Policy ForumInnovation Economics

Where is Silicon Valley?

See allHide authors and affiliations

Science  06 Feb 2015:
Vol. 347, Issue 6222, pp. 606-609
DOI: 10.1126/science.aaa0201

Embedded Image

Although economists, politicians, and business leaders have long emphasized the importance of entrepreneurship (1, 2), defining and characterizing entrepreneurship has been elusive (3, 4). Researchers have been unable to systematically connect the type of high-impact entrepreneurship found in regions such as Silicon Valley with the overall incidence of entrepreneurship in the population (57). This has important implications: Researchers arrive at alternative conclusions about roles and patterns of entrepreneurship (810), and policy-makers are given conflicting recommendations about whether or how to promote entrepreneurship for economic and social progress (11, 12).

To break this impasse, we introduce a new method for studying the founding and growth of entrepreneurial ventures. Whereas most prior studies have focused on the quantity of entrepreneurial ventures (e.g., the number of new businesses per capita in a given region), we focus on characterizing their quality. Rather than assume that all ventures have an equal ex ante probability of success, our method allows us to estimate the probability of growth based on information publicly available at or near the time of founding.

We implement our approach using for-profit business registrations in California from 2001 to 2011 (13), combined with data from the U.S. Patent and Trademark Office and SDC Platinum [details on data and methods are in the supplementary materials (SM)]. We estimate outcomes on the basis of a small number of start-up characteristics: (i) firm name characteristics, including whether the firm name is eponymous [named after the founder (14)], is short or long, is associated with local business activity or regionally traded clusters (e.g., dry cleaning versus manufactured goods), or is associated with a set of high-technology industry clusters (15, 16); (ii) how the firm is registered, including whether it is a corporation [rather than partnership or limited liability company (LLC)] and whether it is incorporated in Delaware (17); and (iii) whether the firm establishes control over formal intellectual property (IP) rights within 1 year of registration (18).

Estimating entrepreneurial quality by city.

Each bubble represents a city. Bubble size reflects city population. Bubble color varies according to quality scale at bottom of figure. Each row represents distinct geographic region. See SM.

To ensure that our estimate reflects the quality of start-ups in a location rather than assuming that start-ups from a given location are associated with a given level of quality, we exclude location-specific measures from the set of observable start-up characteristics. We estimate entrepreneurial quality as the probability of achieving a meaningful growth outcome—defined as an initial public offering (IPO) or an acquisition (19) within 6 years of founding—as a function of these start-up characteristics. This predictive, location-agnostic algorithm can then be used to independently characterize the entrepreneurial quality of firms and locations.

ESTIMATING ENTREPRENEURIAL QUALITY. We estimate entrepreneurial quality through a logit model with a randomly selected sample of 70% of all firms registered in 2001–2006 (keeping the other 30% as a test sample). Our model incorporates business registration and IP factors in a single regression, with all coefficients significant at the 5% level (20) (table S1). When we look at firm name characteristics, eponymous firms are more than 70% less likely to grow than noneponymous firms, whereas firms with short names are 50% more likely to grow than firms with long names, and firms that include words associated with hightechnology clusters are 92% more likely to grow than others. Looking at legal form and IP, corporations are >6 times more likely to grow than noncorporations, and firms with trademarks are >5 times more likely to grow than nontrademarked firms. Patenting and Delaware jurisdiction play an outsized role: Each alone is associated with a >25 times increase in the probability of growth relative to not being present. When both are present at the same time, there is nearly a 200 times increase in the probability of growth.

As a validation test, we estimate entrepreneurial quality for the test sample withheld from the original regression and so compare our predictions of entrepreneurial quality to the actual outcome distribution. Our estimate of entrepreneurial quality is strongly related to out-of-sample outcomes: 76% of all growth outcomes in the test sample are within the top 5% of the distribution of estimated entrepreneurial quality, with 56% drawn from the top 1% of that distribution (fig. S1). Highlighting the extreme uncertainty associated with entrepreneurship, growth is still rare: Even within the top 1% of estimated entrepreneurial quality, the average firm has only a 5% chance of realizing a growth outcome. This is consistent with recent findings that start-up growth is skewed relative to overall firm growth—Gibrat's law (21).

Mapping estimated entrepreneurial quality by ZIP Code.

San Francisco Bay area. Dots indicate single-address ZIP Codes. See SM.

