issue was that without a proper, prearranged workflow everything becomes reactive. Cummings drew his graph in desperation, not to head something off. This is, of course, ammunition for those who argue the first lockdown was too late. That is likely something that will — because of poor data collection in those early days — always remain a known unknown. 

What do you do when your data is poor and you need to predict the future? You ask the statisticians to produce a model.

You do this even when your data is good — and you can test the predictions properly. I used to do magnetohydrodynamical simulations of star formation which, strictly speaking, are not statistical models. We did, however, run tens of them (unlike Covid-19 modelling, mine took about nine months to run). And we did something very important: we rejected any model where the results were physically unreasonable. My observational colleagues had produced terabytes of data on what star-forming regions were like, so we were able to be quite efficient at rejecting outputs.

Notice I say “terabytes” of data. Not a fax machine. This is where the use of models went fundamentally wrong. As others have written, the models themselves left rather a lot to be desired. One in particular is, shall we say, not code I ever want to see again. But, things were doomed anyway due to the lack of high quality data- gathering; not just to start things off, but to keep re-adjusting too. 

This is a continual process — and requires a well-made plan for how it all fits together. Instead, everyone in positions of influence was doubly spooked. There was Cummings’s exponential graph, and a model from Imperial College London all but predicting the end of the world, and one fax machine of actual data. The public health experts, meanwhile, had been for several weeks saying that either lockdowns don’t work, or that they are unnecessary. Probably in the face of these two scary things, SAGE had their much commented-on volte-face.

I’ve seen what afflicted SAGE happen in the world of software development. One often finds teams who are adamant that their subsystem is doing what it should, and it’s only in the face of overwhelming evidence that they switch sides, usually totally, to the reverse. Rarely though, are they totally wrong.

SAGE was hesitant-to-downright-against stringent lockdowns because of the perceived negative effects they have. Here the lack of a plan bubbles up, again. That the expert committee needed to revise its views is not wrong in itself; what is wrong is a total reversal, almost in an instant. A robust plan would have always considered the possibility of needing a stringent lockdown. Then, those negative effects would have been properly explored by academics for years beforehand.

We can’t fully blame the academy for being blind to the concept of lockdowns; just about everyone else was, too. This, though, is why good planning for emergencies is key. And this is one of the questions Lady Hallett’s Covid inquiry should be asking, but isn’t: why did no one’s “plans” — and equally no one’s research — include a lockdown before March 2020? What underlies this failure to predict?

Modern inquiries are uniquely unsuited to answer such a question because it requires asking people to admit to something they will be blamed for, at least in the “Yes, we dropped the ball” sense. But that omission needs teasing out. If lockdowns had always been on the cards, the pressure to produce a proper plan would have been so much greater. Importantly, research into how to use them effectively and proportionately would also have occurred in advance.

So, no plan, scary graphs and the experts have changed sides in a panic. Human nature means that now the decision- makers will overreact. We see this in programming. Someone finds a nasty bug in a released code, and the next thing you know there’s a panic on the mailing list to get a fix out. It usually pays, in my experience, not to rush, lest one replace defect A with defects B, C, and D. 

In March 2020, Boris Johnson and those around him did not have the luxury of not rushing. Bereft of a proper pre-made strategy and thus bereft of good data, all they had was apocalyptic models and Cummings’s not truly incorrect extrapolation. I pick on that extrapolation not to criticise per se, but because it demonstrates the mess so succinctly.

So, an extraordinarily severe lockdown was imposed, which cannot have been based on any actual evidence or data, which did not exist. Someone pulled out of the ether the idea that it would be reviewed in three weeks. The Prime Minister mumbled something about squashing Sombreros. Flattening the curve, which he was referring to, was an established epidemiological idea. SAGE had just, having been spooked, totally resiled from it. Nails, it’s all about nails to keep the shoes on.

The statistical models were mostly rubbish. They did not cause the lockdown. They did not help with keeping decision-makers grounded, but the real cause was a total information asymmetry, and no plan.

Eventually, things calmed down. Now we turn to Wales for a moment. The Welsh ministers initially copied England — one baffling difference relating to physical exercise aside. When it became possible on even the weak, partial, metrics available to ease restrictions, they decided to “play it safe” and impose this strange “stay local” rule. Unenforceably vague, and grounded in not a sliver of evidence, it was the first hint that the feedback loop was broken. 

A good inquiry would take a laser-like view of this small episode, since a lot of information is hidden within it. Sadly, it doesn’t involve any interesting gotchas, and it may require people to admit to dropping the ball, again. We see this in programming, too. I have been guilty of it: bitten by a bug, there is an instinct to over-defend against it and others. So, one finds code littered with assertions and other error traps, often in pointless places. The proper thing to do in that scenario is to use analysis tools.

The public health equivalent of that is to feed data from the coalface into the models, but also to analyse it in its own right. That way we would avoid over-reacting in the name of “playing it safe”, and instead do things measure by measure. This, of course, didn’t happen. Cummings and those around him should have predicted this, and begun to build the structures required for nuanced measures later in 2020, and into 2021.

That failure is why, when things got vexing in the autumn, the approach was simply “slam all the restrictions back on”. And why, when it was realised that the wheels were starting to fall off society and the economy, increasingly finessed exceptions were carved out. These were clearly never driven by an iota of data. Funerals are always the example cited here. Cummings did so himself. But he, and all the others, miss the real sin. Some illicit funeral gatherings still went on. Those are data points and should have been mined.

Ditto for clandestinely open pubs, or random gatherings on the beach. Or even schools. These could all have been treated as pilot projects for analysis, but this was not done in any systematic way — if at all. A proper plan would have involved doing as since the emergency started to unfold in January, well in advance of needing to do it for real.

Instead we got spooked policy makers with useless models

Instead we got spooked policy makers with useless models and experts who, having changed sides, refused to countenance switching back. A sort of anti-correcting groupthink. This causes harm. Now, we don’t have a three-week lockdown to make a plan and work it out; we have a never-ending lockdown and no clear exit path. 

Worse, the first lockdown “worked”, and became the only reliable data point in existence. Luckily vaccines were quickly developed, which ultimately ended the mess. As far as I can divine, no attempt was ever made to produce the high quality data needed to carefully impose minimal restrictions.

The parlous lack of data also worked against determining the negative effects of the lockdown. Again, a proper plan would have had provisions for gathering this data, too. If you are going to incarcerate the whole population, you should start systematically measuring the effect of it. 

To prevent this happening again, Lady Hallett’s Covid inquiry will need to go back to the 2001 foot-and-mouth crisis. That, too, was a plan-free mess: I remember seeing the pyres from my bedroom window. The relevant government departments must have learned all those years ago what “no plan” and “no data” looks like — black smoke, an acrid smell and disinfectant troughs everywhere. For some reason they chose not to make one all the same.

To do so will require calling a generation of civil servants and asking them why they dropped the ball. It may require a bit of blame, too. If Lady Hallett wants our response to the next pandemic to improve upon its failures during the last one, she should turn all her energies to finding out why the UK’s pandemic rider had no horse, let alone shoes and nails.