Abstract
Although knowledge flows create value, the market often does not price them accordingly. We examine “unintended” knowledge flows that result from the cross-border movement of inventors (i.e., flows that result from the move, but do not go to the hiring firm). We find that the inventor's new country gains from her arrival above and beyond the knowledge flow benefits enjoyed by the firm that recruited her (National Learning by Immigration). Furthermore, the firm that lost the inventor also gains by receiving increased knowledge flows from that individual's new country and firm (Firm Learning from the Diaspora). Surprisingly, the latter effect is only twice as strong when the mover moves within the same multinational firm, suggesting that knowledge flows between inventors do not necessarily follow organizational boundaries, thus creating opportunities for public policy and firm strategy.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
Research on social capital provides a useful framework for understanding knowledge-sharing networks more generally. This research has been impressively multidisciplinary, with important contributions by sociologists (Burt, 1992; Coleman, 1988; Granovetter, 1973), political scientists (Putnam, 2000), and economists (Glaeser, Laibson, & Sacerdote, 2002; Knack & Keefer, 1997). In particular, the concepts of structural holes (Burt, 1992) and weak ties (Granovetter, 1973), which highlight the special role of individuals who provide access to non-redundant networks, offer a useful conceptual framework for explaining why the international movers studied here impact on knowledge flows so significantly: they provide access to knowledge networks that neither the receiving firm and country nor the source firm and country might otherwise have.
Jaffe et al. (1993) find that knowledge flows disproportionately within the city, state, and even the country of the inventor.
No gender assumptions should be inferred from our hypothetical inventor being female; references to the feminine should be understood to include the masculine and vice versa.
We use the term “diaspora” in this paper to describe groups of individuals who share a common history in terms of the firm by which they used to be employed and by the country in which they used to live. So, for example, the IBM Canada diaspora refers to the former employees of IBM Canada who now work for other firms and perhaps in other countries. Other scholars have used the term “diaspora” in a similar context, such as Kapur and McHale (2005).
These papers focus directly on the relationship between labor mobility and knowledge flows. However, other empirical papers address the related link between social relationships and knowledge flows, such as those by Zucker et al. (1998) and Singh (2005). These papers also relate closely to our topic of interest, since we conjecture that labor mobility matters because of the effects of residual social relationships that persist after separation.
We treat member nations of the European Union as distinct countries.
By allowing each dyad member to have its own intercept in the regression specification, we control for omitted time-invariant variables, such as geographic distance and cultural characteristics.
These measures include the number of citations made by the source firm to the mover herself, as well as the number of citations made by the mover to the source firm.
Although we make considerable efforts to minimize measurement errors with respect to identifying movers, our process is by no means perfect. We offer three points regarding the nature and implications of this measurement error. First, we intend for the technology field matching process to remove from the sample individuals who share the same name, but who do not work in the same technology area. By employing this process, we do not, for example, falsely identify Michelle Scott as a mover if there is actually one Michelle Scott who works in textiles in Canada and another who works in electrical connectors in the United States. However, if both Michelle Scotts work in electrical connectors, we still will wrongly identify her as a mover.Second, measurement errors, such as the Michelle Scott example, will bias our main result downwards. In other words, if we erroneously assume an individual is a mover when two different people actually exist, then we will increase the mover variable, but we cannot reasonably expect a related increase in knowledge flows. This will weaken the estimated coefficient on movers. Also, errors in the other direction (we miss actual movers if they spell their names differently before vs after the moves, for example) will add noise to the data but will not systematically bias the results in favor of our hypotheses.Third, to offer the reader some sense of the potential magnitude of the measurement error, we have calculated the fraction of “suspicious” instances where the same name from our sample patented during the same year from two different organizations (1.32%). We also have calculated the fraction of “suspicious” mover instances where an inventor moves from firm A to firm B and later moves back to firm A (2. 2%). While the small fractions calculated here do not prove a small error (in fact, the measured phenomenon might indicate moves that actually occurred), they offer comfort that the potential error is not obviously large.
If the same inventor moves from country A to country B and then to country C, we observe two direct moves (from A to B and B to C) and one indirect move (from A to C). We do not distinguish between direct and indirect moves for the purposes of this analysis.
