South Dakota

Should We Be Surprised That Case Numbers Have Been Falling?

“Scientists are scratching their heads over the precipitous decline in daily COVID-19 infections”, says a recent article in the journal Nature. “A sharp fall in the number of people testing positive has surprised scientists”, says a piece in the FT. According to the epidemiologist John Edmunds, “Nobody really knows what’s going on.”

Should scientists really be surprised by the fall in case numbers? Yes, some remaining restrictions were lifted on July 19th – the U.K.’s supposed ‘Freedom Day’. But cases have fallen in the absence of restrictions many times before. It’s therefore hardly surprising they would do so again.

To identify previous examples where infections fell in the absence of restrictions, I utilised the Oxford Blavatnik School’s COVID-19 Government Response Tracker. Specifically, I looked for examples where cases fell from a peak at a time when there were no mandatory business closures in place, and there was no mandatory stay-at-home order.

I was able to identify nine examples. (And note: one’s ability to identify examples is limited by the fact that almost all countries have had either mandatory business closures or a mandatory stay-at-home order in place during each successive wave of the virus.)

The nine examples are as follows: Sweden in the spring of 2020; Japan in the spring, the summer and the winter of 2020; North Dakota in the winter of 2020; South Dakota in the winter of 2020; Wyoming in the winter of 2020; Utah in the winter of 2020; and Iowa in the winter of 2020.

In all nine cases, infections fell in the complete absence of either mandatory business closures or a mandatory stay-at-home order. (Though in some of the cases, there were restrictions on large gatherings, or other less intrusive measures in place.)

It should be noted that all these locations other than Japan have relatively low population densities – which presumably equates to lower transmission, all else being equal. (And Japan’s “success” in dealing with the virus may be due to some cultural or biological factor that is common to every country in South East Asia.) Nonetheless, differences in population density are of degree not of kind.

So what explains the declines – did people just change their behaviour voluntarily? Not necessarily, as I’ve noted before. In South Dakota, cases began falling rapidly in mid November, despite almost no government restrictions and little change in people’s overall mobility. How could this happen?

One possible explanation is super-spreaders. We know there is substantial variation in transmissibility across individuals. Most people don’t transmit the virus to anyone; but a few people spread it to many others. Perhaps cases start declining once enough of these super-spreaders have been infected.

Whatever the true explanation, lockdowns are not necessary for infections to start falling (even if they may cause this to happen slightly earlier or slightly faster than otherwise). Why, then, are the scientists so puzzled?

One reason, as Philippe Lemoine noted in our recent interview, is that some epidemiological models simply assume that only lockdowns can have a large effect on transmission. Not particularly scientific, you might say, but that’s modelling for you.

The fact that infections have been falling in the U.K. is actually even less surprising than I’ve suggested so far. That’s because over 93% of Britons now have Covid antibodies – acquired from either vaccines or natural infection (whereas in the examples listed above, the numbers were far lower).

In summary, a decline in case numbers is only surprising if you’re reasoning from a flawed model.

If Lockdowns are Needed, Why Did More People Die in U.S. States Which Locked Down Than Those Which Did Not?

One of the great things about America is that it has 50 states that can set their own policy across a broad range of areas, including on public health and lockdowns. This has allowed some to resist the stampede to impose swingeing restrictions on normal life in the hope of limiting transmission of SARS-CoV-2, and this provides us with a valuable control group in the great lockdown experiment that can give us an idea what might have happened if we hadn’t made some intervention or other.

During the autumn and winter a new surge in Covid infections prompted most US states, like most Western countries, to reimpose restrictions. But a few resisted. Eleven states did not impose a stay-at-home order and left people at liberty to leave their homes whenever they wished. Of these, four – Florida, Georgia, South Carolina and South Dakota – did not impose any restrictions at all and treated it pretty much like any other winter.

Although there are various differences between states that might have affected Covid outcomes, because they all form part of one country there are enough similarities to make comparisons useful. In particular, if lockdowns are effective and necessary to prevent hundreds of thousands of extra deaths (or the equivalent for the size of the population), then those states which didn’t lock down should have a far worse death toll. If the death tolls are not much worse, but about the same (or better), then lockdowns cannot be having a large impact on preventing Covid deaths.

