OK so I am apparently obsessed with this theme, but I guess that is what makes me a blogger.

Research, like a lot of collaborative activities, encourages specialization.  Successful co-authorships often combine people with differentiated skills.  So successful co-authors are complementary which means that your co-author’s other co-authors are substitutes for you.   This should imply that you are less likely, other things equal, to have a successful co-authorship with your co-author’s co-authors than with, say a randomly selected collaborator.

If we tried to look for evidence of this in data the difficulty would be in holding other things equal.  You are more likely to talk to and have other things in common with your co-author’s co-author than with a random researcher so this would have to be controlled for.

These issues make me think there is some really interesting research waiting to be done taking data from social networks, like patterns of co-authorship or frienship relations on Facebook and trying to simultaneously identify (in the formal sense of that word) “types”  (e.g. technician vs idea-man)  and preferences (e.g. whether these types are complements or substitutes.)  The really interesting part of this must be the econometric theory saying what are the limits of what can and cannot be identified.