Pandemics kill people in two ways, said Chris Whitty at the start of the Covid outbreak: directly and indirectly, via disruption. He was making the case for caution amidst strong public demand for lockdown, stressing the tradeoffs. While Covid deaths were counted daily, the longer-term effects would take years to come through. The only real way of counting this would be to look at ‘excess deaths’, i.e. how many more people die every month (or year) compared to normal. That data is now coming through.
Using the most common methodology, Sweden is at the bottom – below Australia and New Zealand, which had plenty of lockdowns but very few Covid deaths. Here are the graphs that we have just published on The Spectator data hub.
Other graphs are available: so why did we use this one? On our data hub we try, wherever possible, to minimise editorialising, i.e. whereby our assumptions would in any way influence the resulting figure. But in rare instances where we do create our own figures, we use cautious assumptions and then follow the most common UK methodology: typically that of the ONS. We will also explain it fully, for maximum transparency. (An ONS article on this is here).
There is no internationally-agreed methodology for excess deaths, so we used the methodology used by the ONS and applied it to all developed (OECD) countries. A weekly snapshot is not much use when factoring in Covid and non-Covid deaths – you’d have to use a cumulative figure. So here’s our methodology:
- We pulled weekly ‘excess deaths’ data from the OECD database (which runs to ‘week 2’ – Jan 2023 – for most countries). The OECD in turn draws data from Eurostat and national governments.
- We use Jan 2020 as a starting point. (Using Feb or March makes no material change.)
- For the numerator, we have taken the cumulative deaths since Jan 2020
- And denominator: the deaths you would expect over the same period had each country continued as per the five-year period before. This is the standard methodology for ‘expected deaths’.
Like other studies (including one commissioned by Swedish newspaper Svenska Dagbladet from a statistician at the country’s equivalent of the ONS) this puts Sweden at the bottom, with just 3.3 per cent more deaths than were expected. Another way of doing this is to express excess deaths not as a percentage of the previous baseline but as a share of population. So the below chart using OECD data show it per 100,000 population: Sweden is again at the bottom.
Both of the above compare deaths against an old five-year average. But other methodologies use models to calculate how many would have died anyway, and produce a figure ‘excess’ to this modelled prediction.
One is the ‘World Mortality Dataset’ produced using data from the Max Plank Institute in Germany and others. It put Sweden seventh from bottom with an excess of 18 per cent (though this compares three years to expected deaths in one year) and New Zealand in negative territory. The New Zealand data fluctuates, though: its press has been reporting the biggest increase in deaths since the 1918 flu virus. But a country with just five million souls, a small fluctuation can produce a big ratio.
The World Mortality Dataset uses a mathematical model which takes into account seasonal and year-to-year variations to project how many would have been expected to die. The Economist magazine uses its own model to produce its own excess death figures. Its assumptions are here. Others adjust for the different age demographics from country to country; Sweden performs well on these methods too.
So which methodology to use? And should we adjust for age? And if it’s a model, what should the assumptions be? All important questions with, as yet, no consensus. The UK Statistics Authority is understood to be debating this very topic right now: ONS has announced a review into how they do it.
In an ideal world, the World Health Organisation would be doing this very important work to come up with an internationally-recognised standard. Alas, it seems in no rush. There have been some impressive amateur attempts to draw an average of methods used so far. This shows Sweden second from the bottom, bettered by Denmark. But on every measure yet published for excess deaths, Sweden comes either at or near the bottom.
PS One important caveat: since 2016 Sweden has approximately a few thousand deaths each year assigned to that year but no precise date: so they show in the annual but not weekly figures published by Eurostat or the OECD. For example, there were 3,500 undated deaths for 2022 (up to ‘week 50’, the latest data available). Eurostat calls this Sweden’s ‘week 99’ data. We have assigned these weekly deaths at a flat rate for each year: so, 70 a week for 2022. This doesn’t change Sweden’s position in the charts.
Comments