December 7th: granularity is key for levelling up
I have the privilege to be speaking at the Levelling Up With Open Data event on Wednesday, and one of the slides I put together for this actually surprised even me. I had hypothesised (as it is fairly well-rehearsed) that the within-region differences in various measures of prosperity are significantly wider than the between-region differences. [Digging into the granular detail underpinning aggregates is a bit of a theme for me.]
Because most of the data published by the ONS is quite out of date by the time it comes out, I decided to look at claimant count unemployment, which is one of the more timely metrics. The nice thing about claimant count is that it is also available at a very detailed geographic level, for so called “lower level super output areas” or LSOAs. There are around 35,000 LSOAs in England and Wales, with an average 16+ population in 2020 of around 1,400 each. This means that one can also look at the data at the level of “wards” — slightly bigger geographic areas, with an average 16+ population of 6,500.
So, let’s look at Leeds as an example. It’s situated in the English region of Yorkshire and The Humber, where claimant count in October 2021 was 3.9%, so very close to the England and Wales average of 3.7% (left hand panel in the chart). Double clicking on this region and looking at local authority level data (middle panel), we can see that Leeds had a slightly elevated claimant count, at 4.4%, relative to the rest of the region. [This is not unusual: cities tend to have higher unemployment than other areas.]
We can also see that claimant count varies a lot in Yorkshire and The Humber: Bradford’s rate is 6.5%, while Craven’s only 1.3%.
But that’s nothing compared to the variation inside Leeds. The right hand panel further splits down Leeds into wards (the rows) and lower level super output areas (the dots). At the ward level, the differences are already significant: the average claimant count in Gipton & Harehills is 12.5%, while it’s only 1.1% in Harewood. Digging even deeper into the lower level super output areas within Gipton & Harehills, the highest claimant count in October was 17.8% and the lowest 5.2%.
The moral of the story? Well, I think there are several, but three main ones. First, while useful in general conversation, aggregates hide a lot of variation and detail and it is often both important and interesting to look beyond the headline figures. Second, when it comes to “levelling up”, it may be that the unit of analysis — and action — needs to be massively smaller* than what we have gotten used to talking about (e.g., see my essay starting on page 15). Third, data is beautiful!
[* In fact, the appropriate geographic level to consider depends on many other factors, too, and will be different depending on the kind of issue we are trying to fix and the kind of public policy levers that are available. For example, nursery provision is likely to work best at a very local level, whereas transport networks are often something to optimise at a city or national level.]