This is an article by Rick Hayward.
[Note from Hector: this article was updated on Fri 17-7-20 to add the very interesting graph shown as figure 3(2).]
‘Follow the Science’ is the phrase frequently used in both demands and justifications concerning Covid-19.
This basic analysis is not about resolving issues around wider linguistic and philosophical notions of the ‘science’ (however badly the term is abused). Nor is it concerned with promoting false certainties about what is still unclear or unknown. It is simply about clarifying one of the essential foundations of analysis, namely available real data that has a bearing on the wider issues and the policy decisions that follow. It is an attempt to go back to the fundamental evidence that has been largely forgotten in the construction of a narrative that has its own momentum.
At base, however, is the very simple proposition of scientific investigation that, unless evidence reasonably supports a proposition in terms of probability, then the alternative or null hypothesis stands. In everyday terms, clear evidence in terms of probability is required before an assertion about a phenomenon can be sustained.
Specifically, it is incumbent on those who maintain the narrative of the unprecedented danger of Covid-19 (and thus the use of extreme measures) to substantiate their case beyond reasonable doubt. It is scepticism that should be the appropriate norm – not something to be largely excluded from the public discourse.
Even two to three months are significant ‘history’ in the context of the Corona virus and its emergence into public consciousness.
At the beginning of March 2020, this virus was perceived as an unknown quantity that had emerged in China. Clear, unadulterated information was hard to come by, and a ‘safety first’ narrative dominated – largely courtesy of the World Health Organization’s description of the situation. The narrative was given impetus by the output of one particular epidemiological model that forecast a potentially massive and unprecedented pandemic death-toll from the Covid-19 virus.
Although much still remains uncertain about the virus and its impact, key data concerning infection rates and fatality rates – as well as the nature of immunity from the virus – has steadily accumulated since the peak of in hospital deaths at the beginning of April. That accumulating knowledge has clearly pointed to a mortality much lower than originally feared. Additionally, the predictive model that supported those initial fears has been shown to be in error to a startling degree, as such models often are.
Certain ancillary issues also became clearer over time, particularly the steep age-related gradient in mortality, and, conversely, the comparatively mild effects for the majority of the population below the age of 65 in good health. Children were shown to be notably unsusceptible to the virus.
3. The Public Narrative
The tone of the public narrative about the Covid-19 virus was set in the initial weeks of comparative ignorance about its trajectory and outcomes. This ‘worst case’ scenario (apparently confirmed by the situation in Italy and Spain) set the stage for what eventually emerged as the ‘Lockdown’ policy in the UK.
This involved the implementation of measures that were, in several instances, in direct contradiction to the evaluation of evidence of their efficacy made by the WHO in October 2019 in relation to influenza epidemics (‘Non-pharmaceutical public health measures for mitigating the risk and impact of epidemic and pandemic influenza’ – World Health Organization – 2019).
Government policy was explicitly promoted by a media strategy based on the recommendations from a behavioural science sub-group of the SAGE advisory body. The aim of the policy was to influence the public in accepting severe curtailment of normal citizens’ rights with the putative aim of slowing or thwarting the spread of the virus. Fear was explicitly an important key in generating compliance.
In terms of the information provided to the public under this strategy, the following features have been consistently prominent :
(a) A focus on large numbers, quoted on a daily basis, but without any wide context or attempt at capturing relative scale.
(b) Individual cases of people affected by the virus, but again without wider context to indicate how typical or frequent such cases might be.
(c) Assumptions that the observable April spike in mortality was totally attributable to Covid-19 infection, although the revisions to the process of registering deaths had made clear attribution impossible. The mere presence of the virus could be sufficient for a death to be registered as being caused by Covid-19. Worse, in the badly affected care home sector, such an attribution could be on the basis of what was essentially guess-work. More recently, there is an increasing probability that the ‘remedy’ of Lockdown is producing increased health problems and mortality.
(d) Conversely little focus on the historical variation of infections and consequent mortality over wider timescales in order to provide a credible contextual framework. When comparative figures are quoted, there has also been little reference to proportionality in terms of a population that has grown considerably over time. The emphasis has tended to stress the exceptional nature of the current year in terms of arbitrary ‘expected’ norms.