As described in this section, we condition our sample on firms that existed throughout the study period (1980–2000). This results in dropping a significant fraction of observations. As a result, we bias our sample towards movers who leave larger, older firms. This will not obviously bias our estimated relationship between knowledge flows and labor mobility in either direction, but we note this potential concern and offer the caveat that the generalization of our results to firms of all sizes be considered with caution.
Note that if no variation in the dependent variable exists (i.e., the number of citations remains constant over the 20-year period; this usually occurs when a dyad receives zero citations over the temporal period of our sample), then we drop the observation. This explains why the number of observations falls below 42,860 in certain cases. For example, 1958 groups exist in the base specification in Table 4 instead of 2143. Furthermore, when we employ the full specification, we drop additional observations because the technology overlap index cannot always be computed since we base this measure on a five-year moving average, and sometimes either end of the dyad does not produce any patents during that period.
Although similar in spirit to academic article citations, patent citations serve as a more strict measure of knowledge exchange. For academic article citations, including an additional citation costs close to zero. However, for a patent the cost may be higher, since an additional citation may further reduce the scope of the claims over which it grants the inventor protection, thus reducing its value. We therefore expect fewer spurious citations in patents than in academic article citations.
We note that this measure counts citations, not patents. In other words, we count the number of citations to firm i, conditional on the patent having application year t and at least one of the inventors listed as residing in country j. A patent with such characteristics may not cite firm I, or may cite more than one piece of prior art belonging to firm i, and thus such a patent can increment the citation count by an integer value of 0, 1, or more than 1.
Also, if a cited patent lists multiple inventors located in multiple countries, we count each country.
We also use year dummies instead of a year trend variable. The results remain largely unchanged, but we achieve maximum likelihood estimation convergence more consistently using a time trend.
We adopt the NBER patent classification schema (Jaffe & Trajtenberg, 2002), which aggregates the approximately 420 three-digit USPTO Utility Classes into 36 classes. Whereas the USPTO schema is intended to aid patent examiners with prior art research, the NBER schema aims to reflect basic technology application categories. For example, the NBER classification code of 46 corresponds to Semiconductor Devices, which consists of four USPTO Utility classes: Active Solid-State Devices (257); Electronic Digital Logic Circuitry (326); Semiconductor Device Manufacturing: Process (438); and Superconductor Technology: Apparatus, Material, Process (505).
Jaffe (1986) created this index, and referred to it as an “uncentered correlation coefficient”. Whereas we use the index to measure the technological distance between the source firm and receiving country, Jaffe uses it to measure the technological distance between a focal firm and another firm in its industry. Jaffe employs this to develop a measure of the potential spillover pool available to a firm by multiplying the technological distance measure by each dyad member's R&D spending: the closer a focal firm exists to another firm in technology space, the more it will benefit from the other firm's R&D spending. We follow the more recent literature that has built upon this measure to estimate technological positions between two patenting entities (Acs, Audretsch, & Feldman, 1994; Branstetter, 2001; Peri, 2005; Wu, Levitas, & Priem, 2005).
The 26 countries listed in Table 1 represent all nations that movers in our sample moved from.
We find the Poisson assumption of first and second moment equality too strong for these data. While we still obtain consistent parameter estimates through a Poisson regression model, we greatly underestimate the standard errors, making hypothesis testing difficult. Instead, we adopt the negative binomial regression model, which allows the expected mean of knowledge flows to be proportional to the expected variance (Hausman, Hall, & Griliches, 1984).
We do not present the robustness check tables here, but will provide them upon request.
ZINB, as developed by Greene (1994), assumes that the dependent variable consists of two states unknown to the researcher. In the first regime the likelihood of a variable taking on a value above zero is low, while in the second regime the variable follows a Poisson distribution, where the variable can take on values of both zero and greater. As a result, ZINB estimation involves two distinct parts. The first part distinguishes which regime the observation falls into, in turn “inflating” the zero. We follow tradition and estimate this process using a logit regression. We then use a negative binomial regression to provide coefficient estimates.
Movers from the source firm to the receiving country at times t−5, t−4, t−3 and t−2 are all insignificant.
