![Sample our Urban Studies journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/110826/TOC-Subject-Sample-2021-Urban-Studies.jpg"})
Abstract
Geographers and social scientists have probed the effects of agglomeration and spatial clustering on innovation and economic growth. Economists and others have identified the role of knowledge spillovers in driving the innovation process. Although innovation is thus assumed to be a function of proximity, there has been little systematic research on the role of density in innovation. This research investigates density, and more specifically the density of creative workers, as a key factor influencing regional innovation. It uses principal components analysis to create and implement a composite measure of density and presents a model of innovation as a function of creative density. Statistical analyses including multivariate regression find that density and creativity separately and jointly affect innovation in metropolitan areas. The regression analysis finds a positive relationship between the density of creative workers and metropolitan patenting activity, suggesting that density is a key component of knowledge spillovers and a key component of innovation.
Los geógrafos y los científicos sociales han investigado los efectos de la aglomeración y del agrupamiento espacial en la innovación y el crecimiento económico. Los economistas y otras personas han identificado el papel del desbordamiento del conocimiento en el impulso del proceso de innovación. Aunque, por consecuencia, se supone que la innovación es función de la proximidad, se han realizado pocas investigaciones sistemáticas sobre el papel de la densidad en la innovación. En esta investigación se estudia la densidad, y más específicamente la densidad de trabajadores creativos, como un factor clave que influencia la innovación regional. Se utiliza el análisis de los componentes principales para crear e implementar una medida compuesta de la densidad, y se presenta un modelo de innovación como función de la densidad creativa. Los análisis estadísticos, que incluyen regresión múltiple, indican que la densidad y la creatividad juntas y por separado afectan la innovación en las áreas metropolitanas. El análisis de regresión muestra una relación positiva entre la densidad de trabajadores creativos y la presentación de patentes en el área metropolitana, lo cual sugiere que la densidad es un componente clave del desbordamiento de conocimiento y de la innovación.
Acknowledgments
We would like to thank two anonymous reviewers for their comments, and the Annals editor for her careful attention. Elizabeth Currid also read and commented on an earlier draft.
Notes
a Top value refers to 1990 and bottom value refers to 2000.
b Top value refers to 1982 and bottom value refers to 1997.
a To recover the marginal effects of both the composite density index and 1990 percent supercreative employment, we compute the respective coefficients with all other variables at their means (from the second column). When this is done, we observe:
1. Florida's creativity theory attempts to specify precise linkages and mechanisms between tolerance and talent, between talent and innovation, and between innovation and growth. CitationFlorida (2002b) shows these linkages with a path model that specifies the links among tolerance, talent, innovation, and income growth. These linkages are also quite clearly spelled out and developed in CitationFlorida (2005). In contrast to extant theory, Florida's creativity theory says that talent is not a stock with which regions are endowed, but a flow that depends on tolerance or openness. Accordingly, places that are open to artistic innovators will be more likely to produce, retain, and attract innovators of all sorts, including technological innovators. Places that produce, attract, and retain more technological innovators and combine them with Schumpeterian economic innovators or entrepreneurs will be more likely to generate new firms and industries and thus to grow. This article does not attempt to test the entirety of this relationship but focuses in detail on one key component—one central mechanism—and that is the effect of geographic concentration or density on this process. It goes beyond the extant literature on the question by specifying the role of this quintessentially spatial element, density, as a key element of the black box of innovation.
2. Density is what enables frequent, unpredictable, serendipitous meetings and interactions. Density is not subordinate—conceptually or empirically—to interaction as some have suggested, and we thus do not make an empirical distinction between the two in this article. For instance, one might use transport systems as a measure of accessibility, but we would argue that it is unlikely that low-density living combined with good transport systems would have the same “buzz” (CitationStorper and Venables 2004) as high-density locations.
