Friday, 5 February 2021

Do lockdowns work?


Toby Young has responded to my Quillette article about COVID cranks. He says he 'won’t bother responding to [my] detailed criticisms of Ivor Cummins and Michael Yeadon because I don’t think the case against the lockdown policy stands or falls on whether their analysis is correct.'

Fair enough, but it was the claims made by such people that were the focus of my piece. Those who want a more liberal approach have been sullied by their association with such crackpots. That is regrettable because it gives the 'Zero Covid' fanatics the edge in the debate about when to end the current lockdown.

Toby's argument, which has been circulating on social media ever since the 'casedemic' theory collapsed so spectacularly, is that lockdowns simply don't work.

Our contention is that the whole panoply of non-pharmaceutical interventions (NPIs) that governments around the world have used to try and control the pandemic—closing schools and gyms, shutting non-essential shops, banning household mixing, restricting travel, telling people they can’t leave their homes without a reasonable excuse, etc.—have been largely ineffective.

How can this be? As far as I'm aware, one of the few areas of agreement remaining between the sceptics and the mainstream is that SARS-CoV-2 is a communicable disease that is transmitted from human to human, especially in poorly ventilated indoor spaces. If human interaction is severely curtailed, the infection rate can be reduced below 1 and the number of cases will fall.

This does not necessarily require lockdowns. People could do it voluntarily. But if they don't do it voluntarily, I would argue that government coercion is justified under the harm principle, a concept with which classical liberals should be familiar. 
So whilst lockdowns may not be a necessary condition for reducing the infection rate, they are a sufficient condition (assuming, of course, that people abide by the law). Toby is arguing that they are neither. In fact, he goes even further than that by claiming that none of non-pharmaceutical interventions work, not even travel restrictions.

He concedes that there are 'some peer-reviewed studies published in reputable journals seeming to show that these measures reduce COVID-19 infections, hospital admissions, and deaths'. He only links to one of them, but there are many others. In any case, he says that 'most of these rely on epidemiological models that make unfalsifiable claims about how many people would have died if governments had just sat on their hands'. 

Instead, Toby links to a blog post which supposedly reviews 30 studies which provide 'evidence that lockdowns don’t work ... not based on conjecture but on observing the effects of lockdowns in different countries.' I have seen this blog post touted around Twitter for weeks. Of the 30 (actually 31) 'studies', only seven have been published in journals in the last twelve months. Many of them are pre-prints that have not been peer-reviewed. Some are self-published PDFs. Three are newspaper or magazine articles. One is a blog post. Many of them have only a tenuous relationship to the question of whether COVID-19 lockdowns have worked. One of them was published in 2006, another in 2008. Quite a few of them involve the kind of modelling Toby says is no good.

Of the seven published studies, one is only available as an abstract and seems to be more of an opinion piece than a research paper. One is about quarantined US Marines and has little relevance to the question at hand. Another is a research letter that is almost entirely irrelevant. 

Only five of them broadly support Toby's argument. One of them found that various factors, such as life expectancy, are associated with the number of COVID-19 deaths in different countries, but lockdown stringency isn't. This study came to a similar conclusion. This study from New Zealand, where there have been only 25 deaths from COVID-19 so far, makes the bold claim that 'lockdowns do not reduce COVID deaths'. 
Another study looked at the factors associated with higher Covid death rates at the national level and found that 'border closures, full lockdowns, and a high rate of COVID-19 testing were not associated with statistically significant reductions in the number of critical cases or overall mortality'. However, it also found that full lockdowns were associated with higher recovery rates which suggests that 'full lockdowns and early border closures may lessen the peak of transmission, and thus prevent health system overcapacity' (which is the justification for lockdowns in the UK)

Finally, there is this study which concluded that lockdowns suppress the virus in the short term, but lead to more deaths in the second wave 'unless herd immunity is achieved by vaccination, which is not considered in the model'. 

Vaccination is the crucial missing variable. There are two versions of the 'lockdowns don't work' argument. I'll come to the 'hard' version in a moment, but the 'soft' version says that, in the absence of a vaccine, the virus will infect a certain proportion of the population sooner or later so you might as well take it on the chin. Given the social and economic costs of going in and out of lockdown indefinitely, there is a lot of truth to this and the handful of peer-reviewed studies in the blog post Toby links to address that issue.

