As part of its manifesto, the government committed to raising the UK’s R&D spend to 2.4 per cent of GDP by 2025, up from just 1.7 per cent in 2018 to boost productivity and create well-paid jobs.1The latest year the data is available. But framing additional R&D as part of its levelling up agenda to boost economically lagging areas and retain ‘Red Wall’ constituencies risks politicising an otherwise sensible economic policy and wasting public funding when the public finances are already under strain.
This blog presents new analysis offering a granular view of regional R&D capabilities. It shows why we need first to strengthen areas’ R&D capacity – its local innovation ecosystem – to put new money to good use before the funding follows.
Different Regions, Different Strengths
Historically, UK government spending on R&D has been disproportionately allocated to the ‘golden-triangle’ (the South East, London and the East of England). In theory, this concentration shouldn’t matter, as spillover effects raise the prosperity of the whole nation. In practice, however, spillover effects decay sharply with distance widening regional inequality.
In addition to disparities in the levels of R&D spending, different regions have different specialisms and different local innovation ecosystems supporting those specialisms. By using data on business expenditure on R&D and methods from economic complexity research, we can identify specific regional R&D advantages and the similarities between them. (A full description of the methodology is at the bottom of this post).
The first thing to note from Figure 1 is that areas with high R&D intensity are often very different in terms of the sector focus of that investment. The black squares on the heatmap indicate where a region has R&D strengths in a particular sector (i.e. R&D intensity greater than the national average), with both regions and sectors automatically ordered based on their similarities. For example, regions with similar R&D capabilities, such as Inner London and Surrey, will be grouped closer together, but further away from regions with very different R&D specialisms, like the West Midlands. Likewise, sectors are automatically ordered so that those that are similar, such as automotive and aerospace, appear closer together.
Figure 1 – Mapping the strengths and similarities of regional R&D capabilities
Black squares indicate sectoral strengths for each region, while bars indicate total business expenditure on R&D (BERD) for each region/sector. Sectors starting with ‘M’ are manufacturing sectors.
From Tech-Specialists to Jack of All Trades, classifying the regions
High-spending regions are often very different in their R&D capabilities. The bars in Figure 1 indicate the total amount of business expenditure on R&D (BERD) for each region and sector. Here we see that regions with high R&D intensity achieve this in different ways. Berkshire, Buckinghamshire and Oxfordshire are the largest spenders, but focus quite narrowly on ICT sectors and science. East Anglia is the second largest spender, but investments in R&D cover a much more diverse range of industries. Building on this insight allows us to categorise and analyse clusters of regions in the heat map above into four types: Tech specialists, Polymaths, Jacks of all Trades and Manufacturers (see Table 1).
Table 1 – The four categories of region according to their R&D strengths/weaknesses
Category name |
Description |
Regions |
Tech specialists |
Specialised strengths in tech sectors like telecoms, digital creative services and information services. Often strong in science, and high-overall BERD. |
All South East regions except Kent London (Inner and Outer) East Scotland |
Polymaths |
Diverse strengths, including both tech and manufacturing sectors. BERD is moderate to high. |
All East of England regions West Yorkshire Greater Manchester |
Jacks of all Trades (masters of none) |
Diverse strengths like polymaths, but little tech specialisation. BERD is low. This is the largest category. |
Kent South Yorkshire Merseyside South Western Scotland Highlands and Islands East Wales Lincolnshire Cornwall and Isles of Scilly Leicestershire, Rutland and Northamptonshire Devon Dorset and Somerset East Yorkshire and Northern Lincolnshire Shropshire and Staffordshire North Eastern Scotland Tees Valley and Durham |
Manufacturers |
Strengths in manufacturing areas, some strength in science, with little tech. BERD varies, with more heavily specialised regions tending to spend more. |
Gloucestershire, Wiltshire and Bristol/Bath area West Wales and The Valleys Northumberland and Tyne and Wear Northern Ireland North Yorkshire Cheshire Derbyshire and Nottinghamshire Lancashire Cumbria Herefordshire, Worcestershire and Warwickshire West Midlands |
And there is a clear relationship between the categories and the level of R&D going on. Figure 2 shows more clearly how total BERD spending varies by category. The Jacks of all Trades clearly lag all other categories, although the largest Jack spenders are comparable to the weakest Polymaths and Manufacturers. The largest spenders are all either Tech Specialists or Polymaths.
Figure 2 – Tech Specialists, Polymaths tend to spend more on R&D, total BERD spending by category, 2017
The link between R&D category and economic prosperity is very strong. Figure 3, shows total BERD for each region against their median gross domestic household income (GDHI), a measure of living standards in each region. We can see clearly that all the Jacks are clustered in the bottom left and are joined by the weaker Manufacturers and Polymaths: this means that regions with lower levels of business R&D spending tend to be more deprived areas with lower living standards.
Not only that but the relationship has got stronger in recent years. Figure 4 shows that regions with greater BERD have experienced substantially better growth in GDHI between 2012 and 2017.2Logarithmic growth is calculated as the difference in the natural logarithm of the GDHI in 2017 compared to 2012. At changes close to 0, this is very similar to the percentage change. Logarithmic changes are preferable because they are symmetric for increases and decreases. The Jacks had some of the weakest growth in living standards of any regions, a serious issue given these regions already had some of the lowest GDHI. No Jacks are present in the top-10 fastest growing NUTS 2 regions, and they make up seven of the bottom 10. Regions in other categories with reasonably high BERD (greater than ~£600m) have all experienced growth at or above the median.