3. Storper and Venables (2004, 353–54) differentiate between standard (codifiable) and tacit (noncodifiable) knowledge when they write: Codifiable knowledge has a stable meaning which is associated in a determinate way with the symbol system in which it is expressed, whether it be linguistic, mathematical, or visual. Such information is cheap to transfer because its underlying symbol systems can be widely disseminated through information infrastructure, sharply reducing the marginal cost of individual messages. … By contrast, uncodifiable information is only loosely related to the symbol system in which it is expressed. This includes much linguistic, words-based expression. … Bateson (1973) refers to the “analog” quality of tacit knowledge: communication between individuals which requires a kind of parallel processing of the complexities of an issue, as different dimensions of a problem are perceived and understood only in relation to one another. [Face-to-face] encounters provide an efficient technology under these circumstances, by permitting a depth and speed of feedback that is impossible in other forms of communication. In addition, Allen (2000, 18) differentiates between the “codified knowledges embedded in the global networks identified by Manuel Castells” and “tacit knowledge production based upon convention and customary use.”
4. Therefore, as explained by Kiesler and Cummings (2002, 67), geographically distributed collaborations such as e-mail are not a substitute for physical proximity, especially in tasks where tacit knowledge is central, where work is uncertain, and where interactions need to coordinate interdependent groups. They write: Today, one hears many stories of people forging close work relationships at a distance through electronic communication. Some researchers argue that over time, electronic communication allows for sufficient spontaneous communication to support the development of new close ties. … However, the evidence thus far suggests that physical proximity, with its many spurs to spontaneous communication, serves this purpose better. Work collaborations are more likely to be created and sustained, and are likely to be more satisfying and productive, than distributed (geographically distant) collaborations.
5. This file has population data for metropolitan areas and their components for 1990 and 2000, using 1999 MSA definitions.
6. A number of important points need to be made clear about land area. First, it is assumed that the component land area (county, town, etc.) does not change much over time. Thus county or town land area data from the 2000 Census Factfinder are taken to pertain equally well to 1990. Changing much, however, are the MSA and PMSA definitions across years, in this case 1990 to 1999. These changes are primarily reflected in differences in the components that comprise the MSAs. Counties are often added and dropped from MSAs, and we have accounted for these changes in our calculations of MSA/PMSA land areas for these two years. In making these changes, two issues arose. First, in 1990, some regions were defined as an MSA, but in 1999 were subsumed under an existing MSA or Consolidated Metropolitan Statistical Area (CMSA). When this happened, we conclude that the MSA or CMSA existed in 1990 (without the subsumed MSA), and thus have them both included as data points in 1990. Second, in several cases, regions existed as CMSAs in 1990, but then became MSAs in 1999. Given that no new counties are added or dropped, we simply use the MSA definition for both 1990 and 2000.
7. As documented in the CitationFulton et al. (2001) report, urban land is defined by the NRI as a land cover/use category consisting of residential, industrial, commercial, and institutional land; construction sites; public administrative sites; railroad yards; cemeteries; airports; golf courses; sanitary landfills; sewage treatment plants; water control structures and spillways; other land used for such purposes; small parks (less than ten acres) within urban and built-up areas; and highways, railroads, and other transportation facilities if they are surrounded by urban areas. Also included are tracts of less than ten acres that do not meet the preceding definition but are completely surrounded by urban and built-up land. Two size categories are recognized in the NRI: areas of a quarter-acre to ten acres, and areas of at least ten acres. The authors' 1982 population data come directly from Census estimates, and their 1997 estimates are based on a straight-line interpolation of the 1990 and 2000 Census estimates. The authors also make use of New England County Metropolitan Area (NECMA) definitions for several New England regions including Boston, New London, CT; Hartford, CT; Springfield, MA; Lewiston-Auburn, ME; Pittsfield, MA; Portland, ME; Providence, RI; and Bangor, ME. Given that we use MSA/PMSA definitions in our data set, we are forced to use their NECMA estimates of density in our data set. More detailed information about their methodology is available in their report.
8. Comparing a metro's urban density across years gives an idea of the relative rates at which they are adding population and urban lands. If a metro is urbanizing land faster than it is adding population, its urban density will decrease across years. Conversely, if a metro adds population faster than urbanized land, urban density will increase across years. Also, a metro's marginal density will always be positive if it adds population, even when density decreases across years. Often when land is urbanized faster than population grows, however, marginal density will be small. High marginal densities, however, are often correlated with the size of the city, so that places that have already large population bases and that add more population might tend to have larger marginal densities. So, for an individual metro, more information is revealed by comparing the urban density measures across years. Finally, marginal density will be negative if a metro loses population.