MAPPING ENTREPRENEURSHIP. The centerpiece of our analysis focuses on recent cohorts before a growth outcome has occurred (i.e., all start-ups from 2007 to 2011). We estimate the entrepreneurial quality for each firm and then calculate the average estimated quality of firms by city and, separately, by ZIP Code. These scores can be interpreted as the expected number of growth outcomes per 1000 start-ups in the 2007–2011 cohorts.

Average quality across municipalities is shown in the first figure. Silicon Valley stands out from other regions across California: Startups in Menlo Park, Mountain View, Palo Alto, and Sunnyvale have 20 times the average quality of the median city and 90 times that of the lowest-ranked cities in California. Among large cities, San Francisco registers an entrepreneurial quality level nearly 8 times that of Fresno.

Entrepreneurial quality is mapped for the San Francisco Bay area at the ZIP Code level in the second figure. The quality of entrepreneurial activity is distinctively higher in the area that ranges just north of San Jose through San Francisco, with a contiguous mass of intense entrepreneurial quality from just southeast of Google (and the founding location of Fairchild) through Milbrae and Burlingame. In contrast, the Los Angeles region has a much lower level of entrepreneurial quality (fig. S2). Large economic areas can vary significantly in their quality. We investigated the statistical relation between quality and quantity (fig. S3): At best, the relation is weak and noisy. Intriguingly, across regions, entrepreneurial quality is centered around research institutions, such as universities and national laboratories. Stanford is at the heart of Silicon Valley, and University of California (UC) Berkeley; Lawrence Livermore; Caltech; University of California, Los Angeles (UCLA); and UC Irvine each host a region of distinctive entrepreneurial quality.

IMPLICATIONS. By focusing on entrepreneurial quality, we can evaluate more clearly the role of location and institutions in firm growth. For example, our method allows us to estimate a locational entrepreneurship “premium” as the difference between realized and expected growth outcomes for a region. Between 2001 and 2006, Silicon Valley had 60% more actual growth events than predicted by our model, whereas Los Angeles registered 13% fewer than predicted.

Our method can be extended to evaluate entrepreneurial quality at arbitrary levels of geographic aggregation (e.g., a specific street in Palo Alto) (fig. S4). This facilitates finegrained analysis of entrepreneurial dynamics (22), distinguishing empirically (although not causally) between locations at a high level of granularity.

Finally, beyond our characterization of Silicon Valley in the aggregate, our results highlight the role of research institutions as centers of entrepreneurial quality. Characterizing the two-way relation between entrepreneurial quality and scientific research activity is a promising agenda for future research. Although one would need to be cautious about using these estimates as a policy tool (for example, one could imagine “gaming” of various sorts), clarifying the conditions that facilitate positive growth outcomes has important implications for policy-makers and regional stakeholders.

References and Notes

  1. Formal registration includes corporation, LLC, limited partnership, and general partnership.
  2. We define traded and local industries in line with the definition used in the economic cluster literature [e.g., (15)], and high-technology clusters are drawn from the U.S. Cluster Mapping Project [see (15)].
  3. Many firms with the intention to grow register in the state of Delaware, where corporate law is beneficial owing to a large legal canon. Venture capitalists often prefer companies to incorporate in Delaware.
  4. Our use of firm names builds on a basic assumption that entrepreneurs choose firm names conscientiously to serve as a signal to consumers, investors, and employees and that there are costs in impersonating a different type of firm.
  5. An IPO or acquisition represents a significant and observable equity growth outcome from the perspective of the founders. Our ongoing research agenda also explores alternative growth outcomes in terms of employment, firm revenues, and so on.
  6. Results are similar when we look at business registration and IP factors alone; see columns 1 and 2 in fig. S1.
  7. Acknowledgments: We thank the Jean Hammond (1986) and Michael Krasner (1974) Entrepreneurship Fund and the Edward B. Roberts (1957) Entrepreneurship Fund for financial support. We thank participants in the Massachusetts Institute of Technology (MIT) Innovation and the Digital Economy Seminar, the MIT Regional Entrepreneurship Acceleration Program, the Micro@Sloan Seminar, the Asian Innovation and Entrepreneurship Association–NBER Conference, and three anonymous referees for comments. I. Cockburn, M. Delgado, J. Gans, H. Varian, and C. Fazio provided valuable advice. We thank R. J. Andrews for excellent development of figures and visualizations, A. Carracuzzo and MIT Libraries for data support, and I. DiMambro for editorial assistance.
View Abstract

Navigate This Article