(e) Exclusion, and in some cases direct censorship, of alternative viewpoints.
The result (whether intended or unintentional) of this media narrative has been to produce a high degree of fear and compliance amongst the population at large. Both the narrative and the consequent fear has continued, largely unabated by a changing evidence base which shows both a lesser degree of mortality (and, by extension of infection) than had been initially feared.
Bluntly: perception and reality have diverged markedly.
4. The Basis of Analysis
Against this background, and the associated perceptions generated by the partial narrative, some reference to actual comparative mortality figures is useful in order to provide context, based on real data.
The aim of the associated bar charts, therefore, is to provide some such clarity, using the actual figures for England and Wales as a basis. The reference point is simple : a description of yearly variation from the median value of mortality, aiming to provide a clear and proportionate historical context for the year of 2019/2020. The statistics involved are simply descriptive bar charts – although wider inferences do naturally arise.
No attempt is made to specifically identify deaths ‘due to’ Covid-19, since this has become impossible following the confusions that result from changes in the death registration requirements.
The figures used are those for ‘all cause’ mortality for England and Wales, issued on a weekly basis by the Office for National Statistics. These show variations over time in relative mortality which will indicate any exceptional rise related to this particular virus.
The term ‘excess mortality’ is avoided, since this tends to suggest a predictable ‘natural’ level for a given year when, in reality, it simply describes a deviation from the projection of a past trend. Turning it on its head, ‘excess mortality‘ is in fact no more than a measure of confirmed error in modelling. Death rates – as is the case for many natural phenomena – actually vary both systematically and stochastically.
The sampled time period is that of the 27 years from 1993/94 to 2019/20 and, within those years, each Winter/Spring season of 28 weeks – up to Week 20 of the second year. The total mortality for each season is the basis for analysis.
None of the following analysis depends upon inferential statistics : it is purely and simply descriptive, using data and graphics that are accessible to anyone. That is the point – the picture is so obvious and uncomplicated that it raises fundamental questions about the current standard of investigative mainstream journalism – even to the point of whether it exists in any recognisable form.
5. The Data
Figure 1 shows the actual number of deaths from all causes, by year, for each Winter/Spring period.
Over that period, seasonal deaths range from 275,488 to 359,141 – i.e. a range of 83,653 around the median value of 302,030 seen in 2000/01.
This 27-year range in deaths is equivalent to about a quarter of the maximum value – which maximum is, indeed, seen in the 2019/20 season, with registered deaths at a level of 57,111 (19%) above the median.
Figure 2 focuses in on the differences between seasons, detailing the variation of each year’s mortality from the median value. Again, the data are the raw figures for cumulative mortality.
This again clearly shows the 2019/20 as the season with the highest level of absolute mortality. The next season in terms of high mortality occurred in 1995/96, with almost 30000 fewer deaths in the equivalent period.
Thus, in terms of the absolute number of deaths recorded, it is indeed the case that the winter/spring season of 2019/20 exceeded any of the preceding 27 years. This is essentially the basis upon which the ‘dominant narrative’ (above) has been formed.
Figure 3 demonstrates the marked effect of taking relative population size into account.
Between 1993/94 and 2019/20, the population of England and Wales increased by almost 9 million, or about 18 per cent. Adjusting for population has a critical effect on the comparisons between years, as can be seen in this bar chart, which mirrors to the previous chart, but with the crucial standardisation that presents seasonal deaths as a percentage of the relevant population.
It can be seen that during the defined winter/spring period, on average, about 0.55 per cent of the population (c. 5,500 per million) died in England and Wales, with a variation of between 0.64 and 0.48 per cent.
Within this context, it can be immediately seen that the mortality of 2019/20 is not the highest for the 27 years analysed, but instead actually ranks eighth in terms of the percentage of seasonal deaths. This clearly presents a crucially different view of the relative severity of the last season, and, by extension, the impact of the virus.
Whatever might be the proportion of deaths to be accurately attributed to Covid-19, the mortality for the year when shown proportionately drops from the 100th to the 73rd centile overall. Clearly this level of overall mortality is not unprecedented in showing the impact of a specific virus.
Figure 3(2) makes a direct comparison with Figure 1, again showing total mortality for the time period, but this time related to population size. The standardised data gives the more accurate contextual picture.
