Booming Bohemia? Evidence from the US High-Technology Industry

Abstract

This paper assesses the effect of Richard Florida's creative class on economic growth and development at two levels of spatial aggregation. First, I examine the dynamics of economic growth across US metropolitan regions and investigate how they relate to regional specialization and the concentration of talent in the high-tech industry. In addition to evidence of significant high-tech clusters, I identify important complementarities with regard to the interaction between the three Ts of regional development (talent, technology and tolerance) and regional growth dynamics. Using firm-level data, the regional analysis is then complemented by exploring the location of new high-technology plant openings and their relationship with university research and development (R&D) and the creative class. Specifically, I test the hypothesis that both university R&D and the presence of “creativity” generate spillovers which are captured locally in the form of new high-tech establishments, after controlling for important location factors such as local cost, demand and agglomeration economies. While the marginal impacts of increased R&D funding on county probability for new firm formation is modest, the mix of creativity and diversity—as proxied by the Florida measure—appears to be a key driver in the locational choice of new high-tech firms. Separate estimates indicate that these findings hold up across the major high-tech industries in the USA.

Keywords:

• Regional growth
• firm location
• creative class
• high-tech industry
• R&D

Acknowledgements

The author would like to thank the guest editors, Kevin Stolarick, Charlotta Mellander and Richard Florida, and two anonymous referees for their helpful comments. The author is grateful to Nick Kuminoff and Heike Mayer for insights, discussions and vital suggestions on an earlier draft. Inputs from participants at the 2008 Southern Regional Science Association Conference and the 2008 Annual Meetings of the Mid-Continental Regional Science Association are also acknowledged. The author is indebted to Paulo Guimarães for sharing the data on new firm formations.

Notes

¹ Most of the innovation literature uses knowledge generation and high-technology industries as synonymous. However, Hirsch-Kreinsen (2008) shows that “low-tech” industries are far from being characterized by firms that are not innovative. Innovations in sectors with little R&D activity and those in R&D-intensive sectors are to a great extent interdependent, implying that innovation in low-tech industries should not be seen as a contradiction in terms.

² A modified and extended version of this path model and its relevance to the current context are explored in the Appendix.

³ Specifically, I follow Fraley and Raftery (2002, 2006) who implement a cluster analysis based on parameterized Gaussian mixture models. Five clusters are obtained with a BIC outcome of − 8,325.8 using the VVI parameterization (diagonal, varying volume, varying shape) for the component covariance matrix.

⁴ The history of high-tech development in Portland, Boise City and Kansas City, their potential as future leading hubs for the high-tech industry as well as policy options for economic development are discussed in Mayer (2009).

⁵ This finding is particularly interesting in the context of low household savings and metropolitan housing stress as the origins of the current financial crisis. However, further exploration is beyond the scope of this paper and will be pursued elsewhere.

⁶ I am grateful to one of the reviewers for suggesting this specification.

⁷ In fact, this approach is related to, albeit less complex than the structural equation model estimated in Florida et al. (2008b). A modified version of their model—augmented by quality-of-life variables and regional GDP growth—is also estimated for exploratory purposes. The results are presented in the Appendix.

⁸ While overdispersion has consequences similar to those of the presence of heteroskedasticity in linear regression models, the CLM estimator is still consistent if the conditional mean is correctly specified.

⁹ Hellerstein and Mendelsohn (1993) provide a general description of count data models for microeconometric applications.

¹⁰ The summary statistics of the full dataset and a detailed description of the construction of the WFG variables are presented in the Appendix of an earlier version of this paper (Bieri, 2008).

¹¹ WFG use a very broad measure of human capital which is defined as the population percentage of high school graduates. In line with the majority of the literature, however, I define human capital as the share of university degree holders for the remainder of the analysis. Both measures are tested and the results are largely stable for both specifications of human capital. See Table 4 for the summary statistics of the two measures.

¹² In Table 7, the value of δ radius is set equal to 60 miles, which is the value that maximizes the log-likelihood function.

¹³ In line with WFG, I excluded SIC 29 and SIC 34 from the industry analysis, because of the small number of plant openings in these sectors (see Table 5).

¹⁴ I am holding δ constant at the overall optimal level of 60 miles, instead of re-computing optimal levels for all industries as is done in WFG.

¹⁵ See Brakman et al. (2009) for a discussion of how theories that explain the uneven spatial distribution of economic activity (mainly urban economics and New Economic Geography) differ with regard to the relevance they assign to spatial linkages across various levels of spatial aggregation.