But there are two big problems with it. Firstly, many lockdowns - including the ones in Britain - were not so much designed to reduce the overall death toll as to prevent the humanitarian disaster of having many millions of people infected simultaneously and overwhelming the health service (which would itself lead to more people dying). If you accept that avoiding that is as a legitimate policy goal, using lockdowns to flatten the curve makes sense.

Secondly, we have multiple vaccines now. Unlike in March and November, it can no longer be argued that lockdowns merely delay the inevitable. Lockdowns do work if you have a destination.

Saying lockdowns don't work because they don't get rid of the virus is to argue against something nobody has claimed. We always knew that the spring lockdown would only push the problem into the winter. Neil Ferguson's much hated (by sceptics) model for Imperial College in March made that perfectly clear. The government was buying time, hoping that salvation would come in the form of Test and Trace (it didn't) and vaccines (it did). 

The 'hard' version of the argument says that lockdowns don't even work in the short term. They don't reduce the infection rate, even temporarily. Given the existence of germ theory, this is an extremely difficult idea to swallow and yet Toby seems to have done so. With reference to his '30 studies', he writes:

What these data seem to show is that the SARS-CoV-2 epidemic in each country rises and falls—and then rises and falls again, although less steeply as the virus moves towards endemic equilibrium—according to a similar pattern regardless of what NPIs governments impose.

This is not true. Even the studies that can be used to support the soft version of the anti-lockdown argument do not support this argument, which amounts to 'viruses gonna virus'.  

If lockdowns work, you’d expect to see an inverse correlation between the severity of the NPIs a country puts in place and the number of COVID deaths per capita, but you don’t
Toby is now conflating the hard and soft arguments. The test of whether lockdowns reduce the infection rate is whether the infection rate falls after a lockdown is introduced, not whether the country has a high or low COVID death rate overall. Lockdowns can work (in the sense of reducing the number of cases in the short term) in countries that have already had many COVID deaths, as Britain and much of Europe have shown in the last twelve months. Countries lock down hard precisely because they've got high death rates.
On the contrary, deaths per million were actually lower in those US states that didn’t shut down than in those that did—at least in the first seven-and-a-half months of last year.
This is not relevant to the question of whether lockdowns reduce the caseload, but it's worth noting that 'at least in the first seven-and-a-half months of last year' is doing a lot of heavy lifting here. Toby's source is a blog post from the American Institute for Economic Research published in August. It looks at 'the only seven states to never issue stay at home orders' (Utah, South Dakota, North Dakota, Nebraska, Iowa, Wyoming, and Arkansas) and claims that they are 'amongst the top performers in terms of minimizing deaths per capita'.
It didn't age well. By November, the pandemic in North Dakota and South Dakota was 'as bad as it gets anywhere in the world'. Both states have an even higher COVID-19 death rate than the UK. Arkansas's COVID death rate has risen from 21 per 100,000 to 166 per 100,000. Iowa's is 160 per 100,000. Nebraska and Wyoming have a similar death rate to California, which is to say a high one. With the exception of Utah, all the 'top performing' states have seen COVID-19 kill at least one in a thousand citizens so far. 
(Incidentally, the current smiley favourite, Florida, has a COVID death rate of 127 per 100,000, putting it on a par with Spain and Peru. Not a massive success story.)
So do lockdowns reduce the infection rate or is the virus gonna virus?  Thanks to mass testing, we have accurate data on case numbers for many countries. People tend to get tested around 5-10 days after they get infected (ie. when they get symptoms) so we can see whether infection rates fell after a lockdown was introduced by seeing what happened to the case numbers a week or so later.
Here is France, for example. The lockdown began on 31 October and cases started to fall steeply from 8 November. The lockdown began to be eased on 28 November after which case numbers began to rise.


This is Israel. It had a big spike in cases in August and September (contrary to the theory that COVID-19 is a seasonal virus) which it addressed with a lockdown beginning 18 September. Cases began falling from 27 September. In the subsequent wave, it introduced a hard lockdown on 7 January and case numbers fell from the 17 January. 


