A project to estimate the effect of "corona" (as it was referred to the most in Dutch; in English better known as covid) on the weekly death counts in the Netherlands, by making no assumptions for it being there (so the impact is estimated in a corona-agnostic way).

To run this, you need to get data from the CBS (the Dutch statistics office) and put it in data/raw/ under the name Overledenen__geslacht_en_leeftijd__per_week_{date}_{time}.csv.

Note that the original analysis was done in 2020, and has not been tested with data after October 2020. A secondary analysis and write-up can be found in comparison_pkgs, although the focus is on comparing three model estimation methods/packages than on the application to this covid analysis. Read more here.

For the period up to 2020, the death counts are assumed to be made up of a weekly and a yearly component:

• the weekly component reflects weather patterns throughout a year, e.g. it being cold in the winter which will lead to more deaths, especially among the elderly
• the yearly component reflects medical advancements over time

Then, starting from a specified start date (which is the 9th week of 2020 in the code) there is a weekly additional effect estimated. This weekly effect should be 0 (or not significantly different from 0) if there is no effect of covid.

Currently, Numpyro is used to estimate this model, but Stan code to do a similar model (including some variants with extra assumptions) is also included.

By running this project, we can get output like in example_output.pdf. This shows the posterior distribution of additional deaths for the first 32 weeks following week 9, in this case for people aged 65 to 80. It can be seen that there was a clear surplus in deaths for the weeks 2 to 9, which was not accounted for by normal weekly and yearly patterns. After that, the deaths mostly follow their normal pattern again, so the estimated effects are roughly 0, although at the end you can see the start of the second wave appearing. This aligns with the pattern on https://www.worldometers.info/coronavirus/country/netherlands/

The conclusion of this is that there was a statistically significant increase in the amount of deaths during the first wave of the covid crisis. It was not the case that "people who were already dying were being attributed as covid deaths", as some people suggested back in 2020.