Structure k-cohesion experiment
From InterSciWiki
See Cooperation : The New Palgrave Dictionary of Economics on Panchanathan, K. & Boyd, R. Indirect reciprocity can stabilize cooperation without the second-order free rider problem and the review of their work by James Fowler: Panchanathan, K. & Boyd, R. on Human cooperation Second-order free-riding problem solved? Nature 432, 499–502 (2004). Here the Karthik Panchanathan & Rob Boyd (2005) reply: Human cooperation: Second-order free riding problem solved? (reply). Nature 437: E8-E9.
http://intersci.ss.uci.edu/wiki/pw/Experiment_k-cohesion.pdf (draft 0)
http://intersci.ss.uci.edu/wiki/pub/k-cohesion1.pdf (draft 1 - incorporates comments from Aljaz Ule through Jeroen Bruggeman)
http://intersci.ss.uci.edu/wiki/pub/k-cohesion2.pdf (draft 2 - Jeroen Bruggeman improves DW revision 1)
1. In the first paragraph you seem to say that "Reputation ... is better avoided", which I don't understand; why?
I thought that recent experiments cast some doubt on the generic effectiveness of punishment in establishing cooperation - or have these authors never dealt with small children? In any case, if reputation can in part establish cooperation, then less punishment is necessary to establish cooperation (Rockenbach and Milinski’s result), which avoids players to rely strongly on punishment (and us to deal with it in our explanation) – betting on the safe side it is, rather than avoiding punishment altogether.
2. You also say there that a clique is a "maximally cohesive group". Is cohesion then defined only in terms of the structure of interaction? Reading above this sentence I understood that cohesion involves shared norms and emotions - something that may not be captured simply by a structure of the network. Also, the idea that k-connectivity captures cohesion also suggests that it all reduces to the network structure.
In Douglas White’s work, there is a distinction between the ideational and the relational components of cohesion, although I doubt that one can distinguish them completely. For sure, if there is higher k-connectivity relational-wise, White would hypothesize that there is also a higher level of shared norms and emotions; the latter are hard to measure, though, while the k-connectivity is in the hands of the experimenter. The effect of higher shared norms and emotions can be measured, though, as the contributions to the collective goal. The reason why one can’t distinguish the two components of cohesion is this: a network tie is a set of expectations and memories, possibly incorporating shared norms and emotions; once there is a tie, the shared norms, expectations and emotions can’t be cut away (without violating all ethics on brain surgery during game experiments, that is).
3. It is not clear to me why you intend to use Rockenbach and Milinski's game appended by limited possibility for social pressure. That game had a specific intention to compere the varying roles of rewards and punishment. It was not suggested to be a realistic game of social cooperation. The complicated game might make analysis of experiments very demanding.
Let’s keep the experiment as simple and as cheap as possible! Please ignore my ignorance on these matters; I used their paper as a template, lacking experience in doing experiments.
4. Did you plan that the public goods game is played across all players or between neighbors only? Is the whole collection of games played once or is it repeated?
Indeed the interactions should be limited to network neighbors according to the figures. Some repetition seems necessary, though, because then players can anticipate that a bad reputation, or selfish action, in a current round may harm them in future rounds.
5. Do you already have any specific research questions in mind? What about hypotheses?
Question: does k-connectivity increase cooperation in a public goods game? Hypothesis: for a small group, the higher k-connectivity, the higher the yield in the public goods game. For large groups (much larger than in the proposal) I would expect a non-monotonic effect, because people would feel ‘suffocated’ by too many ties, get annoyed, and reduce their contributions.
6. Finally, my experience is that many treatments lead to lots of complications in analysis and interpretation. Why not simply focussing on the last comparison between networks in Figures 4 and 5? If any difference in cooperation is observed there you can clearly conclude it's due to the k-connectivity factor.
JB: You are right - of course the two groups in Figures 4 and 5 are enough.
Seems to me, though, we might have a seven-person circle k=2 in addition to figs 2 and 3. That provides not only a contrast between a 2 subgroups with broker graph (fig 2) and a single k=3 group (fig 3) both with same density, average path distance, degree centralization, group size --- but a contrast with a moderate-cohesion group (k=2) with no centralization at all. If that comes up as "higher" on some outcome than fig 3, then the effect is not cohesion but centralization....
Doug
The circle (k=2) seems excellent!
I can also try at Oxford, I know very well a clever Chicago postdoc (very network oriented political scientist) who just got a job there and told me they have a lab, and she and I got some European funding to pay for network related research, which can be spent on the experiment (within Europe, that is, not if it's in the USA). Shall I await your request first?
Jeroen.
Version of the document at 30 September 2008: 7-cycle (k=2) replaced by graph (k=3) that is comparable to the other graphs, due to its having the same average path distance and density; to the paper some minor changes, and references added by Jeroen. http://users.fmg.uva.nl/jbruggeman/k-cohesion3.pdf (draft 3 - Jeroen Bruggeman improves revision 2)
