MANAGEMENT SCIENCE Informs ® Vol
Total Page:16
File Type:pdf, Size:1020Kb
MANAGEMENT SCIENCE informs ® Vol. 51, No. 5, May 2005, pp. 756–770 doi 10.1287/mnsc.1040.0349 issn 0025-1909 eissn 1526-5501 05 5105 0756 © 2005 INFORMS Collaborative Networks as Determinants of Knowledge Diffusion Patterns Jasjit Singh INSEAD, 1 Ayer Rajah Avenue, 138676 Singapore {[email protected], http://www.jasjitsingh.com} his paper examines whether interpersonal networks help explain two widely documented patterns of Tknowledge diffusion: (1) geographic localization of knowledge flows, and (2) concentration of knowledge flows within firm boundaries. I measure knowledge flows using patent citation data, and employ a novel regression framework based on choice-based sampling to estimate the probability of knowledge flow between inventors of any two patents. As expected, intraregional and intrafirm knowledge flows are found to be stronger than those across regional or firm boundaries. I explore whether these patterns can be explained by direct and indirect network ties among inventors, as inferred from past collaborations among them. The existence of a tie is found to be associated with a greater probability of knowledge flow, with the probability decreasing as the path length (geodesic) increases. Furthermore, the effect of regional or firm boundaries on knowledge flow decreases once interpersonal ties have been accounted for. In fact, being in the same region or firm is found to have little additional effect on the probability of knowledge flow among inventors who already have close network ties. The overall evidence is consistent with a view that interpersonal networks are important in determining observed patterns of knowledge diffusion. Key words: knowledge spillovers; technology diffusion; social networks; interpersonal ties; innovation History: Accepted by Scott Shane, technological innovation, product development, and entrepreneurship; received May 4, 2004. This paper was with the author 2 months for 1 revision. 1. Introduction separate identification of the impact of direct versus Acquisition of knowledge can be crucial for economic indirect ties on knowledge flow. success of firms and for innovativeness and growth Extending existing research methodology for using of geographic regions (Grossman and Helpman 1991). patent citation data to measure knowledge flows (e.g., However, knowledge acquisition is not easy. Even Jaffe and Trajtenberg 2002), I employ a novel regres- though ideas are intangible in nature, they can be sion framework based on choice-based sampling to extremely hard to transmit across regional or firm estimate the probability of knowledge flow between boundaries. In particular, two patterns of knowledge inventors of any two patents. Consistent with past diffusion have been documented. First, knowledge research, I find that knowledge flows are stronger flows are geographically localized (Jaffe et al. 1993). within than between regions or firms. In particular, Second, knowledge diffuses more easily within a firm even after carefully controlling for technological spe- than between firms (Kogut and Zander 1992). In this cialization of regions (Thompson and Fox-Kean 2005), paper, I formally examine interpersonal networks as knowledge flow between two inventors from the the driver behind these patterns. Although other fac- same region is found to be 66% more likely than that tors such as institutions, norms, language, culture, between two inventors from different regions. Like- or incentives could also influence how knowledge wise, all else being equal, knowledge flow between diffuses, I estimate how much of the empirical pat- two inventors is three times as likely within than be- tern of knowledge diffusion can be explained sim- tween firms. ply by the fact that people within a region or a firm A rich literature in sociology has emphasized infor- have close interpersonal ties. Although the main focus mation flow through interpersonal networks (Ryan of this paper is to study the role of interpersonal and Gross 1943, Coleman et al. 1966, Granovetter networks in determining knowledge diffusion pat- 1973, Burt 1992, Rogers 1995). This has motivated terns, its contributions to the literature also include the current study investigating whether such inter- an improved methodology for analyzing microlevel personal networks are behind the observed patterns knowledge flows, a richer data set than is typically of knowledge diffusion mentioned above. Whereas it employed in studies on knowledge diffusion, and a can be difficult to obtain large-scale systematic data 756 Singh: Collaborative Networks as Determinants of Knowledge Diffusion Patterns Management Science 51(5), pp. 756–770, © 2005 INFORMS 757 on interpersonal relations, we can infer such relations of relatively small magnitude