Figure 4 steps back from annual data to look at the overall pattern of mortality over the full 27-year period.
A simple 3-year moving average has been calculated. Moving averages are frequently used as the basis for defining ‘excess mortality’ at a given point in time. This graph doesn’t attempt to replicate such an exercise, but is rather meant to illustrate the caveats about such use – as has been made previously.
The picture that emerges is a description of the general pattern of the data ‘as is’, showing both a wave-like regularity and year-to-year irregularities within that overall pattern. To use it to suggest a definite predictable value of mortality would be, essentially, only mathematical guess-work – not an illustration of what ‘ought’ to be.
The inherent variation in predictions of ‘excess’ mortality are well captured in a recent item from the Oxford Centre for Evidence Based Medicine that compares the predictions of a model using using a simple moving average with that using ‘harmonic regression’ (more appropriate for time series with wave-like features) :
“The ONS 5-year average is 235,293 deaths, which gives an excess of 51,486;
Using the harmonic regression trend predicts 257,081 deaths in 2020 an excess of 29,698”
Clearly, definitions of ‘excess mortality’ are open to a wide range of magnitude; they are not sufficiently stable so as to provide a clear indication of what might be considered a normative baseline. This is a fact that is perfectly well appreciated by anyone (except ICL modellers) who has attempted to make future predictions based on past data; error bands tend to rapidly diverge.
That said, the shape of the curve in Figure 4 does indicate two features that provide a further confirmation of the context for the scale of deaths in 2019/2020 :
(a) The recent trend indicates a generally rising trend of mortality, following a period (from about 2005) of overall lower mortality. The higher level in the current year is not inconsistent with that overall pattern.
(b) The relatively higher level of mortality in 2019/20 follows a year when mortality was relatively low. It is quite possible that there is a balancing connection between the mortality levels of the two years as deaths among the vulnerable population (such as we have seen in the Covid-19 period) may have been delayed until the following season.
It is interesting that the mean of the mortality in the two seasons (at 0.55% of population) is very close to the central figure for the 27-year period overall. Which would reinforce the notion of an unexceptional balancing over the two seasons)
6. Summary and Conclusion
Undoubtedly, the response to the Covid-19 pandemic has been unprecedented. That unprecedented response has been framed by a narrative that has been subject to confirmation bias of initial fears about the virus as it migrated to Europe.
The fundamental question remains as to whether the actual effects of the virus itself are ‘unprecedented’.
Although emerging data and analysis (from a starting point of relative ignorance) has tended to revise downwards any rational assessment of the severity of the virus’s impact, the dominant narrative has remained biased towards sustained hyperbole and exclusion of alternative analyses.
Now that infection appears to have fallen below what is defined as an ‘epidemic’ level, emerging and more nuanced analytical viewpoints tend to contradict self-sustaining and simplistic justifications for what may have been a reasonable precautionary response at the outset of the ‘pandemic’.
A simple examination of essential time-series data for England and Wales (countries that have been relatively badly affected), corrected for population size, shows that the Winter/Spring season of 2019/20 displays relatively high mortality but does not merit the typification of being a particularly notable exception within the longer framework of the last 27 years.
The higher level of recent mortality may also be related to the low level of the previous year, in terms of the effect on the vulnerable population. But, certainly, a range of questions remain – not least that concerning the differences between countries. Advocates of the unchanging (or little changed) policy response tend to immediately resort to the explanation that such differences are attributable to variation in ‘lockdown’ and ‘social distancing’ measures – again, an example of post-hoc justification rather than evidence-based analysis, which does not support the hypothesis.
The increasing detachment from reality is seen in the mythical status attributed to mythical beasts such as a hypothetical ‘second wave’ or the status of ‘R’ values. A brief overview of the headlines in a newspaper such as The Guardian illustrates clearly the way in which an exclusive ‘Marvel Comic’ narrative has far outpaced any reference to the much-abused ‘Science’ in any meaningful sense.
In general, much remains unknown about this virus, and only time will clarify some key issues. But it is clear that even a fairly basic analysis of essential data does not support the worst case view that has been taken, or the degree of inchoate fear that afflicts a majority of the population.