This blog is mostly about the measurement of excess deaths from COVID-19. The bottom line is that the crude measures one encounters are likely to be overstated by amounts like 54-63%. In looking at this technical question, the paper I found with sophisticated measures also contains this remarkable and truly sad statement:
“Third, and relatedly, our analysis suggests that the UK’s lockdown has had a net positive impact on mortalities. That is to say, it resulted in more, not less, deaths. Intuitively, this may be due to the unintended consequences of the lockdown (for example, a substantial reduction in the provision of, or access to, other forms of critical healthcare) dominating its intended consequences.”
This finding is at least as important, and likely moreso, than getting better measures of excess deaths from COVID-19.
Michael McNeill asked me an excellent question that motivated this blog:
“I read your article and others on LRC regarding the CDC’s Updated Covid numbers.
“While I tend to agree that the numbers have been grossly overestimated. at the same time,
“I came across this article related to excess mortality rates.
“What do you make of this information?”
I replied as follows with some slight editing:
“Dear Michael,
“I thought about this last night before retiring. My thought was that excess deaths should be a good measure in theory, but only if the statistical [stochastic] processes causing deaths are understood. Their predictions form the baseline of ‘expected’ deaths. There are such processes. Statistics deals in some “birth and death” processes. They are likely to be Poisson processes, or some other time series process that can be measured. There are problems with ascertaining them. The base time period is a variable. There are numerous death causes and associated [distinct] processes, so one actually has a mixture of processes to contend with. I thought that using a multi-year average would be too crude, and one would end up introducing measurement error. I thought that there would be an issue in choosing time period, whether day or week or months. [Deaths that would have occurred in the near future might be brought forward in time. What period should be used?]
“That was as far as I got. This morning I searched on measurement of excess deaths. The entry I went to is this:
“I went there because SSRN is a reputable source founded by my thesis advisor. All my work is on SSRN. You can read the abstract and the paper to get an idea of what’s involved. The authors compare the current ‘crude’ excess deaths measures with their more sophisticated measures using several factors and a Poisson process. This is along the lines of what I thought last night. Their result is of much interest:
“‘Results from two sets of identifiers indicate that, over the periods when our weekly estimates of total COVID deaths and the current excess deaths measure differ (week ending 17th or 24th April 2020 – week ending 8th May 2020), the former is considerably below the latter – on average per week 4670 deaths (54%) lower, or 4727 deaths (63%) lower, respectively.’
“Their measure results in a 54-63% reduction in excess deaths as compared with the crude measures being bandied about.
“That’s as far as I’m going with this issue, other than maybe I’ll blog it for others because the crude measures look quite startling on graphs and they seem worthy, when actually they may well not be.”
I recommend the linked SSRN paper for those who are interested in this subject.
8:46 am on August 31, 2020
