Written evidence submitted by Jenevieve Treadwell (IGR0094)

(Policy Fellow, Leading for London, London School of Economics)

This submission is in response to the question, ‘How regional Cluster growth can best be measured, mapped, and monitored to help inform local leadership and evidence-based policymaking in Whitehall.’ This is based on my ongoing research into London’s industrial clusters as the London Policy Fellow at LSE. I am more than happy to provide further details on any points raised or discuss them in person.

Anecdotally, identifying London’s clusters is simple: from Hatton Garden in Faringdon to Finance in Canary Warf. But data-driven examples of clustering are rarer. UK cities have great comparative advantages in many different industries, but there is a lack of information about where these clusters are and how they are developing over time. Without an understanding of where clusters of ‘growth industries’ exist, it is hard to quantify local and regional strengths and therefore hard to efficiently direct policy interventions.

Why is it hard to identify and monitor industrial clusters?

Firstly, the quality of the available data presents a challenge. The official survey used to identify employment by industry is the Business Register and Employment Survey (BRES). This survey has a reasonably high response rate, about 83% in 2023, although it suffered during COVID-19 (dropping to 73%), making longitudinal analysis less reliable.

The geography at which BRES is available also presents a challenge. Instead of collecting detailed location data (like postcodes), employment is attributed to administrative geographies like Lower Super Output Areas, Regions and Local Authorities. Because clusters do not obey administrative boundaries, employment may be spread across two geographies, weakening the appearance of a cluster.

The classification used to determine the industry to which a business belongs is out of date and no longer fit for the modern economy. Industry is defined in terms of Standard Industrial Classification (SIC) codes. SIC codes were created in 2007 and reflect the economy at that time.

For example, there is no SIC code for AI or Game Development. Instead, AI companies list under SIC codes like ‘Business and domestic software development’ (62012) or ‘Other information technology service activities’ (62090), and Games Developers may come under ‘Information technology consultancy activities’ (62020) or ‘Other Software Company’ (58290). FinTech companies have also struggled, with half unable to classify themselves within SIC codes.[1]

Alternative data is available through private companies. For example, Data City uses machine learning and data scraping of Companies House and firm websites to create ‘Real Time Industrial Classifications’ (RTICs) and organises firms together that self-describe in a similar way.

Consequently, RTICs are better able to reflect the modern economy. Under the RTICs classification system, AI companies not only have a category for themselves, but multiple sub-RTICs, like ‘AI Technologies and Applications: Blockchain’ or ‘Green Tech,’ that allow a significantly more detailed picture of AI’s role in the economy.

However, there are limitations to the data. The data is a count of business rather than of employment, with employment figures partially estimated or missing. As it records businesses, not employment, it is harder to account for a ‘headquartering effect.’

The headquartering effect is when a company performs activities like R&D or innovation in one place, but is registered elsewhere for administrative purposes. This means that the innovation or tradable activity may be attributed to a company’s headquarters rather than its R&D site.

For example, Rolls Royce is listed as having a London office with RTICs listed as ‘Research and Consulting - Physical Science’s and ‘Engineering: Engineering Research,’ reflecting much of the work that Rolls Royce does. However, the only activity performed at the London office is sales.

Also, there can be some misattribution of RTICs. For instance, where consultancies provide support to a range of industries, they may be given the RTICs of one of their clients. If a consultancy company’s website lists its experiences supporting Clean Energy, Game Development and MedTech firms, the model may misattribute these RTICs to that consultancy.

This is not to say that either source is of poor quality. Rather, both enable a different understanding of the UK’s economic geography.

What do we mean when we say cluster?

Clusters are concentrations of related firms or industries within a given geography.[2] This happens because firms tend to benefit from being near other firms that do similar things. The benefits that accrue to firms within the cluster can spill over to the region as a whole, resulting in economic growth.

How could we measure, map and monitor clusters?

There are many ways to measure clusters, both in terms of the data source used and the statistical techniques that can be employed.

Employment: employment in an industry as a share of total employment in an area, or number of jobs relative to the area (e.g. jobs per SqKm).

Productivity: GVA per person employed in a given industry.

Firms: firms in a given industry as a proportion of all firms in an area, or number of firms relative to the size of the area (e.g. firms per SqKm).

There is no golden dataset or metric that will indisputably identify a cluster. Instead, we can triangulate from multiple data sources. Employment count, number of firms and productivity can all be used to identify a cluster. But what they find might be different.

These variables can be simply mapped to establish rough clusters or using cluster-analysis techniques (like Moran’s I or K-means clustering). It is very important to emphasise that the type of technique used depends on the type of available data that has been collected. Similarly, data needs to be longitudinal to monitor cluster growth, change or decline.

While clusters can occur naturally, their growth is not inevitable. They can exist in a more or less supportive policy environment.[3] Without understanding where clusters are, it is difficult for either Whitehall or local leaders to target interventions and plan for the future.

24 January 2025