Too much of a good thing? Network brokerage within and between regions and innovation performance

ABSTRACT

The paper offers a multilevel understanding of the effect of brokerage on regional innovation. We develop a typology of regional networks based on the extent to which the inventors of a region connect the otherwise disconnected group of local inventors (internal brokerage) or connect the local network with other regions (external boundary-spanning). Using data on regional innovation performance and co-inventing networks within and between US metropolitan statistical areas (MSAs) between 2000 and 2014, we show that configurations balancing high (low) internal brokerage and low (high) external boundary-spanning lead to higher innovation performance than those where brokerage occurs at both levels of analysis.

KEYWORDS:

• networks
• brokerage
• boundary-spanning inventors
• multilevel networks
• patents
• innovation

JEL:

• O31
• R11

ACKNOWLEDGEMENTS

We thank Stefano Breschi (Bocconi University), Corey Phelps (University of Oklahoma), Sen Chai (ESSEC Business School) and Raymond-Alain Thietart (ESSEC Business School), who all provided feedback on the broad project comprising this paper as members of Amit Kumar’s dissertation committee. We are grateful to Fabrizio Fusillo (ESSEC Business School, University of Turin) for research assistance.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Notes

1. Our framework applies to regions defined as medium-large metropolitan areas with established administrative boundaries. We operationalize regions using US metropolitan statistical areas (MSAs). We do not make explicit assumptions concerning the specialization of economic activities within a given region, but we control for it in the empirical analyses. In this respect, our work differs from research on clusters, which focus on the spatial concentration of actors specialized in the same industry.

2. Scholars refer to the first type of externalities as ‘Marshall–Arrow–Romer (MAR) externalities’, while they use the label ‘Jacobs’ externalities’ to characterize spillovers stemming from knowledge diversity.

3. This paper focuses on the advantages and disadvantages of internal brokerage and external boundary-spanning, irrespective of the type of actor (private firm, university) occupying this network position. However, in the descriptive statistics we provide some insight about the relative propensity of different types of actor to broker within and between regions.

4. See http://www.nlsinfo.org/usersvc/NLSY97/NLSY97Rnd9geocodeCodebookSupplement/gatt101.htm#cbsa8/.

5. The dependent and independent/control variables build on non-overlapping sets of patents. The network variables are built using patents filed between t – 3 and t – 1, and the control variables are based on patents filed in the same time window. By contrast, our dependent variable is the count of patent applications filed in year t weighted by the forward citations these patents received until t + 5.

6. In unreported analyses, we also controlled for network transitivity and hierarchy. The results remain unchanged. As these controls increased the VIFs without increasing model fit, we excluded them from the main analysis.

7. Out of a full sample of 5181 observations (MSA–state–year), 5053 observations met these conditions, while 181 observations (2.47% of the full sample) were dropped. The results based on alternative thresholds, such as five and 15 inventors, are consistent with those discussed in the next section.

8. Citation-weighted patent counts are also estimated using a conditional fixed-effect negative binomial (e.g., Fleming, 2001). We repeated the analysis using this econometric model. The results remained robust.

9. To make sure that the examples provided are comparable, we provide only examples from the top quartile of MSAs with respect to size. We discriminate between high and low values of external boundary spanning and internal brokerage using t < 33% percentile (low) and > 66% percentile (high). We consider an MSA to belong to a quadrant if it was positioned in that quadrant for at least nine years between 2003 and 2014.

10. We also recomputed these measures using a 50- and a 75-mile radius from the focal MSA. In line with Kang and Dall’Erba (2016), these variables have a stronger effect when considering a broader geographical radius. The results concerning our key variables remain unaltered.

11. We used the coordinates of the centre of the MSA (latitude, longitude) to compute distances between pairs of collaborating inventors, using the Stata geonear routine.

12. We also performed the analysis using the maximum number of collaboration ties maintained by inventors in the focal MSA. The results remain similar in terms of size and statistical significance to those discussed in the paper.

 

Additional information

Funding

This work was supported by the Agence Nationale de la Recherche [grant numbers ANR-18-CE26-0005 and ANR-16-IDEX-0008, and convention number 2018-EP-55].