MY late mother, rather gloomily, once told me that death was the ‘perfect statistic’: it came to us all and was completely unambiguous. You were dead or you weren’t. (The only known modern exception to this rule is Theresa May, whose bloodless countenance suggests that she has spent rather too long in the company of a notorious Carpathian Count.)
Covid-19 proved my mother wrong. Death, unambiguous, final death, has unsurprisingly become the salient statistic by which the success or otherwise of Covid-19 policy is measured. It is nonetheless disastrous that it is so, as I’ll explain in what I apologise for in advance is a very geeky and, to some, no doubt cold-hearted blog.
By themselves, recorded deaths are data, not information, information being data with value. When starting out on modelling anything, you first have to ask yourself what is the ‘value’ of this data? What information I am trying ultimately to glean from it? That may sound ridiculously obvious, but as a data quality engineer, I can assure you that both what the data actually measures, and what value you wish to extract from it, are very often simply assumed. Even having decided these questions, before modelling can begin you must know whether the dataset you have collected is up to scratch. How complete is it? Has it got missing values? Are there intrinsic biases? If so, can they reasonably be corrected?
After all the brouhaha of the past few weeks, this vital truth has been missed: Professor Ferguson has rightly got a kicking for his execrable coding standards and the uncertainties it has introduced into his modelling, but the fact remains that the model was modelling the wrong thing.
At the risk of sounding callous, the fact that someone died while infected with coronavirus is not particular useful: much more informative is did they die with coronavirus, or from coronavirus? The distinction plainly matters enormously, and the failure of the government to make it, as well as the irresponsibility of the media, means our economy will now almost certainly undergo the worst recession for 300 years.
Let’s assume in some better parallel universe that the government did avoid disaster by conveying the truth that coronavirus is not a grave threat to the young and healthy; even then, our work is not done. Is public policy really to be oriented towards stopping people dying? No, afraid not. Death is certain one day. What we should really be trying to minimise is life shortening, and when we view coronavirus policy in such a light the glaring holes appear: most of those who die with it – though by no means all – appear to be frail and in their twilight years. However, there have been frightening reports of a surge of non-Covid-related excess deaths during the lockdown period, plus a huge drop in cancer screening, no doubt eventually costing thousands of lives. Were these non-Covid and Covid-related cohorts medically and demographically different? Can they be modelled? Is the continuing lockdown, never mind the huge recession to follow, therefore going to lead to more aggregate life-shortening than ending it now would?
Whether or not lockdown was justified originally, it is surely crucial to ask these questions now. However, I fear that we will not do so: in our rich and plentiful lives, death is regarded not as an inevitability but as a scandal. That does not, of course, absolve politicians from treating us like adults and levelling with us about the hard choices that must be made. Nor, I am afraid, does it absolve the even more appalling performance of the Church, which for once had a crucial role to play in comforting the population and preparing them to face these intensely painful realities, but instead opted to excommunicate itself from the British people.
Of course, there are even larger and better philosophical questions concerning lockdown, some even more important than individual life and death: the long-term effect on freedom, state power and our culture, quality of life for example, and so on. However, even in the relatively narrow confines of modelling we are failing: the most sophisticated, most brilliantly coded models, fed with tip-top quality data, are worthless if the question you ask is wrong.
