FEBRUARY 15, 2021
The lockdown is causing death, mental crisis, and decades of financial chaos. To be worthwhile, infection rates should at least rise and fall when lockdowns are lifted and applied. They don’t.
Science is brutally simple. Either a hypothesis explains the past and predicts the future, or it doesn’t. If it does, we tentatively accept it. If it doesn’t, we abandon it. If we don’t abandon it, we call the result pseudoscience. Like astrology, or water divining.
The lockdown hypothesis supposes that mandatory business closures and compulsory stay-at-home orders (“lockdowns”) provide significantly more control over the rate of spread of infection than can be obtained by people voluntarily altering their behaviour in response to common sense and straightforward guidance. If that is true, then at least three predictions must be observable:
Since, by the Government’s own estimates, lockdowns kill hundreds of thousands of people,1 if lockdowns didn’t do these things, we wouldn’t tolerate them. And the benefits would have to be so obvious that they hit you between the eyes.
Science tests its predictions with experiments. To test these predictions, we might: take a population of 66 million people; introduce a novel lethal virus; apply and lift restrictions regularly; and look for evidence of a relationship between the date of changing restrictions and the rate of fatal infections.
As it happens, we’ve done precisely that. So what were the results?
But first, why fatal infections, and not “cases”? Because we can’t harm people who would not be affected by the disease in the name of people who might be affected by the disease with interventions that we aren’t certain work. It’s the most fundamental ethical standard, summed up in the Hippocratic oath that every doctor makes: “I will abstain from all intentional wrong-doing and harm”.2
The “D” in COVID-19 stands for “disease” — the manifestation of significant symptoms arising from infection. “Cases” are not “disease”. SAR-CoV-2 and its mutations cause no- or mild symptoms in the vast majority. It may cause long term health issues in a relatively small number, but lockdown causes the deaths of hundreds of thousands of people. So we need to look at fatal infections when comparing costs and benefits.
The Office for National Statistics doesn’t publish data for the date on which people acquire fatal infections. They only publish it for the day their death is registered. But we know that the median elapsed time between exposure to infection and dying from it is around 23 days,3 and that the median elapsed time between occurrence of death and registration of death in ONS statistics is around 4 days.4 So we can work out the rate of fatal infection quite easily. It’s simply the number deaths, shifted backward 4 weeks.
To get our eye in before we look at whether lockdown events influenced the rate of fatal infection, it helps to understand some simple maths.
Infectious disease is potentially dangerous because it’s an exponential phenomenon. The daily increase or decrease in infections is proportional to the number already infected. Like bank interest rate, very small changes in infection rate give rise to very large changes over time in the number infected. The problem is, since the changes are relatively small and the time period is relatively short, it’s very hard to see from a graph of daily infection numbers whether the infection rate is changing. Is the curvy graph more curvy, or less curvy? That’s hard for our eyes to detect.
But there is a simple maths technique for making these tiny changes visible to our eyes. Because the dynamics are governed by power laws, when we take the log of the daily infection rate, the tangent (‘slope’) in the resulting graph corresponds to the rate that infection growth is changing at that moment.5 And our eye is very much better at seeing ‘slope’ than ‘curviness’. When the number being infected is rising, the slope points upward. If that number increases, the slope steepens upward. When the number being infected is constant, the slope is level. If the number of infected is falling, the slope points downward. If that number falls faster, the slope steepens downward.
A sketch will make this more obvious. Figure 1 illustrates what the lockdown hypothesis predicts should happen to hypothetical death rates when restrictions are not applied, or applied and lifted, both as a normal curve of daily deaths, and the log-transform of those deaths. See how the slope is easier to see in the log-transformed (red) version? It’s like an X-Ray of our data.
Figure 2 shows the standard (blue) and log-transformed (red) graphs of daily fatal infections, from the start of the epidemic in February 2020 to the annual rise in seasonal illness that begins at the end of every Summer. The first recorded death in the ONS data is 9th March, giving the first estimated date of fatal infection on 11th February.
Prediction 1 says: “When no controls are applied, the rate of infection should not fall”. Remember—this means that a log-transformed plot of daily fatal infection before lockdown is imposed should be a straight line. It’s not. Figure 3 shows the fatal infection rate from the start of the epidemic until the first lockdown on 23 March. By the beginning of March, the slope of log-transformed infection rate was falling . By the night of Mr Johnstone’s “stay at home” TV broadcast, without any mandatory controls in place, growth in new infections had fallen to zero. This should not surprise you: after you heard about people eating infected pangolins, did you wait until Mr Johnson told you to before washing your hands more often and taking more care around other people? No. Neither did 66 million other people.