Andrew Rose kindly provides these data on his website: http://faculty.haas.berkeley.edu/arose/
Recall that the source firm may have multiple movers to Country 2. In this case, “patent stock” is the sum of the patent stocks of each recipient firm in Country 2.
As we examine the effect of movement on knowledge flows in an aggregate sense, we remain agnostic as to the motivation for the move. However, different reasons for moving certainly may result in different flow patterns. For example, if an individual leaves a firm on unfriendly terms because of a falling out, she may sever ties with former colleagues and thus be much less likely to facilitate knowledge flows back to the source firm than another inventor who leaves due to a spouse relocating or for other such reasons. Thus motivation for moving may play an important part in terms of predicting the resultant knowledge flow patterns.
Another example is developing nation “catch-up” policies. It is possible that such policies influence the behavior of both knowledge flows and labor flows in our data set, since we study a reasonably long period (1980–2000) during which many countries made explicit efforts to increase their participation in the innovation-oriented economy. However, since we use dyad fixed effects in our estimations, we take a conservative approach and consider only within-dyad variation. So, empirically, we have no reason to discount or control for public policies that “artificially” increase labor flows to a particular country, which in turn cause an increase in knowledge flows to that country. We want to capture that and attribute the increase in knowledge flows to the increase in immigration. That will not introduce bias into our measure. However, if the policy directly increases both immigration and knowledge flows (in other words, some policy mechanism separate from immigration increases knowledge flows), this presents a problem. For such an effect to bias our measure, the policy would have to increase knowledge flows the year after it increases immigration, and some mechanism other than immigration would have to influence those knowledge flows. We note this possibility but do not consider it a likely occurrence in our data.
References
Acs, Z. J., Audretsch, D. B., & Feldman, M. P. 1994. R&D spillovers and recipient firm size. Review of Economics and Statistics, 76 (2): 336–340.
Agrawal, A. 2006. Engaging the inventor: Exploring licensing strategies for university inventions and the role of latent knowledge. Strategic Management Journal, 27 (1): 63–79.
Agrawal, A., & Cockburn, I. 2003. The anchor tenant hypothesis: Exploring the role of large, local, R&D intensive firms in regional innovation systems. International Journal of Industrial Organization, 21 (9): 1227–1253.
Agrawal, A., Cockburn, I., & McHale, J. 2006a. Gone but not forgotten: Knowledge flows, labor mobility, and enduring social relationships. Journal of Economic Geography, 6 (5): 593–617.
Agrawal, A., Kapur, D., & McHale, J. 2006b. Birds of a feather – Better together? How co-location and co-ethnicity influence knowledge flow patterns. Mimeo, University of Toronto, Toronto, ON.
Almeida, P., & Kogut, B. 1999. Localization of knowledge and the mobility of engineers in regional networks. Management Science, 45 (7): 905–917.
Audretsch, D. B., & Feldman, M. P. 1996. R&D spillovers and the geography of innovation and production. American Economic Review, 86 (3): 630–640.
Branstetter, L. G. 2001. Are knowledge spillovers international or intranational in scope? Microeconometric evidence from the US and Japan. Journal of International Economics, 53 (1): 53–79.
Burt, R. S. 1992. Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press.
Coleman, J. S. 1988. Social capital in the creation of human capital. American Journal of Sociology, 94 (Supplement): S95–S120.
Crane, D. 1969. Social structure in a group of scientists: A test of the “invisible college” hypothesis. American Sociological Review, 34 (3): 335–352.
Dasgupta, P., & David, P. 1987. Information disclosure and the economics of science and technology. In G. Feiwel (Ed.), Arrow and the ascent of modern economic theory: 519–542. New York: New York University Press.
Dasgupta, P., & David, P. 1994. Towards a new economics of science. Research Policy, 23 (5): 487–521.
Glaeser, E. L., Laibson, D., & Sacerdote, B. I. 2002. An economic approach to social capital. Economic Journal, 112 (483): F437–F458.
Granovetter, M. S. 1973. The strength of weak ties. American Journal of Sociology, 78 (6): 1360–1380.
Greene, W. 1994. Accounting for excess zeros and sample selection in Poisson and negative binomial regression models. NYU Department of Economics Working Paper Number EC-94-10, New York University, New York.
