Do small world networks promote innovation?

After reading the HBS Working Knowledge interview of HBS Professor Lee Fleming, I was happy to connect with Lee in person early this week. We discussed his working paper, "Small Worlds and Regional Innovative Advantage," co-authored with Charles King and Adam Juda.



This paper seeks to answer the question: Do small world networks promote innovation?



[By "small world" they mean that people tend to connect to their neighbors (hence we feel like local actors) but there are enough other links so that there is a short path of relationships between any two people, even across great distance.]



Fleming et al note that many have argued "yes" to this question but with scant evidence. So they have collected an impressive body of data to put the question to a more rigorous test.



Their results? My reading of the paper is that small world networks (at least the ones they studied) aren't so closely related to innovation after all. When I put that to Lee, he declined to sound so definitive and said, "The jury is still out."



Roughly speaking, here is what they did: Lee's team collected records of 2.5 million US patents filed by 2 million inventors over the last 25 years. Then they created a multi-component network by linking inventors with jointly filed patents. Finally, they looked at the largest connected component in each of 337 metropolitan areas, calculated 19 different characteristics of each (from both economic and network perspectives) and determined the correlations of these characteristics across all 337 regions.



Which characteristics correlate to more patenting? The strongest predictors of a productive patenting network of inventors are (1) one type of technology dominates the network of inventors, rather than many different kinds of technology, (2) young technology, (3) one firm dominates the network, rather than many different firms.



As for small world characteristics, Fleming et al found that clustering (connectedness of neighbors) has a negative relationship with innovation. Short path length has a statisticaly robust positive relationship, but its magnitude is several hundred times weaker than the positive effect of a regional monopoly (#3 above), for example.



Lee has left the "30,000-foot" view of this research, and his next project is to investigate a more closeup view that studies individual inventors in much greater detail. So far he is finding similar results (clustering is bad, short path lenghts marginally good) but again, the jury is still out.



After reading how BzzAgent succeeds at viral marketing without even considering SNA, and now reading Lee's research, my sense of social network analysis has just become significantly more abstract. These results are more than a little discouraging to anyone hoping to sell SNA as a engine of productivity. But the SNA practitioner in me is still encouraged, on balance, by results such as Burt's study of Raytheon managers. And the SNA researcher in me is excited to see how these studies continue to unfold, either way.