Figure 3 - 2017 BERD against 2017 median GDHI, coloured by innovation category
Figure 4 - 2017 BERD against 5-year median GDHI log growth rate
But while Jacks and some Manufacturers are some of the poorest regions in the UK, simply raising the level of public sector R&D funding is unlikely to be enough begin to redress long-standing regional disparities.
Every region’s ability to absorb and use additional R&D money to create economic prosperity depends on its R&D capacity: the local innovation ecosystem of its universities and research institutions, its scientists and engineers, and its innovative businesses and start-ups.
Rather than simply pouring public money into larger or more diverse R&D programmes, the government needs to also prioritise funding to build-up the wider capacity in these relatively deprived areas to put that R&D activity to good use. Without this, the government is likely to find the seeds of additional R&D funding fail to bear fruit for its levelling-up agenda.
TECHNICAL APPENDIX
Data
Business expenditure on R&D (BERD) data comes from Eurostat, and can be broken down by NUTS2 region, and by sector according to 2- or 3-digit NACE industry codes (dependent on sector). Eurostat does not, however, provide a crosstab necessary for this analysis.
To resolve this, we estimate the crosstab by using GVA data from the ONS3 https://www.ons.gov.uk/economy/grossvalueaddedgva/datasets/regionalgvanuts2 (which does provide a crosstab), and by applying a technique called the Furness method.
Let $$ X$$ be a matrix where each element $$ {X}_{ij}$$ represents the total business R&D expenditure in region $$ i$$ in sector $$ j$$. The Eurostat data tells us the row sums $$ {\sum}_{j}{X}_{ij}$$ and the column sums $$ {\sum}_{i}{X}_{ij}$$, but not the individual elements. Let $$ {X}^{\left(0\right)}$$ be a matrix with the same dimensions of $$ X$$, but where the individual elements are the GVA for each region and sector.
The Furness method enables estimation of the elements of $$ X$$ iteratively. Let the elements of matrix $$ {X}^{\left(1\right)}$$ and $$ {X}^{\left(2\right)}$$ be defined as follows:
$$ {X}_{ij}^{\left(1\right)}=\frac{{X}_{ij}^{\left(0\right)}{\sum}_{j}{X}_{ij}}{{\sum}_{j}{X}_{ij}^{\left(0\right)}}$$
$$ {X}_{ij}^{\left(2\right)}=\frac{{X}_{ij}^{\left(1\right)}{\sum}_{i}{X}_{ij}}{{\sum}_{i}{X}_{ij}^{\left(1\right)}}$$
In simple terms, in the first iteration, we divide all the entries in $$ {X}^{\left(0\right)}$$ by the row sums of $$ {X}^{\left(0\right)}$$, and multiply them by the row sums of $$ X$$, which means the row sums of $$ {X}^{\left(1\right)}$$ and $$ X$$ are the same. We then repeat this process with the column sums to create $$ {X}^{\left(2\right)}$$. This process converges rapidly, such that after 12 iterations, the row/column sums of matrix $$ {X}^{\left(12\right)}$$ are within 0.1 percent of the row/column sums of $$ X$$.
Revealed comparative advantage
The sectoral strengths of each region are identified by calculating revealed comparative advantage (RCA) from the previously defined matrix $$ X$$. RCA for region $$ i$$ in sector $$ j$$ is a measure of how much region $$ i$$ invests in sector $$ j$$ compared to the average region. It is defined as:
$$ {RCA}_{ij}=\frac{{X}_{ij}/{\sum}_{j}{X}_{ij}}{{\sum}_{i}{X}_{ij}/{\sum}_{ij}{X}_{ij}}$$
RCA values greater than 1 indicate that the region has more business R&D investment in that sector than the average region. We can then construct a matrix $$ M$$ showing the strengths of each region, where:
$$ {M}_{ij}=\mathrm{}\left\{\begin{array}{c}1,{RCA}_{ij}\ge 1\\ 0,otherwise\end{array}\right.$$
Which is represented as a heatmap in figure 1.
Similarity indices
The regional similarity matrix $$ \stackrel{~}{M}$$ is defined as:
$$ {\stackrel{~}{M}}_{ii\mathrm{\text{'}}}={\sum}_{j}\frac{{M}_{ij}{M}_{i\text{'}j}}{{u}_{j}{d}_{i}}$$
Where $$ {d}_{i}={\sum}_{j}{M}_{ij}$$ and $$ {u}_{i}={\sum}_{i}{M}_{ij}$$. These measures are known as diversity (the number of sectoral strengths a particular region has) and ubiquity (the number of regions that have a strength in a particular sector). Likewise, a sectoral similarity matrix $$ \widehat{M}$$ can be defined:
$$ {\widehat{M}}_{jj\text{'}}={\sum}_{i}\frac{{M}_{ij}{M}_{ij\text{'}}}{{u}_{j}{d}_{i}}$$
The similarity indices used to order the regions and sectors in figure 1 are defined as the eigenvector associated with the second largest eigenvalue of $$ \stackrel{~}{M}$$ and $$ \widehat{M}$$ respectively.
For further information on interpreting these similarity matrices, and why the second eigenvector produces a similarity index, see Mealy et al., 2019.4 https://advances.sciencemag.org/content/advances/5/1/eaau1705.full.pdf
Footnotes
- ^ The latest year the data is available.
- ^ Logarithmic growth is calculated as the difference in the natural logarithm of the GDHI in 2017 compared to 2012. At changes close to 0, this is very similar to the percentage change. Logarithmic changes are preferable because they are symmetric for increases and decreases.
- ^ https://www.ons.gov.uk/economy/grossvalueaddedgva/datasets/regionalgvanuts2
- ^ https://advances.sciencemag.org/content/advances/5/1/eaau1705.full.pdf