This is the Czech Republic, or Czechia as it now prefers to be known. It introduced a lockdown on 22 October and cases started falling after the 27 October. A subsequent lockdown began on 27 December, with case numbers peaking on 7 January.


This is Ireland. It had a huge outbreak over the Christmas period which was rapidly brought under control by a lockdown that began on 31 December. Case numbers peaked on 8 January. 

This is Wales. It introduced its 'firebreak' lockdown on 23 October. Cases peaked on 29 October before falling by half. 

I could go on, and you can look at the figures yourself if you still doubt me, but let's finish with England. England had a lockdown on 5 November, case numbers peaked on 9 November and then fell by half. It has been claimed that infections started falling just before the lockdown, but that's not obvious from this data. I remember thinking at the time that rates looked pretty flat and being suspicious of Chris Whitty's dodgy dossier, which is why I opposed that lockdown, but there was no sign of rates falling and there was certainly no reason to think they would fall by half. Why would they?

The third lockdown began on 6 January, but case numbers started falling at the same time which suggests that infections peaked at the very end of December or the very start of January. Other evidence confirms this and sceptics have leapt on this as evidence that lockdowns don't work.

Actually, it is evidence that lockdowns are not the only thing that works - and the other things that work are very close to being lockdowns. London and large parts of the south east were put into Tier 4 on 20 December. The schools were closed for the Christmas holidays at around the same time. This meant that the most heavily populated parts of the country were already in lockdown in all but name. Everywhere else was in Tier 3. There was not a pub or restaurant open in the country. Many workplaces were closed over Christmas. 

We can see from the mobility data that things quietened down over the Christmas period. It was a very unusual Christmas.

If you want to argue that closing all the schools and most of the workplaces and putting the whole country in Tiers 3 and 4 has similar effects to a full lockdown, you won't get much argument from me, but don't pretend infection rates were falling under a laissez-faire regime.

Alternatively, you can argue - as Toby appears to be doing and many smileys certainly do - that infection rates are unaffected by any restrictions on human interaction and that it is sheer coincidence that lockdowns have been accompanied by a sharp decline in case numbers in the UK and elsewhere time and time again. He seems to think it was a fluke in January, a fluke in November and a fluke in March.

Incidentally, Snowdon’s claim that the first British lockdown reduced COVID infections is easy to debunk. You just look at when deaths peaked in England and Wales—April 8th—go back three weeks, which is the estimated time from infection to death among the roughly one in 400 infected people who succumb to the disease, and you get to March 19th, indicating infections peaked five days before the lockdown was imposed.

This theory remains contentious as we didn't have mass testing at the time. When BBC More or Less looked at the issue, they found it depended on whether you used the mean or modal average. But let's say it's true. People were told to avoid unnecessary travel and contact with others on 16 March. Vulnerable people were told to shield themselves for twelve weeks. Pubs, restaurants, gyms and schools were closed by law on 20 March.
It is possible that this was enough to bring R below 1, but it is equally possible that R would have gone back above 1 after the initial panic subsided and people resumed human interaction. Either way, you have to accept that human interaction spreads the virus. If you're worried about the social and economic damage done by people not going out mixing and socialising, it shouldn't really matter whether it is the result of voluntary or mandatory measures. If you don't think lockdowns make much difference to people's behaviour, they can't make much difference to the economy. You can't have it both ways.
If, on the other hand, you want people to be mixing and socialising, don't pretend that the way people behave in the days leading up to a lockdown are representative of how they would behave if there was no lockdown at all. 

And if your argument is that human interaction is irrelevant and the virus is gonna virus, then there's nothing left to be said. Wallow in ignorance if that's your scene, but don't expect be taken seriously by anyone outside the smiley cult. 

The number of cases doubled in England in December to 30,000 a day, which is more than we had when the November lockdown was announced. In the second half of the month, they doubled again. We have seen the consequences of that in recent weeks, with more than a thousand COVID deaths a day and nearly 40,000 people in hospital with the disease. 

Imagine if it had doubled again. And again. At some point it would have petered out, but with 80% of the population vulnerable to catching it - and it being the middle of winter - there was no reason to think that exponential growth wouldn't continue for at least a little while longer. 