Hausman, J., Hall, B. H., & Griliches, Z. 1984. Econometric models for count data with an application to the patents–R&D relationship. Econometrica, 52 (4): 909–938.
Henderson, R., & Cockburn, I. 1996. Scale, scope, and spillovers: The determinants of research productivity in drug discovery. RAND Journal of Economics, 27 (1): 32–59.
Jaffe, A. B. 1986. Technological opportunity and spillovers of R&D: Evidence from firms' patents, profits, and market value. American Economic Review, 76 (5): 984–1001.
Jaffe, A. B., & Trajtenberg, M. 2002. Patents, citations, and innovations: A window on the knowledge economy. Cambridge, MA: MIT Press.
Jaffe, A. B., Trajtenberg, M., & Henderson, R. 1993. Geographic localization of knowledge spillovers as evidenced by patent citations. Quarterly Journal of Economics, 108 (3): 577–598.
Jaffe, A. B., Trajtenberg, M., & Fogarty, M. S. 2002. The meaning of patent citations: Report on the NBER/Case-Western Reserve survey of patentees. In A. B. Jaffe & M. Trajtenberg (Eds), Patents, citations, and innovations: 379–402. Cambridge, MA: MIT Press.
Kapur, D., & McHale, J. 2005. Give us your best and brightest: The global hunt for talent and its impact on the developing world. Washington, DC: Brookings Institution Press.
Knack, S., & Keefer, P. 1997. Does social capital have an economic payoff? The Quarterly Journal of Economics, 112 (4): 1251–1288.
Nelson, R., & Rosenberg, N. 1993. Technical innovation and national systems. In R. Nelson (Ed.), National innovation systems: 3–22. Oxford: Oxford University Press.
Peri, G. 2005. Determinants of knowledge flows and their effect on innovation. Review of Economics and Statistics, 87 (2): 308–322.
Putnam, R. D. 2000. Bowling alone: The collapse and revival of American community. New York: Simon & Schuster.
Romer, P. M. 1986. Increasing returns and long run growth. Journal of Political Economy, 94 (5): 1002–1037.
Romer, P. M. 1990. Endogenous technological change. Journal of Political Economy, 98 (5 Part 2): S71–S102.
Rosenkopf, L., & Almeida, P. 2003. Overcoming local search through alliances and mobility. Management Science, 49 (6): 751–766.
Singh, J. 2005. Collaborative networks as determinants of knowledge diffusion patterns. Management Science, 51 (5): 756–770.
Song, J., Almeida, P., & Wu, G. 2003. Learning-by-hiring: When is mobility more likely to facilitate inter-firm knowledge transfer? Management Science, 49 (4): 351–365.
Thompson, P., & Fox-Kean, M. 2005. Patent citations and the geography of knowledge spillovers: A reassessment. American Economic Review, 95 (1): 450–460.
Wooldridge, J. M. 2002. Econometrics analysis of cross section and panel data. Cambridge, MA: MIT Press.
Wu, S., Levitas, E., & Priem, R. L. 2005. CEO tenure and company invention under differing levels of technological dynamism. Academy of Management Journal, 48 (5): 859–873.
Zucker, L., Darby, M., & Brewer, M. 1998. Intellectual capital and the birth of US biotechnology enterprises. American Economic Review, 88 (1): 290–306.
Acknowledgements
We thank the JIBS departmental editor, Tom Murtha, and three anonymous reviewers for helpful suggestions and comments. The Social Sciences and Humanities Research Council of Canada (Grant No. 410-2004-1770), Human Resources and Skills Development Canada, and Industry Canada (Grant No. 537-2004-1001) funded this research. We gratefully acknowledge their support. Errors and omissions are our own.
Author information
Authors and Affiliations
Corresponding author
Additional information
Accepted by Thomas Murtha, Departmental Editor, 13 June 2007. This paper has been with the authors for three revisions.
Rights and permissions
About this article
Cite this article
Oettl, A., Agrawal, A. International labor mobility and knowledge flow externalities. J Int Bus Stud 39, 1242–1260 (2008). https://doi.org/10.1057/palgrave.jibs.8400358
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1057/palgrave.jibs.8400358