Do the UK’s cities hide Sweden’s 3x greater spread?

This piece by Daniel Hannan, whom I’d respected for having a deeper understanding on subjects up to this point, has lit my wick. I’m no lockdown fan, but it seems naively predicated on the idea that capacity for spread in Sweden and the UK is entirely comparable, and that measures we took were not needed at all, and that we would have been absolutely fine with just those in Sweden.

At the heart of this flawed argument is an attempt to apply a correct judgement about density, to the wrong data for the two countries.

In considering a comparison, comparing deaths/1m is far better than comparing cases. The former is real, the latter always a sample. As 1st world countries with very similar demographics by age, on the surface the two at 580 for Sweden and 614 for the UK, look similar. Lockdown fans push back with the argument that Sweden has a density of 25/sqkm and the UK 275, miltant sceptics quite rightly refute this saying Sweden is far more uninhabited, and people are more concentrated in urban areas. It doesn’t help that Google confuses matters. A quick search suggests London is 5.7k/sqkm and Stockholm 4.8k (which backs up the sceptics), and feeling confident in this militant sceptics assume that given d/1m are the same London’s extra measures must have been pointless.

However, this is naive. If you take population and density data by district for the two countries, unitise them to make sure they all occupy equal space in the plot, rank them, you get this. The first red band represents Stockholm.

In doing so it becomes clear just how much denser the UK is than Sweden. 10% of the UK, 50% of all of Sweden, is up to 3.5x denser, the rest pretty much matches up, apart from the bottom 20% of the population, which lives at density levels we don’t even have in Wales or Scotland. If you population weight the densities, you get 2155 as the UK average, and 780 for Sweden, about 2.8x denser.

So, as Disraeli would say Google is giving us “lies, damn lies and statistics”.

So how can we go about comparing like for like spread?

Without agreeing that density is a factor, there is no foundation for debate on this subject. So let’s first get to a point of agreement on this.

Here zoomed out is a view of deaths/1m for the UK from this excellent ONS interactive map.

What jumps out immediately is that spread has been no way even, and looking at it in some detail it starts to become apparent that more densely populated and interconnected areas have seen far higher concentrations of death.

The worst Newham in London with 1440d/1m, least South Hams in Cornwall 40d/1m.

Here is a similar (per 1m not 100k) view of the data for Sweden, from this equally excellent interactive map on this Swedish site aggregating govt data from different regions, and it shows in broad terms the same pattern.

So if we concede density has an effect within a country, it does not seem a leap of faith to me to suggest that finding two parts of each country, with similar population demographics, and comparing deaths/1m, we should get an idea of relative spread — in fact militants would argue they are doing so for Stockholm and London.

Let’s consider the cities by first looking at their constituent parts. Newham is part of London, which is a really massive two tier metro area, Greater London (9m), which is part of the London Metro Area (14.4m). Similarly, Stockholm City is the middle of the Stockholm County metro area.

Newham has 350k residents and a density of 9.7k/sqkm. This jumps out straight away, Google had 5.7k for London? We know from the plot above Sweden has no district that dense, but Newham is in fact only the 11th densest borough in London (here is the list for all places in the UK, the top 24 are all more dense), leading the way is Islington with a density of 16.1k/sqkm, near twice that of Newham, and all of these denser boroughs are further in the middle, so something is not right about Google’s number. Add to this that Islington has only seen 810d/1m.

Why so much lower than Newham?

All the epidemological models agree that a huge modifier of r0 is working from home, and this is likely the key, Islington just has far more white collar professionals than Newham, and their employers have just made it far easier for them to earn from home. London has seen 80% home working, and more so in Islington than Newham.

Can we see the same pattern between areas of Stockholm County?

Stockholm city (974k and 4.5k/sqkm) is like an Islington, has seen 65% voluntary home working, far more than other parts of Sweden, with Upplands Vasby (41k and a density of 2.6k/sqkm), being similar to a Newham. Upplands has seen 1389 d/1m, Stockholm only 1050. A similar ratio to that of Islington and Newham. However, the neighbouring suburban municipality of Danderyd (33k and a density of 1k/sqkm), just 1/3 the density of Upplands, has only seen 426 d/1m. So it looks like spread goes up with density, but is brought down by having more white collar workers and an easier capacity to support home working, and this plays out in both countries.

So what parts of the UK compare to Upplands Vasby & Stockholm City?

It is at this point that we realise Sweden is living a marketing dream. They have elevated Stockholm County to a metro area, but it is 6526 sqkms, with 2.4m people, we just don’t live like them. Our closest matching metro area is Merseyside 2.2m, costal, the core city Liverpool urban area is ~900k, so like Stockholm’s 950k, but the Mersey’s metro area is squeezed into 641 sqkms. So about 10x denser.

