Allow me to continue my musings on deaths in Germany. First, as Joel Smalley explained, “there are differences in mortality distributions between sexes (male and female). Stratifying by sex will yield better results than aggregating sex.”
No problem. For example, here is the final diagram from my previous post (average daily deaths per month, scaled per age cohort to the population as of 31.12.2022), but restricted to deaths of women:
And then for men:
The general shapes are very similar. For women, both the February/March peaks of old and the 2022 autumn disaster are more pronounced, reflecting the much larger share of women among very old people.
In both diagrams I introduced yellow for the 2023/2024 season. Late reporting of deaths is less of a problem in Germany; the numbers for July and August can be considered stable. But what is going on there? Aren’t we all supposed to be desiccating, or melting, or boiling?
The Robert Koch-Institut (RKI), anyway, has been trying to transfer some of its pandemic power into the heat deaths business. Their long-running models (h/t Witzbold) have been brushed up and are now producing continuous updates, supplying us with weekly heat deaths porn. More than 70,000 heat deaths are being proclaimed for 2000 to 2023, with annual figures ranging from 100 (in 2011) to 9,500 (in 2003). And 2023 currently ranks 9th of 23, with 3,100 heat deaths and counting – almost at the level of 2022 (4,500, rank 6). What is going on there?
For a first visual check, I repeated last year’s experiment, and plotted maximum daily temperature in Frankfurt (degree Celsius, resolution 0.1 degrees, horizontal axis) against deaths (vertical axis), for all the months of July and August from 2000 to 2023. The grey dots are 2000-2021, the red dots 2022, and the green dots 2023.
Yes, of course there is correlation between temperature and number of deaths. That’s a fact of life, and no amount of pearl-clutching is going to do anything about it. But 2023 is unremarkable, considering that in this diagram there is no correction for size and age structure of the population (the data required are only available, to the general public, at monthly resolution). The summer of 2022, however, was quite deadly, and this can not be blamed on heat alone. And these conclusions are not due to the choice of Frankfurt as reference for temperature; the resulting diagrams are similar for Munich and Berlin, and also for Frankfurt after introduction of time lag (hot today, die tomorrow).
I restricted to July and August in the above diagram because these months are hottest, and most heat deaths can be expected to occur then. Let’s call July plus August (JA) the “bad” (or “red”) summer, and June plus September (JS) the “good” (or “green”) summer. In the following diagram I show JA deaths (red, solid) and JS deaths (green, solid). Now, since July and August have 31 days each, and June and September only 30 days each, it also makes sense to look at JS deaths scaled by 31/30 (green, dotted). And finally, assuming that most heat deaths are JA deaths, I subtracted the RKI figures from JA deaths to get some idea of non-heat deaths (red, dotted). Note that I had to estimate September 2023 deaths by scaling the first 24 days to 30.
Until 2020, the RKI model seems to have done a decent job of ironing out the wrinkles in the JA deaths, bringing the residual (red, dotted) close to what might be expected from JS (green, dotted). In other words, it has tracked the difference JA – JS between “bad” and “good” summer. The following diagram shows this difference (orange, solid) and the RKI heat deaths figures (blue, solid). Since what happened from 2020 to 2022 might in part be attributed to Covid, the dotted orange line adjusts for (official) Covid deaths.
It also makes sense to adjust the orange lines for number of days (factor 31/30):
By visual inspection, it becomes clear that correlation between blue and orange must be high. In fact, Pearson correlation amounts to at least 89% until 2018, drops to around 85% when 2019 and 2020 are included, and falls to 73% until 2023 (maybe 77% after correction for Covid deaths).
Two explanations are possible:
The model is unchanged, and is losing stability (explanatory power through time) because some unknown factors are influencing (causes of) deaths, or
The model has been changed, maybe in order to come up with numbers of heat deaths that provide a better fit with the narrative.
Whatever. My trust in the RKI’s modelling abilities is limited. Note that Matthias an der Heiden is involved here, who also produced formidable nonsense (comment in German, sorry) on the effects of Covid measures on Covid case rates. Thank you for your consistency in the spending of my tax money.
Bravo. Thanks.
Achte darauf, deinen Namen aus den Eigenschaften zu entfernen!