04Why do hotspots appear where they do?

What, then, determines where hotspots emerge? This section uses two types of regression analysis (see Box 6) to understand firstly why clustering is more likely to happen in certain parts of the country than others, and secondly why hotspots are more likely to emerge in certain neighbourhoods than others.

The analyses show that the benefits of agglomeration – that is the benefits that cities provide to businesses through access to workers and access to other knowledge-based businesses – are the main drivers of clustering of these firms. Beyond this universities and large high-technology employers are all shown to play a role.

Box 6: Regression analyses

Regression analysis refers to a group of statistical techniques which can model the influence of different factors (independent variables) have on a topic of interest (the dependent variable). One of the most important advantages of regression models is that they show the impact of a given factor, such as universities, on an outcome, such as the number of new economy firms, whilst holding other things (such as skills) constant. This enables researchers to assess how important different factors are relative to one another and control for the influence of other components.

This report uses two types of regression analysis to understand why the amount of new economy activity varies between labour market areas and why hotspots appear in specific neighbourhoods. The geographical units of interest across the two types of analysis are Travel-to-Work Areas (TTWAs) – largely self-contained labour market areas developed by the ONS – and LSOAs respectively. Unlike LSOAs, TTWAs are rarely in perfect alignment with administrative boundaries. Although this complicates recommendations for local policymakers, TTWAs are a useful unit of analysis because they provide a realistic picture of the geography over which local economies are organised. This is important when trying to understand what new economy firms find attractive. An approach based on TTWAs also aligns with the government’s Investment Zone strategy.31

The first set of regression analyses use linear models to assess how differences in independent variables, such as the number of workers with post-secondary qualifications, determine differences in the amount of new economy (or clustered new economy) activity, measured in terms of the number of NEFs per 10,000 working-age residents, across TTWAs. Data limitations mean Northern Ireland is excluded from these regressions.

An average labour market area has around 17 new economy firms per 10,000 working-age residents, and all 218 have at least one new economy firm. The average number of clustered new economy firms across the 80 labour market areas with hotspots is just 2 per 10,000 working-age residents.

The second set of analyses use logistic regression models. Logistic models are used to understand how different factors, such as proximity to a university, influence the odds that a given neighbourhood (LSOA) has a hotspot. It is necessary to use this type of model because the dependent variable, whether an LSOA has a hotspot, is a binary outcome. The topic of interest is therefore not an amount which varies in size across places; an LSOA either has a hotspot assigned to it or it does not.

Logistic models are harder to interpret than linear models, but for the purposes of this study two factors matter. First, whether a factor increases or decreases the odds that an LSOA has a hotspot. Second, which factors have the greatest influence. Data limitations mean that only England and Wales are considered in these regressions.

No model can explain everything. The results of the regression analyses provide an indication of how a small number of factors influence the relevant dependent variables rather than a complete account of clustering processes. This is especially the case in relation to the logistic regression models. Just 1.1 per cent of LSOAs in England and Wales have a hotspot. Even a 100 per cent increase in the odds a LSOA has a hotspot still means the chance of a given location having one are low.

Further details, including results tables for the models used in this report, are available in the appendix.

The size and skill level of the labour force matters for hotspots

Research has shown that agglomeration and skills are important enabling factors in profitable innovation.32 In the regression models, jobs density – a proxy for agglomeration based on the number of people working in a place relative to the resident population – has the largest impact in almost all models.33 Innovation hotspots do not form in a vacuum; rather, they emerge in dense locations alongside other businesses . Among other advantages, places with comparatively large numbers of workers are ideal locations to benefit from knowledge spillovers. The role of jobs density helps to explain the urban bias of the clustered new economy.

The linear regression models suggest that size and nature of the pool of potential workers is an important factor. Doubling the raw number of workers with post-secondary (‘L4+’) qualifications in a place would boost new economy activity by 8 per cent and increase clustered new economy activity by 30 per cent compared to the average observed across all labour market areas. The models further suggest that similar effects could be induced by increasing the share of the workforce with post-secondary skills in an area by around 8 percentage points.

The fact there are two separate effects – one relating to the proportion of a labour market with skills and another depending on the raw number of skilled workers – means that size matters. All else being equal, the models suggest that a big city will have more new economy firms per 10,000 working-age people than a small one, even if the share of the labour force with post-secondary qualifications is the same.

More evidence on the importance of the size of the skilled labour pool can be gleaned from the logistic regression models, which are used in this report to understand the factors which alter the chances a neighbourhood has a hotspot (see Box 6). To measure the number of skilled workers who can reach a neighbourhood, this report focuses on the ‘quality’ of nearby rail and metro stations. Because most people travel to work by car, rail and metro stations are an imperfect proxy of accessibility.34 However, they are important hubs for different kinds of public transport and hotspots are often located nearby to them.

Box 7: Using stations to estimate neighbourhood-level access to skilled workers

On average, a hotspot is between 0.7 and 1.3 km from a station. In this analysis, stations within 0.8 km are considered to be ‘nearby’. The size of a neighbourhood’s skilled labour pool is measured as the number of people with post-secondary qualifications who can reach the best nearby station by public transport in 30 minutes.35 Of the nearly 35,000 neighbourhoods in England and Wales, 9,000 have a rail or metro station nearby. The average station in England and Wales can be reached by 22,000 people, although this figure varies across the country. The best-connected neighbourhood, located in central London, has access to 340,000 highly-skilled workers via its strongest nearby station; the least well-connected neighbourhood within 0.8 km of a station can be reached by just 2,300.

The regression models suggest that increasing the number of skilled workers who can reach a neighbourhood by around 40,000 increases the odds that the neighbourhood has a hotspot by 27 per cent. As can be seen in Figure 11, the impact of accessibility scales strongly. All else being equal, the odds that a neighbourhood in walking distance of Manchester Piccadilly, which is accessible to around 130,000 highly-skilled workers, having a hotspot are around two per cent. For a place close to London Waterloo, which more than 300,000 highly-skilled workers can reach in 30 minutes, the odds are 4.5 per cent. Considering that just one per cent of neighbourhoods have hotspots, the impact of access to larger pools of labour is considerable.

Figure 11: Predicted probability that a neighbourhood has a hotspot depending on accessibility

Source: ONS, The Data City, and Centre for Cities calculations

Many large cities are not offering the access to skilled workers that their size would suggest

At first glance, these findings appear to contradict the underperformance of large cities observed in Section 3. However, although these cities are large on paper, prior Centre for Cities research has shown that poor transport connections in British cities makes them effectively smaller than their populations suggest, whilst European cities have access to much deeper pools of labour thanks to their strong connectivity.36

It is therefore likely that part of the observed weakness of major cities’ new economies can be attributed to ‘effective size’ problems. In Figure 12, the size of the major cities’ city centre new economies, relative to the urban average, is given in relation to their nominal and effective sizes. The latter metric, displayed in the dark green bar, gives a sense of the size of the new economy of each city centre relative to the number of highly-skilled people who can actually get there, rather than the residents of the city as a whole.

Figure 12: City centre new economies measured against population and effective size

Source: ONS, The Data City, and Centre for Cities calculations

Three important observations emerge from the data. First, major cities have larger city centre new economies than city centres in general. Sheffield and Birmingham are exceptions to this rule, with Birmingham’s city centre possessing an average-sized new economy and Sheffield’s being 10 per cent smaller.

Second, most major cities look stronger when the number of highly-skilled people who can reach the city centre, rather than the total number of highly-skilled people in the whole city, are considered. The difference is particularly striking in relation to Manchester and Birmingham, which move from middle-to-low performers to positions much closer to that of Bristol and Leeds. Among major cities, Manchester’s city centre new economy is second only to that of London when effective size is considered.

Third, the performance of city centres in small successful cities such as Brighton and Reading is much more explicable after taking effective size into consideration. These cities are well connected hubs and, as such, are effectively larger than their raw populations suggest. Using their populations, as is the case in Figure 10, rather than effective sizes, therefore overstates the success of their city centre new economies.

Together, the evidence from Figures 10 and 12 show that when considered in terms of effective size, major cities such as Manchester, Birmingham, and Liverpool have successful new economies which cluster strongly into hotspots. The unfortunate fact is that despite ostensibly possessing large, highly-skilled, labour forces, these cities are effectively much smaller. The best-connected station in Brighton can move more highly-skilled workers into the city centre in 30 minutes than the best-connected stations in Bristol, Leeds, Liverpool, and Newcastle and only around 8 per cent less than those in Birmingham, Nottingham, and Sheffield. While this is great for Brighton, it speaks to a fundamental problem in major cities elsewhere in the country, which draw fewer benefits from their skilled workforces than would be the case with better transport connections.

Universities play a role in organising new economy businesses into hotspots, but their impact is conditional on research quality and distance

Universities have been shown to support innovation through research outputs, incubators, the provision of infrastructure, and the deployment of reputational and intellectual capital.37 The quality of the institutions, measured in relation to their research outputs, is an important factor in determining their effectiveness.38 The Government appears to be applying research in this area to its Investment Zone strategy, which includes a role for research-intensive institutions and innovation catapults.39

The impact of universities on clustering at the spatial scales analysed in this report is, however, mixed. While universities do not appear to increase the amount of new economy activity in their vicinities, they are associated with the organising of the innovative firms into hotspots. Successful research-intensive universities perform this function far more effectively than their peers, although their impact is still relatively modest.

Box 8: Identifying and classifying universities

Data on university locations come from the Higher Education Statistics Agency.40 Following the methodology used by DLUHC in identifying locations suitable for Investment Zones, this report relies on the classification system contained in the annual Transparent Approach to Costing (TRAC) reports in order to gauge institutional quality.41 The system divides universities into six ‘peer groups’ depending on their research incomes and possession of certain departments, such as medical schools.42 For the purposes of this report, only universities in the top TRAC-A group are considered to be ‘research-intensive’.

The primary evidence for the role of universities being confined to the spatial organisation of the new economy comes from the linear regression models. At a labour market level, the impact of universities on new economy activity is statistically insignificant at best. The models even suggest that research-intensive universities actually reduce the per capita number of new economy firms to an extent similar to halving the number of skilled workers in a place. Despite this, universities do increase the amount of clustered new economy activity in their labour market areas, with research-intensive institutions providing a boost nearly eight times larger than other universities. Therefore, although universities do not seem to increase the stock of new economy firms in a place, their boost to the number of clustered new economy firms suggest that they help organise the local new economy into hotspots.

Their impact is, however, mediated by distance. One logistic model shows that neighbourhoods within 5km of a research-intensive university are 48 per cent more likely to have a hotspot than those further away. Another, which considers linear distance, suggests that a 1 km increase in the distance between a neighbourhood and the research-intensive university in its labour market area reduces the odds it has a hotspot by 3.2 per cent. A neighbourhood 6.5 km further away from the university than another is 20 per cent less likely to have a hotspot.

While these effects are sizeable, the rarity of hotspots means that these effects are relatively modest in practice. Holding everything else constant, the linear distance models suggest that the probability that a neighbourhood within 1km of a research-intensive university has a hotspot is 1.5 per cent. The equivalent figure for those 10km away is 1.2 per cent.

One explanation for universities’ relatively small impact is that their role is most prominent at the beginning of a company’s development, and as a result the size and number of associated innovative firms (some of which may be formal spin-off ventures) are small.43 It is therefore possible that the effects observed in this report are a reflection of businesses supported by universities moving away as they grow, limiting the overall number of new economy firms nearby and leaving behind only those businesses still in early stages of development. Data limitations prevent deeper analysis of this question. It may also be the case that these clustered firms perform better because they are close to a research-intensive university, and that the impact of these institutions is felt indirectly over larger distances. But once again further data would be required to explore this.