Prediction 2 says: “If controls are applied, the rate of infection should fall”. Figure 4 shows the period from imposition of controls on 23rd March until the annual rise in seasonal respiratory infection that begins every year at the start of autumn. Infections did indeed fall — at a constant rate. But that was preceded by the rapid fall to zero in the growth rate, without any controls, from a high value before the start of lockdown #1. We have no reason to suppose that the rapidly falling growth rate in infections prior to the imposition of lockdown would not have continued in its absence, and progressed into a fall. Indeed, this is the basis of the Gompertz function model of epidemic growth and contraction (see postscript, below). But such coincidences are the lifeblood of tabloid horoscopes.
At the end of the summer, the Government introduced mandatory mask wearing. With the sensitivity of log-transformed data, we can see changes in the rate of the very small infection numbers around that time. However, by that time, fatal infections had fallen to fewer than 100 per day, excess mortality had fallen below the 5 year average, and the UK was experiencing the highest heatwave excess mortality since 2003. Of this, ONS notes: “Further work is required to explore how the concurrent risk of COVID-19 and heatwaves may have intersected to amplify these impacts”.6 If you are going to run an experiment to see if facemasks protect the vulnerable, don’t run it when the vulnerable are being killed by heatwaves.
Prediction 3 says: “If controls are lifted, the rate of infection should rise”. Here, the impression that the Government were driving a clownshow bus with a disconnected steering wheel is starkest. On three occasions (Figure 4 again), it lifted controls: the partial lifting of lockdown #1 on May 10th; reopening of retail around June 15th; and reopening the country on July 4th. In theory, the infection rate should have risen sharply on each occasion. It didn’t on any occasion. The virus cleared the population in the way viruses have done for millions of years — at its own pace.
Any lockdown enthusiast who views this predictive failure of their hypothesis and says: “Well, if you stand on one leg, shut one eye, and hold the graph up to a full moon, you can see a change on this date” is missing the point. To justify a policy that kills hundreds of thousands of people, the effect should be so clear that it leaps off the page. It is not.
And this is where scientists and pseudoscientists diverge. In science, when we fail to find evidence to support our claim, or find evidence falsifying it, we abandon the claim. In pseudoscience, they invent another claim to explain the failure of the first claim.
In cults, they ban scepticism. But that is for a future essay.
Did any benefits of lockdown “hit you between the eyes”? Were any of its propositions visible in the data? No. Yet lockdowns are bitterly defended with a pseudoscientific worldview exemplified by the moral rectitude of believers such as George Monbiot — a journalist who has never missed a pay check while, from his comfortable metropolitan home, dispensing intolerant prescriptions about the need to ban scepticism about his belief.7
Hungarian philosopher Imre Lakatos summed up the nature of their pseudoscientific world view thus:8
An intellectual crime. And, and time may show, perhaps a material one.
Infection rate in epidemics follow a well understood curve, described mathematically by the Gompertz function. The great majority of COVID-19 outbreaks, including the UKs, can be modelled accurately by it, as shown in Rypdal (2020).9
Rypdal (2020) show that the peak rate of fatal infection can be influenced by the effectiveness of changes in public behaviours that affect the transmissibility of the disease. To the extent that these prevent saturating hospital capacity, this would lower cumulative death rate. However, Bendavid, E. et al. (2021)10 show that enforcement offers no advantage over voluntary behaviour changes in the death rates experienced by countries implementing a wide range in severity of lockdown. Equally, while such considerations might alter the timing over a short period of the dates of deaths, there is no evidence (beside avoiding saturating hospitals) that they alter the total number, as visible in Figure 5.
For an accessible introduction to exponential growth, see for example section 2.1 in Hobbie, R., & Roth, B. (2015). Intermediate Physics for Medicine and Biology (5th ed.). Springer ↩︎
Lakatos, I. (1980). The Methodology of Scientific Research Programmes: Volume 1. Cambridge University Press. For a fascinating lecture by Lakatos on the difference between science and pseudoscience, listen to his BBC Radio Talk here: link ↩︎
Rypdal, K. and Rypdal, M. (2020) ‘A parsimonious description and cross-country analysis of covid-19 epidemic curves’, International Journal of Environmental Research and Public Health, 17(18), pp. 1–23. doi: 10.3390/ijerph17186487. ↩︎
Bendavid, E. et al. (2021) ‘Assessing Mandatory Stay‐at‐Home and Business Closure Effects on the Spread of COVID‐19’, European Journal of Clinical Investigation, pp. 0–3. doi: 10.1111/eci.13484. ↩︎