simply because patent indirectly by studying secondary data on collabora- collaborations capture only a fraction of all relevant tions between individuals (Cockburn and Henderson interpersonal ties. 1998, Newman 2001, Fleming et al. 2004). Follow- This paper is organized as follows. Section 2 moti- ing this approach, I use collaboration information for vates my formal hypotheses. Section 3 describes the patents registered with the United States Patent and data on patent citations and on inventors. Section 4 Trademark Office (USPTO) to construct a rich longi- introduces my citation-level regression framework tudinal database of interpersonal relations among all for estimating probability of knowledge flow, and inventors recorded by USPTO since 1975. This forms describes how I measure interpersonal ties. Section 5 the basis of constructing a social proximity graph for reports and explains the empirical findings. Section 6 over 1 million inventors, which is then used to mea- discusses some open empirical issues and possible sure the social distance between any two inventors. future extensions. Section 7 offers implications and I capture both direct and indirect interpersonal rela- concluding thoughts. tions. For example, if individual X has a direct tie with individual Y , and Y has a direct tie with Z, I allow the possibility that Z might learn indirectly 2. Hypotheses about X’s work through his or her tie with Y .In The analysis in this paper comprises three main parts, the analysis, collaborative ties between inventors are as summarized in Figure 1. The first part formally found to be an important determinant of the proba- establishes that intraregional and intrafirm knowl- bility of knowledge flow, with this probability falling edge flows are indeed more intense than the knowl- with increase in social distance. For example, for edge flows between different regions or firms. The inventors with direct collaborative ties the probabil- second part tests the extent to which existence and ity of knowledge flow is four times that for inventors directness of interpersonal ties between individu- who are not connected, and it is 3.2 times that for als determines the probability of knowledge flow inventors who have no direct tie, but who have an between them. The third part, which is the crux of indirect tie through past collaboration. this paper, combines the constructs from the first two The main analysis of the paper studies whether the parts to examine the extent to which interpersonal above interpersonal networks help explain the pat- networks help explain the intense intraregional and terns of knowledge diffusion discussed earlier. I find intrafirm knowledge flows. that the effect of being in the same region or the Although previous work has documented geo- same firm on the probability of knowledge flow does graphic localization of knowledge flow (e.g., Jaffe decrease once collaborative networks have been taken et al. 1993), recent work raises methodological con- into account. Although the decrease is nontrivial cerns that could have led to overestimation of this in magnitude and statistically highly significant, the phenomenon (Thompson and Fox-Kean 2005). There- magnitude of this effect is not as large as one might fore, before trying to explain the result on intra- expect: Once interpersonal ties have been controlled regional knowledge flows, I first test if the result for, there is a 17% decrease in the effect of geographic does indeed continue to hold even when using a new colocation on probability of knowledge flow, and a empirical approach (as detailed in §4) that addresses 12% decrease in the effect of being within the same some of these concerns: firm on the probability of knowledge flow. However, because only a subset of all interpersonal relations are captured by a network constructed exclusively Figure 1 Summary of Hypotheses from patent collaboration data, these results should be (a) Hypotheses 1 and 2 viewed as a lower bound on the importance of inter- Same region personal networks. This interpretation is strengthened Greater probability of by the additional finding that geographic proximity knowledge flow Same firm and firm boundaries have little additional effect on the probability of knowledge flow between inventors (b) Hypotheses 3 and 4 who are already closely connected in the collaboration network. The regional and firm boundaries continue Close collaborative links Greater probability of to matter more for knowledge flow between inven- between individuals knowledge flow tors with no ties or with only very indirect ties. As explained in §5, this is consistent with a view that (c) Hypotheses 5 and 6 interpersonal networks are actually quite important in Same region causing the more intense intraregional and intrafirm Close collaborative links Greater probability