9. This composite density measure is roughly in terms of standard deviations, with some values greater than zero and others less than zero. Subsequently, we interact this variable with another continuous variable, percentage creative employees in a PMSA. To ensure that each variable is on a similar scale (between 0 and 1), we rescale the composite density index such that all values are between 0 and 1, thus creating a variable that is similar in form to percentage creative employees. The rescaled density measure takes this form:
10. This measure includes the following occupations: education administrators, engineers, architects, mathematical and computer scientists, natural scientists, postsecondary teachers, teachers except postsecondary, librarians, archivists, curators, social scientists, urban planners, writers, artists, entertainers, and athletes.
11. Dividing the high-tech percentage of metro output by the high-tech percentage of national output forms a location quotient (LQ) for a metro. An LQ of 1.0 means a metro's concentration of high-tech output is equal to the nation's concentration, although greater than 1.0 means a metro's concentration is higher than the nation's concentration.
12. Given our use of 1990 PUMS data, we are not able to recover the actual density of creative capital. Doing this would require that we obtain, for the numerator of a density measure, an absolute number of creative workers. Because the PUMS primary geographic unit, the Public Use Microdata Area (PUMA), often spills across numerous PMSAs, however, we are forced to exclude those PUMAs from a final total. Thus, we only use PUMAs that are entirely within a PMSA, clearly complicating any attempt to recover an absolute number of creative workers. The percentage of metro area employment that is creative is more appropriate because we assume that on average, the excluded PUMAs are not different than those included, and thus the percentage creative capital is approximately accurate. Finally, we can attempt to estimate the actual creative density, by multiplying the percent supercreative employment by 1990 population, and then dividing by 1990 Census land area. We did this, and note that its correlation with the creative–density interaction is r = 0.8707. We thus use the interaction term, because it enables use of our constructed composite density index.
13. Beyond the results reported in , we estimated several other regressions using proxies for patents as the dependent variable to test the consistency of the findings. When the 2000 Milken Tech-Pole Index and its components—the high-tech LQ and tech share—are inserted as dependent variables, the regression estimation results are very similar to those using patents as the dependent variables. The creative–density interaction is positive and significant in the Tech-Pole and tech share regressions, and positive and insignificant in the LQ regression. Next, we estimated regressions using citation-weighted and industry-weighted patents. The interaction term once again is positive and significant. Overall, these results provide additional evidence in support of the hypotheses. Also, we undertook several procedures to guard against the possibility that our results were determined or overly reliant on the presence of outliers and influential points. First, we took out the top 5 percent of the observations from 1999 patents/100,000 and the creativity–density interaction term. This resulted in twenty-two observations being removed from the data set. When the trimmed 1999 patents/100,000 is regressed on the trimmed interaction term and the other independent variables, we again observe a positive and significant coefficient on the creativity–density interaction term. Finally, the estimation results of an iteratively weighted least squares robust regression procedure also return a positive and significant coefficient on the interaction term.
14. When we remove creativity and creative density, the effect of bohemians on patents turns from negative to positive, but remains insignificant. The effect of gays on patents had been positive but insignificant with creativity and creative density, and when we remove these two variables the effect of gays becomes significant.
15. Current empirical work, such as that by CitationDuranton and Puga (2001) and CitationFeldman and Audretsch (1999), looks at the role of diversity and city size on innovation, finding that larger, more diverse cities and regions are typically more innovative. Among many others, these researchers have probed this diversity–size relationship. Given the different focus of our research, we instead are interested in whether or not density and city size interact in some important way to promote innovation. As such, we include several preliminary empirical tests probing these effects.
16. An additional issue associated with metro size is that a critical mass or threshold of creative persons must be achieved before their presence can have any discernable effect on innovation. In other words, we would look to see whether the absolute number of supercreative employees matters more than the percentage, and also to see whether this critical mass predominates the effect of density. We felt that although this was an important issue, it was outside the scope of this study, and thus chose to pursue it at a later date.
17. By not using the weighted average density measure, however, and by using the PMSA as our unit, we are able to construct and use the composite density index. The value of this composite index was discussed earlier.