Presumably, those who believe that non-pharmaceutical interventions are 'largely ineffective' believe that the infection rate would have fallen by two-thirds in recent weeks, as it has, if we'd have carried on as normal. They must also believe that rates would have quadrupled in December, as they did, if we'd been in full lockdown from late November. They must think it is just good luck that SARS-CoV-2 is no longer circulating in Wuhan and New Zealand. Virus gonna virus, right? It's got a mind of its own.

It's absurd thinking, but the sceptics have finally found an argument that cannot be categorically disproved. Lockdowns have a scientific rational: you can't transmit a virus to people you don't meet. Contrary to what Toby says in his article, they also have historic precedents: during the Spanish Flu, cities such as Philadelphia closed shops, churches, schools, bars and restaurants by law (they also made face masks mandatory). And now we have numerous natural experiments from around the world showing that infection rates fall when lockdowns are introduced.

In the final analysis, the smileys yelp 'correlation is not causation'. It never is. The link between smoking and lung cancer is still only a correlation, but there comes a point when the correlations are so strong and the alternative explanations are so far-fetched that it becomes proof in the mind of any fair-minded individual.
But even though sceptics lack a convincing hypothesis to explain why lockdowns don’t work, I don’t think we’re obliged to come up with one. Surely, the onus should be on governments to show that lockdowns work if they’re going to suspend their citizens’ civil liberties? For me, as a classical liberal, this is the most decisive argument against the draconian controls democratic governments have imposed in the hope of mitigating the impact of the virus.
To be fair to the government, it has tried to persuade people and it has succeeded in most cases - which is going to have to suffice when a democracy is faced with a collective action problem. Toby complains that the case for lockdowns rests on 'epidemiological models that make unfalsifiable claims about how many people would have died if governments had just sat on their hands'. But you don't need a model to know that an uncontrolled epidemic is going to get out of control. The models try to quantify how bad things will get - not always accurately - but the basic assumption that a communicable disease will spread exponentially unless action is taken should not be up for debate. 
Unless there are certain agreed facts, there is no room for a productive conversation. We have a highly infectious disease that kills around one per cent of those who contract it and which is a huge drain on healthcare resources. Most people have not yet contracted it and therefore do not have immunity to it. It is airborne and spreads via human contact. Whenever restrictions have been relaxed we have seen rates rise, particularly in the winter. Voluntary behavioural change may have been enough to control the virus in some parts of the world, at least for a while, but it has never been enough to produce a sustained fall in the UK. Around 20% of the population have had the virus, of whom over 100,000 have died. If 40% got it, 200,000 would be dead. If 60% got it, 300,000 would be dead. And so on.
This is just science and the remorseless logic of exponential growth. If the sceptics want to argue with biology and mathematics, then I'm afraid the onus is on them to make their case. Both sides are dealing with counterfactuals and therefore with unfalsifiable propositions. Absolute proof is impossible, but those who believe that lockdowns reduce the infection rate have a solid theoretical basis for their view which is backed up by ample empirical evidence from around the world. 

The sceptics, by contrast, have fallen for every crackpot theory going for the last six months and have finally settled on a claim that can never be fully demolished; the idea that whatever happens would have happened anyway. Thanks to mass vaccination, the virus should fade away for good over the next few months and the sceptics can tell us forevermore than the government overreacted to a December spike of a virus that was already on its way out; that a virus which naturally peaked in late March last year naturally peaked in early January this year. And you will never be able to convince them otherwise.
Toby makes a fair point when he says that the government has never produced a solid cost-benefit analysis, but such an analysis would be rejected out of hand by the sceptics because it could only be based on 'unfalsifiable claims about how many people would have died'. Given that the average person who dies from COVID-19 loses, on average, ten years of life, and that much of the economic damage is caused by people's response to the virus rather than the lockdown, any realistic estimate of how many deaths have been prevented - combined with the standard valuation of a quality-adjusted life year - would, I am sure, show that this lockdown has been worthwhile. 

That could change as the lockdown wears on, the vaccinations kick in and the number of deaths being prevented declines. At that point, the public debate would benefit from some level-headed economic analysis from lockdown sceptics, but who, by that time, will still be listening to them?

UPDATE: This debate was continued here.

No comments:

Post a comment

Comments are only moderated after 14 days.