“Inventor Mobility and Knowledge Transmission in Nanotechnology,” J. Kim, S.J. Lee & G. Marschke (2010)

How do firms acquire knowledge? Surely books and consultants diffuse knowledge, but there is theoretical and anecdotal evidence suggesting that long-term, repeated contact is necessary for knowledge to diffuse fully – this is called “tacit knowledge” and is essentially the reason students go to school rather than read textbooks in their houses. Quantitatively describing this effect is difficult, however; how exactly would one measure the diffusion of tacit knowledge?

The authors of this paper examine patent citation data, creating a matched list of inventors and the companies their patents were linked to, and attempt to find increase citation of company A’s patents when an employee moves from A to B. They find just such an effect. Unfortunately, this effect is not convincing for three reasons. First, only 34% of names are matched to patents, and given the way foreign names are transcribed, this biases included patents to those from American inventors and inventors with Western names. Given that there is some racial homophily in companies (firms with many Chinese engineers tend to hire more Chinese, etc.), it’s not surprising that firms with many matched names also cite patents from other firms with easy-to-match names. Second, there is no dummy for geography, and there is widespread evidence suggesting that being located in the same region makes patents more likely to be cited. Since workers who shift firms predominantly shift to another firm in the same area, the geographic effect could be driving the result. Third, and most importantly, the statistical model compares citations of patents from a company where B’s employee used to work to any patent in Patent Subclass 19 – Chemicals, Misc., a broad category. A recent paper from P. Thompson and M. Fox-Kean (http://www.fiu.edu/~thompsop/current/adobe/localization.pdf) showed, in the context of the well-known patent paper by Jaffe et al, that such broad categories will bias the estimator because firms in the same subclass of chemicals are both more likely to hire the same workers and more likely to cite each other’s patents. Using finer and finer partitions of the patent classification can avoid this problem, but it’s tough to get around altogether. I’m still waiting to see a paper with an optimal method for examining large-scale diffusion of knowledge – it’s an important subject to have been left uncovered for so long!


%d bloggers like this: