We have witnessed significant changes in economic geography over the last years. Locations that had very similar income levels or house prices in the 1960s are now quite different. Moretti (2012) terms this the ‘Great Divergence’ and sees the rise of innovation hubs playing a central role. While much attention has been paid to changes in the geography of innovation over recent decades, longer term questions remain unanswered. Are today's levels of spatial concentration of innovation historically unprecedented? Which locations drive changes in the spatial concentration of innovation over time? How persistent is regional innovation leadership? Has persistence increased or decreased over time? We aim to answer these questions below. The longstanding central obstacle in
Andrews, Whalley considers the following as important:
This could be interesting, too:
Mendicino, Nikolov, Rubio-Ramirez, Suarez, Supera writes How much capital should banks hold?
Djankov, Panizza writes How Africa can recover from Covid-19
Alogoskoufis writes The pandemic and Greece’s debt: The day after
We have witnessed significant changes in economic geography over the last years. Locations that had very similar income levels or house prices in the 1960s are now quite different. Moretti (2012) terms this the ‘Great Divergence’ and sees the rise of innovation hubs playing a central role. While much attention has been paid to changes in the geography of innovation over recent decades, longer term questions remain unanswered. Are today's levels of spatial concentration of innovation historically unprecedented? Which locations drive changes in the spatial concentration of innovation over time? How persistent is regional innovation leadership? Has persistence increased or decreased over time? We aim to answer these questions below.
The longstanding central obstacle in answering these questions – the lack of precisely geo-coded patent data over the very long term – has recently been overcome. In a new paper (Andrews and Whalley 2021), we employ patent data from the Comprehensive Universe of US Patents (CUSP) (Berkes 2018), which contains information on all US patents from 1836 to 2016. Crucially for our purposes, the CUSP data contain the location of each inventor listed on each patent. The geographical information is obtained by extracting the name of each inventor's town, county, and state from the patent text. We then determine the latitude and longitude of that location, and assign that location to its current US county. This last step is important in light of changes in municipal and county boundaries over time.
We measure the spatial concentration of innovation based on the ‘dartboard’ approach of Ellison and Glaeser (1997). The idea is to compare the observed spatial distribution of innovation to what would happen if innovation was randomly distributed across space according to the population distribution in order to capture the spatial concentration of innovation intensity.
In Figure 1 we plot our dartboard spatial concentration of innovation index over time using commuting zones as the local geographic area. We choose to begin our analysis in 1866 to exclude disruptions from the Civil War. Regardless of which geographic measure we use, the concentration of innovation exhibits nearly identical patterns over time. Four distinct periods are apparent.
Figure 1 Patent concentration
The period from the end of the Civil War to roughly 1905 is one of largely uninterrupted declining concentration What might explain this decline in spatial concentration? One constant since the middle of the 19th century has been improvements in transportation and communication technologies. A second trend is the widespread increases in higher education. Goldin and Katz (1999) refer to this period as the ‘formative years’ of US higher education. The 1862 Morrill Act, for instance, provided land grants to states to fund agricultural and mechanical colleges in their states. This spurred the creation of universities with a focus on practical science across the country. In most cases, these new universities were not located at existing large innovation hubs, but were instead placed near the geographic centre of the states to be more easily accessible to a largely rural population. Andrews (2020) shows that the establishment of a local college caused an increase in local patenting, although this increase is mostly due to the colleges' role in promoting population growth rather than the direct effects of increasing human capital. Kantor and Whalley (2019) show that proximity to land grant colleges also facilitated the diffusion of innovations, with areas closer to colleges increasing in agricultural productivity faster than more distant areas. Notably, the benefits of proximity to a land grant college decline during the 20th century, coincident with the widespread adoption of the telephone and automobile, further supporting the argument that declining transportation and communication costs led to less spatially concentrated innovation.
Around 1905, the decline in the spatial concentration of invention occurring since the end of the Civil War halted and, if anything, slightly reversed. By the early 20th century, the US railroad network was mature and experienced little additional growth. This period also saw inventive activity increasingly move to large industrial firms, which were more geographically concentrated than independent, democratic inventors.
At the end of WWII in 1945, the decline in the spatial concentration of invention recommenced. WWII was a watershed moment in terms of the government's involvement in research and innovative activity. Locations that received large amounts of wartime funding, some of which were not among the most innovative locations before the war, continued to be major sites for industrial production and innovation after the war ended.
Starting around 1990, the postwar decline in concentration abruptly halted and reversed, leading to 25 years of rapidly increasing spatial concentration of invention. These changes could be due to the fact that internet technology is a complement rather than a substitute for in-person interaction. They may also be driven by the growing importance of highly concentrated education for regional growth and education-innovation complementarities (see Akcigit et al. 2020), among other potential factors.
An alternative way to visualize the spatial concentration of invention is to plot the share of patents from commuting zones in different parts of the patenting distribution through time (Figure 2). In panel (a), we plot the share of patents from the top 5% of commuting zones when ranked by patenting activity in each year, along with the bottom 50%. The share of patents in the bottom 50% of the distribution move far less and, in all years, account for only a trivial share of patents. In panel (b) we further break down patenting in the top commuting zones. The trends for the top 1% share after 1945 closely mirror national trends in spatial concentration in Figure 1. Overall, trends in national innovation spatial concentration appear to be driven by a few elite innovation hubs.
Figure 2 Share of patents, by percentile
Have today's elite innovation hubs always been highly innovative? How much churn is there in the identities of the elite innovation hubs? Many policymakers expect the location of innovation leadership to be highly persistent. If there are increasing returns and cumulative innovation effects at the regional level, then early leads in innovation would be likely to compound, producing persistence.
To investigate persistence, in Figure 3 we examine the correlation between the share of national patents held by a commuting zone at two different points in time. We split our time period in half and plot the correlations between the share of total national patents from each commuting zone in 1866 and 1941 in panel (a) and in 1941 and 2016 in panel (b). We see some evidence of persistence, though for both sets of years the regression line is below the 45-degree line and the data cloud is quite scattered. Further, persistence appears to be falling — the correlations for the last 75 years are significantly lower than over the previous 75 years. The top patenting commuting zones in 1866 tended to be the country's largest cities, with the northeast dominant. Of the six commuting zones making up the top 1% in 1866, four were still in the top five in 1940, and all were in the top seven. In 2016, the West Coast, and especially Silicon Valley, is nearly as dominant among the top 1% of commuting zones as the northeast was in 1866.
Figure 3 Share-share correlations
Hence, the geographic concentration of inventive activity observed today is not unprecedented and the US had a similar level of geographic concentration at the start of the Industrial Revolution as it has today.
However, the changing degree of geographic concentration over time, and especially the substantial turnover in the identities of top inventing places, suggests that the location of invention is not fixed by geography or historical accident, and that policy may have a role to play in promoting local or regional innovation.
Akcigit, U, J Pearce and M Prato (2020), “Tapping into talent: Coupling education and innovation policies for economic growth”, VoxEU.org, 10 October.
Andrews, M (2020) “How do Institutions of Higher Education affect Local Invention? Evidence from the Establishment of U.S. Colleges”, working paper, University of Maryland.
Andrews, M and A Whalley (2021), “150 Years of the Geography of Innovation”, Regional Science and Urban Economics, forthcoming.
Berkes, E (2018), “Comprehensive Universe of U.S. Patents (CUSP): Data and Facts”, working paper.
Ellison, G and E L Glaeser (1997), “Geographic Concentration in U.S. Manufacturing Industries: a Dartboard Approach”, Journal of Political Economy 105(5): 889-927.
Goldon, C and L F Katz (1999), “The Shaping of Higher Education: the Formative Years in the United States, 1890 to 1940”, Journal of Economic Perspectives 13(1): 37-62.
Kantor, S and A Whalley (2019), “Research Proximity and Productivity: Long-term Evidence from Agriculture”, Journal of Political Economy 127(2): 819-854.
Moretti, E (2012), The New Geography of Jobs, Houghton Mifflin Harcourt Publishing Company.