Box 9: Finding places with large, high-technology employers

Two challenges exist in relation to the identification of important employers. First, data on specific firms are tightly controlled by the ONS in order to prevent disclosure. Second, publishable data are only available at the level of middle layer super output areas (MSOAs), which are reasonably large geographical units containing four to five LSOAs. These factors necessitated careful choice in metrics and a strategy based on estimation.

The approach taken by this report is as follows. First, public MSOA-level counts of firms and employees across industries (using two- and three-digit levels of aggregation of the SIC) are compared to determine instances where suppression has been used to prevent disclosure. The total number of suppressed firms in each industry are then divided evenly among MSOAs where suppression has probably occurred (places where employees in an industry are recorded despite the reported number of firms being zero).

Second, the number of employees in a given industry in a specific MSOA are divided by the estimated number of firms. This gives the average size of firms in that industry in the MSOA. Third, an MSOA is deemed to possess a relatively large employer in an industry if the average size of the firms is above that seen both in the MSOA in general and the industry nationally. This is by no means a perfect metric, but it gives an indication of places in which a particular industry’s firms are relatively large.

Some industries are more important than others. For the purposes of this analysis, the 747 MSOAs (10 per cent of all MSOAs) with relatively large firms in nine high-technology industries (SIC codes 21, 26, 30.3, 59-63, and 72) are considered.44 Among the industries are pharmaceuticals, computer hardware, spacecraft, broadcasting, telecommunications, software, and scientific research. Because MSOAs are large, the statistic used in the logistic regression analyses is the distance (in km) between an LSOA and the closest MSOA with relatively large high-technology firms.

At a local level, high-technology employers are stronger anchors than universities

Large, high-technology employers have been shown to play important roles in translating local academic research into product development.45 Evidence from the biotechnology sector also suggests that large innovative companies can act as anchors around which start ups and small enterprises can benefit.46

The logistic models show that high-technology employers have a positive impact on the odds that a neighbourhood has a hotspot. As Figure 13 shows, holding all else equal, the odds that a neighbourhood within 1km of a major high-technology employer has a hotspot are just over two per cent, while the chance that those 10km away is just over one per cent.

Penalties for being far away from these major employers are therefore large. A 1km increase in a neighbourhood’s separation from a high-technology employer implies a 15 per cent reduction in the probability that it has a hotspot. Although the strength of the penalty wanes with distance, the impact is still significant. A neighbourhood 6.5 km further away from a major employer than another is 67 per cent less likely to have a hotspot.

Figure 13: Predicted probability that a neighbourhood has a hotspot by distances from a large high-technology employer

Source: ONS, The Data City, and Centre for Cities calculations

Large, high-technology employers can therefore act as anchors for clustered new economy activity, and they seem to perform this function more strongly than research-intensive universities. Like universities, however, the anchoring effect is strongest in the immediate vicinity. As is the case in Stevenage (Box 4), places which host major institutions are ideal locations for innovative companies looking to share inputs and benefit from relationships in the supply chain or in the wider labour market.

Firms are willing to pay a premium to enjoy the benefits of hotspots

Evidence from rateable value data suggests that innovative firms are willing to pay a premium to form hotspots in desirable locations, even in the parts of the country where floorspace costs are already high.47 As can be seen in Figure 14, the per metre rateable value of office space in areas with hotspots is generally higher than that in areas without across PUAs and non-urban local authority districts.

The average premium on office space is between 11 and 13 per cent, although this varies across the country. In London, which has the most expensive office space in the country, the difference in cost between places with and without hotspots is around £150 per square metre. The premium on hotspot office space is also large in Reading, Oxford, Birmingham, and Manchester. Despite being somewhat elevated in places with hotspots, floorspace costs in Liverpool, Newcastle, and Sheffield are much lower.

Figure 14: Rateable value of office space and hotspots and elsewhere across local authorities

Source: Valuation Office Agency, The Data City, and Centre for Cities calculations

The rateable value of floorspace and the various premiums give an indication as to where the development of hotspots may be encountering constraints imposed by the built environment. London, and other cities and towns in the Greater South East such as Reading, Oxford, and St Albans are obvious examples. There, office space is generally expensive, and space in hotspots even more so.