In the chart above I have used data from Worldometer to plot the current total Covid deaths per million for each state. I have coloured the 11 states which did not lock down (i.e., impose a stay-at-home order) this winter in red. I have also calculated the average for the two groups of states, those which did not lock down over the winter and those which did, and coloured them in yellow.

As you can see, states which did not lock down over the winter, far from having many times more Covid deaths, have actually had fewer – 1,671 vs 1,736 deaths per million. There may be demographic or other reasons that some states have a higher or lower number of deaths than others so we shouldn’t read too much into the precise differences. But even so, if lockdowns are supposed to suppress the virus to low levels and thus prevent ‘hundreds of thousands’ of deaths (or the population equivalent), then how is this possible? The only conclusion is that lockdowns do not work as intended and do not suppress the virus.

This conclusion is reinforced by looking at the death tolls in the four states which imposed no restrictions at all over the winter, the average of which is 1,716 deaths per million, which is still below that of those which imposed lockdowns (1,736). Florida reopened in the autumn, Georgia and South Carolina in the spring of 2020, and South Dakota never closed. Yet overall they have suffered fewer Covid deaths per million than the states which imposed stay-at-home lockdowns this winter.

Those academic teams which produce models predicting doom for places which don’t impose the measures they recommend should be challenged to apply their models to these states and hindcast the last winter. Any model which cannot accurately reproduce the known outcomes for these states should be calibrated until it can. Otherwise, if it can’t get the answer right for the past, why should we trust it for the future?

The modelling teams at Warwick, Imperial and LSHTM can be found on Twitter (as can LSHTM’s Adam Kucharski) if anyone feels like putting these questions to them.

Boris Is Wrong: The Lockdown Has Not Been “Overwhelmingly Important”

Yesterday, the Prime Minister said that the reduction in cases, hospitalisations and deaths “has not been achieved by the vaccination programme”. Rather, he claimed, “it’s the lockdown that has been overwhelmingly important in delivering this improvement in the pandemic and in the figures that we’re seeing”. While the lockdown may have had some impact on the epidemic’s trajectory, we should be very sceptical of the Prime Minister’s claim.

First, as Will Jones pointed out yesterday in Lockdown Sceptics, there are several US states where numbers fell dramatically in the absence of any lockdown: Florida, Texas, Georgia, South Dakota, South Carolina and Mississippi. And to this list, one could add Sweden. As shown below, the trajectory of deaths per million in Sweden is strikingly similar to that in the UK, even though the country has never gone into lockdown. (It should be noted, of course, that measures not based on age-adjusted excess mortality can be misleading.)

These examples do not show that lockdowns have no impact on the epidemic’s trajectory. But they do show that lockdowns are not necessary for case and death numbers to decline. Hence it is wrong to assume that, if numbers decline after a lockdown is introduced, it must have been the lockdown that caused the decline. (It might have been, but this cannot simply be assumed.)

Second, the most convincing study of the UK’s lockdowns of which I am aware (now published in Biometrics) concludes that each one was introduced only after the corresponding peak of fatal infections.

In particular, the statistician Simon Wood sought to reconstruct the actual time course of infections in England, based on available data. He notes that reported case numbers are subject to various forms of bias (e.g. non-representative samples, changes in the amount and type of testing) and that “under normal circumstances” statisticians would not “recommend attempting to estimate the effective reproduction number of the pathogen from such data”.

As an alternative, Wood used hospital death numbers (which, though imperfect, are less comprised than case numbers). In order to reconstruct the time course of infections, he combined these with the distribution of fatal disease durations (i.e., the number of days between infection and death), which he derived from the published literature.

His results are shown in the chart below. The grey dots are hospital deaths; the black line is inferred fatal infections; and the red lines are the lockdowns. As you can see, the peak of fatal infections occurs before the corresponding lockdown in each of the three cases. This finding casts serious doubt on the Prime Minister’s claim that the third lockdown has been “overwhelmingly important”.

Wood’s findings are consistent with those of economist David Paton, who notes that seven separate indicators all appear to show infections declining before the start of January’s lockdown. (Though it should be noted that parts of England were already under quite heavy restrictions when the lockdown began, and these may have contributed to the epidemic’s retreat.)

