Covid-19: Reported Deaths Aren’t Currently Exceeding Actual Deaths!

I can become irrationally angry when I hear “experts” spouting crap in the media. There is such distrust of authority these days that surely it is critical that those still respected don’t allow their biases to corrupt their reason.

The More or Less program on Radio 4 last week criticised (from about 1:32 in) the reporting of 1325 deaths on Saturday 2nd Jan as the “highest daily toll”. They were correct that the number is merely the highest reported on a particular day, not the highest number that occurred on a particular day.

But MoL went on to say:

“…there’s often a delay between when a death occurs and when it is reported… If we look at the date these deaths actually occurred the highest figure so far during his second wave was on the 6th of January when 799 people died… The peak in the spring was the 8th of April when there were 1,072 deaths. Thankfully we haven’t got there quite yet.”

Wrong!!! When the program was first broadcast on 13th January we had probably already exceeded 1,072 deaths on at least one day. I’ll explain why.

There are a few issues here.

First, the reason particularly high figures are reported on particular days is that fewer deaths are registered at the weekend, so that far fewer deaths are reported on Sundays and Mondays, as is clear from the bottom graph in Fig 1 (this is official UK data):

So only the 7-day running mean is really meaningful (but watch out for Bank Holidays). I would have thought this was obvious. The latest 7-day average number of deaths reported (from Fig 1 – go to the website and hover!), for 13th Jan (i.e. for the period 10th-16th Jan) is 1,103.1.

I also recently heard recently David Speigelhalter on the radio pooh-poohing the figure of 1564 deaths that was reported on 13th Jan (the spike a few days from the right of the bottom graph in Fig 1), dismissing it as due to “lags” (sorry, I can’t find the clip – there’s no simple way to search BBC radio news output, nor an index that I’m aware of). He didn’t bother to mention the lower reporting at weekends, let alone the 7-day mean. I would have expected something rather more coherent from the Winton Professor for the Public Understanding of Risk and Fellow of Churchill College at Cambridge University.

Second, MoL didn’t fully take account of the lags in reporting deaths. Most are registered within a few days, but it takes weeks for all of them to be registered. The number of deaths reported to have occurred by 6th Jan in the latest data (published 16th Jan, as shown in Fig 1) is 884, not the 799 reported on MoL on 13th Jan. This obvious point wasn’t mentioned on the radio programme. The total for 6th Jan will probably increase further.

Third, again fairly obviously, when deaths are rising, the reported deaths, because of the lags in reporting, will underestimate the actual deaths. If this isn’t immediately apparent, imagine there are 4 deaths on Monday, 8 on Tuesday, 12 on Wednesday etc and each day you report half the deaths the previous day and a quarter of those on the day before that. So on Wednesday you’d report 5 (not the 12 that actually occurred that day or even the 8 that occurred on Tuesday), on Thursday 8, and so on.

Or we can look at the graphs in Fig 1. It might be worth emphasising that the two graphs in the figure show exactly the same deaths, just representing them differently, i.e. by date reported and by date of death. On the website (by hovering your mouse) you can see, for example, that, on 1st April, when deaths were rising, the 7-day mean of reported deaths was 544.3, but of actual deaths 724. On 1st May, though, past the peak of the first wave, the 7-day mean of reported deaths was 597, but of actual deaths was 524.1. The same thing happened in the second wave. On 1st November, when deaths were increasing, the 7-day means were 295.3 and 334.9 for reported and actual deaths, respectively. This is not coincidence. It’s a statistical effect. It will be happening in the current surge (I won’t call it a third wave, since the second wave was never really brought under control).

One point MoL might have mentioned, though, was that, in the first wave which peaked in April, the deaths due to Covid by the particular measure of those within 28 days after a positive test were understated because a lower proportion of those who died of Covid were actually tested than is currently the case. ONS death certificate data gives a better comparison, but even that is an underestimate, presumably because, at the time, some doctors just recorded “pneumonia” and suchlike as the cause of death, and didn’t mention Covid. But that wasn’t the argument the programme was actually making.

It seems to me that a possible reason for downplaying the severity of Covid-19 is unconscious (or possibly even conscious) justification of a particular position on how to deal with the epidemic. Those who are anti-lockdown and want to prioritise the economy (as they might see it – I consider the approach to be misguided ahort-termism) are – I posit – more likely to underestimate the effects of Covid-19, for example, the number of deaths caused. All the more important, then, that programmes like MoL exercise a little more care.