Core-Periphery Segregation in Evolving Prisoners Dilemma Networks

Dense cooperative networks are an essential element of social capital for a prosperous society. These networks enable individuals to overcome collective action dilemmas by enhancing trust. In many biological and social settings, network structures evolve endogenously as agents exit relationships and build new ones. However, the process by which evolutionary dynamics lead to self-organization of dense cooperative networks has not been explored. Our large group prisoner’s dilemma experiments with exit and partner choice options show that core-periphery segregation of cooperators and defectors drives the emergence of cooperation. Cooperators’ Quit-for-Tat and defectors’ Roving strategy lead to a highly asymmetric core and periphery structure. Densely connected to each other, cooperators successfully isolate defectors and earn larger payoffs than defectors. Our analysis of the topological characteristics of evolving networks illuminates how social capital is generated.

💡 Research Summary

This paper investigates how dense cooperative networks self‑organize when individuals can both sever existing ties and form new ones, using large‑scale laboratory experiments of the two‑person Prisoner’s Dilemma (PD) with partner‑choice and exit options. Four experimental sessions were conducted, each with 35–40 participants playing 20 rounds. In each round participants first propose partners; a link is formed only if the proposal is mutual, after which a PD game is played on that link. Maintaining k links incurs a quadratic cost (4.4(k‑1)² in the low‑cost condition, 8.8(k‑1)² in the high‑cost condition), limiting the number of profitable connections.

Across both cost settings, cooperation rose steadily: the proportion of cooperative choices (%C_total) increased from roughly 55 % in the first round to 60 % (low‑cost) and 76 % (high‑cost) by round 19, before a sharp end‑game decline to about 21 % in the final round. Positive assortment—excess of CC and DD dyads relative to a random baseline—also grew over time, and the Pearson correlation between partners’ individual cooperation rates increased, indicating that cooperators increasingly paired with each other.

Network‑level analysis revealed that modularity was lower than in degree‑preserving random networks, suggesting the emergence of a core‑periphery rather than modular community structure. To locate nodes on the core‑periphery spectrum, the authors applied k‑shell decomposition, assigning each node a coreness index k_s. Nodes with high k_s formed a densely connected core, while low‑k_s nodes occupied a sparse periphery. The correlation between a player’s %C and k_s strengthened over rounds, confirming that high cooperators gravitated toward the core.

Behaviorally, two distinct networking strategies drove this segregation. Cooperators employed a “Quit‑for‑Tat” rule: they never abandoned cooperative partners but immediately cut ties with defectors. Defectors used a “Roving” strategy, frequently dropping existing partners—even cooperators—to seek new victims. Consequently, cooperative dyads persisted with a 95 % continuation probability, whereas mutual‑defection dyads continued only ~13 % of the time and mixed (C‑D) dyads ~21 %. Links involving high‑k_s nodes became increasingly stable, while peripheral links grew more volatile.

Earnings followed a similar trajectory. Early rounds favored defectors, but as the core solidified, cooperators earned higher average payoffs because they enjoyed many mutually cooperative links, whereas defectors had few links and most of those yielded low or negative returns. Thus, the endogenous network formation acted as a self‑organized punishment mechanism: the structure itself marginalized defectors without requiring reputation systems or indirect reciprocity.

The findings were robust across low‑ and high‑cost conditions, indicating that the core‑periphery segregation and cooperation enhancement are not artifacts of a particular cost level. The authors conclude that preferential partner choice alone can generate dense cooperative clusters, providing a concrete topological realization of social capital. They raise open questions about why some societies fail to develop such cores and under what circumstances peripheral defectors might transition into the cooperative core, suggesting fruitful directions for future research on the co‑evolution of strategy and network topology.