There is a large amount of evidence that lockdowns are neither necessary nor, in every case, sufficient to bring case and death numbers under control. This does not mean they have no impact on the epidemic’s trajectory, but it does mean that claims of “overwhelming” efficacy should be met with skepticism. And the best available evidence for England suggests that the infections were already declining when the third national lockdown was imposed.

Stop Press: Simon Wood, the author of the Biometrics study mentioned above, has written a piece for the Spectator responding to the Prime Minister’s comments, as well as the claim made by Imperial College that infections were surging right up until the first lockdown was imposed in March 2020.

This post has been updated.

South Dakota – the Least Restrictive State in the Western World, Yet Covid Deaths Now Averaging One a Day

As I noted back in March, South Dakota may have taken the least restrictive approach to COVID-19 of anywhere in the Western world. Its conservative governor, Kristi Noem, has been a stalwart opponent of lockdowns: when the state’s epidemic burgeoned at the end of August, there were practically no restrictions in place. Despite this, case numbers fell rapidly after reaching a peak in mid November. And by late February, they were in the low triple digits.

How has it fared since then? Case numbers have remained low, averaging about 170 per day:

And deaths have continued to fall. The last seven days saw an average of just one death per day:

Although South Dakota has the eighth highest death rate among US states, its epidemic retreated without any government intervention. (And no new restrictions were introduced in March or the first week of April.)

What’s more, South Dakota has done better economically than most other states and Western countries. According to the Bureau of Economic Analysis, its GDP fell only 1.7% in 2020 – the seventh lowest among US states. (Britain’s GDP plunged nearly 10% last year.) In addition, South Dakota currently has the lowest unemployment rate out of all 50 states – at just 2.9%.

One might say, “You can’t equate GDP and unemployment with human life”, but that simply isn’t true. The money to pay for the NHS (and programs like Medicare and Medicaid in the US) comes from taxing economic activity: the less economic activity, the less money there is available. And that’s before you even factor in the social costs of unemployment and bankruptcy.

One should be cautious about extrapolating from South Dakota to countries like Britain, given the state’s low population density. However, the trajectory of its epidemic casts serious doubt on the models that led us into lockdown.

What Happened in South Dakota?

We’re publishing a piece today by Dr Noah Carl, an independent scholar, on South Dakota. As Noah points out, South Dakota had some of the lightest restrictions in the Western world and its death toll is high compared to other US states – the eighth highest, in fact. But it has also seen cases decline rapidly since November in spite of the Governor’s laissez-faire approach, which is puzzling given that the herd immunity threshold doesn’t appear to have been reached. In the following Extract, Noah speculates about why this could be.

So, why did case numbers fall in South Dakota? I’m afraid I don’t have the answer. But here are a few possibilities. First, the herd immunity threshold is lower than 66%. This could be because the threshold has been overestimated in general, or because it is lower specifically in South Dakota, perhaps due to the state’s geography.

Second, the Google mobility index is a poor measure of the behaviours that drive transmission (as Philippe Lemoine has suggested). Perhaps South Dakotans were extra careful to practice social distancing during the month of November, even though they didn’t stop going out for retail and recreation. Weighing against this interpretation is the fact that there were dramatic changes at the start of the pandemic. Notice the precipitous decline in the retail index, and concomitant rise in the residential index, on the left-hand side of the chart.

Third, the level of immunity at which cases start declining (even if true herd immunity has not yet been reached) is much lower than 66%. This could be the case if there is substantial heterogeneity in the behaviours that drive transmission. Suppose that 80% of infections are caused by 20% of people. (Perhaps these ‘super-spreaders’ are particularly sociable, careless, or likely to interact with others by nature of their work.) Once a large enough share of the 20% has been infected, case numbers may begin falling rapidly. (This point has been made by David Dowdy.)

Worth reading in full.

Noah’s piece originally appeared in his Substack newsletter, which is worth subscribing to. He writes regularly about the pandemic.

What Happened in South Dakota?

by Noah Carl

South Dakota provides an interesting case study of what happens when the authorities make practically no attempt to check the spread of COVID-19. The state’s governor, Kristi Noem, is a stalwart small-government conservative, who has been even more defiant in her refusal to impose lockdowns than Sweden’s Anders Tegnell. She has argued that “the people themselves are primarily responsible for their safety”, and at one point claimed that a “very prominent national reporter” had praised her for proving that lockdowns were “useless”.