The premium on the cost of space in South Cambridgeshire is the highest of any labour market area. Where these hotspots are, though, is revealing, with most of the area’s clustered new economy firms being found in close proximity of Cambridge itself. This, coupled with the high cost of commercial space in Cambridge points to a chronic undersupply of commercial premises. This undersupply, especially laboratory space, in Cambridge and Oxford has received significant attention in the last few months.48 Among the prevailing concerns is that a lack of facilities is pushing valuable activity overseas and hurting national growth prospects.49 In response to these challenges, the Government has announced the formation of a Cambridge Delivery Group to oversee the expansion of the city over the long term.50


  • 31 Department for Levelling Up, Housing & Communities (2023), Investment Zones Place Selection: Methodology Note, His Majesty’s Stationery Office
  • 32 Leiponen A (2005), Skills and innovation, International Journal of Industrial Organization, 23 (5-6): 303-323
  • 33 In the linear models, only the control for Greater South East location is larger (see Appendix)
  • 34 Le Vine S, Polak J and Humphrey A (2016), Commuting trends in England 1988-2015, London: Department for Transport
  • 35 The figures are computed from output areas containing stations and census data. Analysis confined to England and Wales for data availability reasons. Sources: ONS (2022), UK Travel Area Isochrones; ONS (2023), Census of Population 2021
  • 36 Rodrigues G and Breach A, Measuring up: Comparing public transport in the UK and Europe’s biggest cities, London: Centre for Cities
  • 37 Jaffe A (1989), Real effects of academic research, The American Economic Review, 79 (5): 957-970; Porter M (1990), The Competitive Advantage of Nations, New York: Free Press; Heaton S, Siegel D and Treece J (2019), Universities and innovation ecosystems: a dynamic capabilities perspective, Industrial and Corporate Change, 28 (4): 921-939
  • 38 Fritsch M and Slavtchev V (2007), Universities and Innovation in Space, Industry and Innovation, 14 (2): 201-218
  • 39 Department for Levelling Up, Housing & Communities (2023), Investment Zones Place Selection: Methodology Note, His Majesty’s Stationery Office.
  • 40 Higher Education Statistics Agency (2021), Unistats dataset
  • 41 Department for Levelling Up, Housing & Communities (2023); Investment Zones Place Selection: Methodology Note, His Majesty’s Stationery Office
  • 42 Office for Students, Peer groups for annual TRAC, TRAC fEC and TRAC(T) benchmarking 2021-22, London: TRAC
  • 43 Bagchi-Sen S, Baines N and Lawton Smith H (2022), Characteristics and Outputs of University Spin-offs in the United Kingdom, International Regional Science Review, 45 (6): 606-635
  • 44 The choice of sectors is akin to that used in other clustering research, e.g. Advanced Oxford, Oxfordshire’s Innovation Engine 2023, Oxford: Advanced Oxford
  • 45 Agrawal A and Cockburn I (2003), The anchor tenant hypothesis: exploring the role of large, local, R&D-intensive firms in regional innovation systems, International Journal of Industrial Organization, 21 (9): 1227-1253
  • 46 Feldman M (2005), The Locational Dynamics of the U.S. Biotech Industry: Knowledge Externalities and the Anchor Hypothesis, in Quadrio Curzio A and Fortis M (eds.) Research and Technological Innovation, Heidelberg: Physica-Verlag
  • 47 These costs are excluded from the regression models because of high collinearity with jobs density, which is a far more important explanatory variable
  • 48 Howard T, Science superpower dreams thwarted by lack of laboratory space, The Times 14 February 2023
  • 49 Foster P, Lab space shortage threatens life science boom in Oxford and Cambridge, Financial Times 1 August 2022
  • 50 Gove M (2023), Long-term plan for housing: Secretary of State’s speech, https://web.archive.org/web/20230810164405/https://www.gov.uk/government/speeches/long-term-plan-for-housing-secretary-of-states-speech