Regardless, if we carry on comparing:

• Upplands Vasby: is like the Wirral, the closest commuter district to Liverpool, 2k/sqkm, 480 d/1m, about 1/3.
• Stockholm City: is similar to the Liverpool urban area which has had about 450 d/1m. If we look to similar density areas in the UK to be sure, Leicester and Sheffield, both of which are at about 350d/1m, which is 1/3. You may ask why not Birmingham? It is as dense, but has twice the metro area. Manchester too is more dense, and has a 50% bigger metro area. Metro area matters as 2x the volume of people commuting in and out makes a big different. Even so their centres are at about 800 d/1m, so still not as high as Stockholm’s, and are far more like London.

So what does all this mean? How do we think about it?

• 10% of the UK is far denser, and has seen d/1m rates similar to Sweden
• 20% of the Swedish population lives at density levels we don’t even have, and they have had very low rates of spread compared to the rest of the Sweden, and this drags their average down.
• In like for like areas Sweden has seen about 3x the d/1m rate we have.

When you average all these areas together you end up with national averages that look similar, but this hides the reality. Had we relaxed the lockdown to the same degree they managed, we would have likely seen 3x the spread and deaths.

What makes London’s 5.7k vs Stockholm’s 4.8k so incomparable?

The former is for Greater London, a metro area, the latter only Stockholm City. Worse than this when you look up deaths for the two Google give you deaths for Greater London, but in Sweden it gives you deaths for Stockholm County, 26 municipalities with a density of only 360/sqkm, apples and oranges. Plus, the 4.8k is an estimate from 2016, and Stockholm’s wikipedia page cites 4.5k from a 2019 measure… sigh.

If we search in more detail we find the following (bear in mind each larger area is inclusive of the previous).

For the 10 central London boroughs that match Stockholm County’s population of 2.4m London is at least 3.4x and most 34.7x as dense, dramatically different, and incomparable. Yes, d/1m look similar, but compare Stockholm to parts of the UK of similar density, we again get about 1/3 the spread.

The reality of the UK vs Sweden is if you take each part of Sweden, and find a comparable part of the UK, you create a sub-country of the UK, that has had about 1/3 the spread of Sweden. The remainder is all denser, with a population larger than the whole of Sweden, and has had more spread, hiding this fact.

Are there other countries that provide a useful frame of reference?

If we look at other parts of the world for comparisons on demographics that bring this picture into a little more clarity, the best two to consider are Copenhagen and New York.

Why you ask?

• Copenhagen: is very similar on density to Stockholm, both at a City and County level. They have had a stronger lockdown than the UK and have only seen 210d/1m in Copenhagen, about 1/5 that of Stockholm.
• New York: is not comparable to Stockholm at all, but what makes it so interesting is it demonstrates what London could have been. The argument there is some mythical Tcell driven 20% ceiling of infection is shown to be false hope. NY has had 1800d/1m for the whole city, which means on an IFR of 0.5% it has got to 36% infection, near twice that of London.

What starts to form is a pattern between density, the level of home working achieved, and the deaths/1m that ensue.

What is the takeaway here?

The reality is there are 3 separate countries at play in comparing Sweden and the UK:

• UKtiny: 20% of Sweden is at density levels we don’t even register. Places like Skye or the Herbides are likely similar, but they are such a tiny proportion of our population they don’t even register
• UKsparse: the remaining 80% of Sweden maps onto these areas of the UK, with all like-for-like comparisons showing we have 1/3 the spread.
• UKdense: this mainly London, Birmingham and Manchester, for which there is no parallel in Sweden, and it has had 3x the spread relative to UKsparse, with the compulsory not voluntary measures we introduced only just getting these under control.

The rate of growth by mid-March as an average for both was 60% faster than Sweden, but all this growth was happening in UKdense which saw 2x. The UK’s choice to elevate Sweden’s 65% voluntary lockdown to an 80% enforced one was being made as a blunt instrument for UKdense, and really did not help UKsparse move closer to true herd immunity like Sweden managed to, all these place in the UK have about 1/3 the spread. Sure UKdense needed a window to build test & trace, measures in Germany that more permanently suppress spread, but our need to generate a united effort had consequences.

On the flip side, people should not assume this was Sweden’s plan all along, they were just as worried, and as this article shows they got close to our measures, there was no ideological red line, they would have. We simply had UKdense and they did not.

It is utopian idealism that leads to this sort of analysis from Hannan. He is not a scientist, trying to run a narrative as one, just like Toby Young, and misses all the true teaching moments in the data.