South Dakota’s epidemic began in the late summer of 2020; the state having been spared the first wave. Yet between August 1st 2020 and February 28th 2021 (212 days), there were only 64 days of mandatory school closures; seven days of workplace closures; seven days of bans on large events; zero days of bans on large gatherings; zero days of restrictions on public transport; zero days of mandatory stay-at-home orders; and zero days of restrictions on internal movement. In addition, Noem never issued a mandate on the wearing of face masks. The Mount Rushmore State, as it’s known, may have taken the least restrictive approach to COVID-19 of anywhere in the Western world.

As things presently stand, South Dakota is ranked eighth highest among US states for the number of confirmed COVID-19 deaths per million people – so not the worst, but well above average. (And the state may have to some extent benefited from its comparatively low population density.) However, my aim here is not to criticise or praise South Dakota’s approach to COVID-19. Rather, it is to examine the trajectory of its epidemic, and to consider what this tells us about the epidemiology of COVID-19 more broadly.

The chart below plots the daily number of cases in South Dakota since March 31st 2020, alongside the Google mobility index for residential, and the Google mobility index for retail and recreation. Each of these indexes quantifies the change in people’s movement relative to the baseline, based on smartphone data. For example, the mobility index for residential tells us how much more or less time than usual people were spending at home (–20 means 20% less; +20 means 20% more). Each line represents a seven-day moving average, so as to make the chart easier to read. (The two troughs in the red line around 350 on the x-axis correspond to Thanksgiving and Christmas, respectively.)

Looking at the blue line, case numbers accelerated up to a peak in mid November, and then fell rapidly again afterward. But this prompts a question: given the almost total lack of restrictions, what caused case numbers to fall? By the date of the peak, South Dakota had had about 65,000 cases, which equates to 7.3% of the total population. Even if, say, two thirds of those who’d been infected were not counted, that still only gets us to 22% of the population. And the herd immunity threshold for COVID-19 is thought to be around 66%, or even higher.

Another possibility is that people changed their behaviour voluntarily or in response to local measures. While this almost certainly happened to some extent, the Google mobility data does not reveal any dramatic shifts around the peak in mid November. The retail mobility index fell substantially during August and September, but then decreased gradually over the next two months. There was no sharp drop that could explain the sudden decline in cases. Likewise, the residential mobility index was mostly flat over the months of October, November and December. I checked the other Google mobility indexes (e.g. for workplaces), and none of them showed a dramatic change in mid November.

So, why did case numbers fall in South Dakota? I’m afraid I don’t have the answer. But here are a few possibilities. First, the herd immunity threshold is lower than 66%. This could be because the threshold has been overestimated in general, or because it is lower specifically in South Dakota, perhaps due to the state’s geography.

Second, the Google mobility index is a poor measure of the behaviours that drive transmission (as Philippe Lemoine has suggested). Perhaps South Dakotans were extra careful to practice social distancing during the month of November, even though they didn’t stop going out for retail and recreation. Weighing against this interpretation is the fact that there were dramatic changes at the start of the pandemic. Notice the precipitous decline in the retail index, and concomitant rise in the residential index, on the left-hand side of the chart.

Third, the level of immunity at which cases start declining (even if true herd immunity has not yet been reached) is much lower than 66%. This could be the case if there is substantial heterogeneity in the behaviours that drive transmission. Suppose that 80% of infections are caused by 20% of people. (Perhaps these ‘super-spreaders’ are particularly sociable, careless, or likely to interact with others by nature of their work.) Once a large enough share of the 20% has been infected, case numbers may begin falling rapidly. (This point has been made by David Dowdy.)

Under the governorship of Kristi Noem, South Dakota may have taken the least restrictive approach to COVID-19 of anywhere in the Western world. Although the state has racked up a high death count, its epidemic receded long before herd immunity was reached – despite almost no government restrictions, and little change in people’s overall mobility. This may be because the herd immunity threshold has been overestimated; the mobility data don’t capture changes in social distancing; or cases start declining as soon as enough ‘super-spreaders’ have been infected.

Noah Carl writes about COVID-19 and other topics in his Substack newsletter (where this article was originally published). You can follow him on Twitter @NoahCarl90.