In the expected range, yes, and pulled forward regarding the main peak, yes, but 2022 had two additional peaks in summer and autumn. see my earlier post which has a diagram based on weekly data. Here's an update of that diagram:
More precisely, what you see is the result of the method used. As a rule of thumb, 90% of deaths involve people over 60, so this group dominates. Unusual is the succession of several bad seasons. Other disturbing results are revealed when one looks at the younger age groups as well and at what times excess mortality occurred there. This is very disturbing overall. A good metric that takes the younger cohorts into account a little better is life expectancy. Here, there has been a clear trend of improvement for decades. If life expectancy fell in a super-mortal year, it rose all the more in the following year. This pattern has been broken since 2021. Life expectancy fell in 2021 and 2022, and these declines were statistically significant.
I am a bit wary of life expectancy because it is a derived concept (in other words, a model). Shouldn't whatever shows up in life expectancy also be detectable in the mortality rates by age group?
Of course, age-graded death rates are also a model.
In the case of LE, the years of life lost are indirectly included. Not so for death rates. There, old and young are treated equally, but the mortality risk increases exponentially from 30 on. The mortality risk of e.g. the 30-40 year olds could be increased by 100% for 3 months, which would undoubtedly be a catastrophe, but in the overall mortality rate this would possibly be lost in the noise.
Insurance companies use LE, and they do not go insolvent.
However, it is as you say. You can see it in the age cohort mortality rates in appropriate time resolution.
Yes, resolution is key, with respect to both time and age. Unfortunately, we are not living in CSI world where it is always possible to zoom into video footage until the number plate can be identified. Some things can not be detected by statistical methods, but only by study of individual cases.
Thanks CM, great post. I've been wondering for a while whether a harvesting effect was to be expected after 2020/21 instead of two more over-mortal seasons.
Amazing work. So, it looks like deaths are well in the expected range for every year but pulled forward 2 to 3 months of each of the pandemic years?
In the expected range, yes, and pulled forward regarding the main peak, yes, but 2022 had two additional peaks in summer and autumn. see my earlier post which has a diagram based on weekly data. Here's an update of that diagram:
https://www.file-upload.net/download-15180129/weeklydeathsGermany.jpg.html
More precisely, what you see is the result of the method used. As a rule of thumb, 90% of deaths involve people over 60, so this group dominates. Unusual is the succession of several bad seasons. Other disturbing results are revealed when one looks at the younger age groups as well and at what times excess mortality occurred there. This is very disturbing overall. A good metric that takes the younger cohorts into account a little better is life expectancy. Here, there has been a clear trend of improvement for decades. If life expectancy fell in a super-mortal year, it rose all the more in the following year. This pattern has been broken since 2021. Life expectancy fell in 2021 and 2022, and these declines were statistically significant.
I am a bit wary of life expectancy because it is a derived concept (in other words, a model). Shouldn't whatever shows up in life expectancy also be detectable in the mortality rates by age group?
Of course, age-graded death rates are also a model.
In the case of LE, the years of life lost are indirectly included. Not so for death rates. There, old and young are treated equally, but the mortality risk increases exponentially from 30 on. The mortality risk of e.g. the 30-40 year olds could be increased by 100% for 3 months, which would undoubtedly be a catastrophe, but in the overall mortality rate this would possibly be lost in the noise.
Insurance companies use LE, and they do not go insolvent.
However, it is as you say. You can see it in the age cohort mortality rates in appropriate time resolution.
Yes, resolution is key, with respect to both time and age. Unfortunately, we are not living in CSI world where it is always possible to zoom into video footage until the number plate can be identified. Some things can not be detected by statistical methods, but only by study of individual cases.
Thanks CM, great post. I've been wondering for a while whether a harvesting effect was to be expected after 2020/21 instead of two more over-mortal seasons.