Complex Network Analysis in Cricket : Community structure, players role and performance index

This paper describes the applications of network methods for understanding interaction within members of sport teams.We analyze the interaction of batsmen in International Cricket matches. We generate batting partnership network (BPN) for different teams and determine the exact values of clustering coefficient, average degree, average shortest path length of the networks and compare them with the Erd\text{"{o}}s-R\text{'{e}}nyi model. We observe that the networks display small-world behavior and are disassortative in nature. We find that most connected batsman is not necessarily the most central and most central players are not necessarily the one with high batting averages. We study the community structure of the BPNs and identify each player’s role based on inter-community and intra-community links. We observe that {\it Sir DG Bradman}, regarded as the best batsman in Cricket history does not occupy the central position in the network $-$ the so-called connector hub. We extend our analysis to quantify the performance, relative importance and effect of removing a player from the team, based on different centrality scores.

💡 Research Summary

The paper applies complex‑network theory to the analysis of batting partnerships in international cricket, constructing a Batting Partnership Network (BPN) for each national side. In a BPN, every batsman is represented as a node and an edge is placed between two nodes whenever the two players share a partnership during an innings; the edge weight reflects the number of runs scored together. After building team‑specific BPNs from a large corpus of match data, the authors compute fundamental topological metrics: clustering coefficient (C), average degree (k̄), and average shortest‑path length (L). By comparing these values with those of Erdős‑Rényi random graphs that have identical node and edge counts, they demonstrate that all cricket BPNs exhibit the hallmark of a small‑world network—high clustering together with short global distances. Moreover, the degree‑assortativity coefficient is negative, indicating a disassortative structure where high‑degree (core) batsmen tend to connect with low‑degree (peripheral) batsmen, a pattern that mirrors the real‑world reliance of star players on a rotating pool of partners.

Centrality is examined through four complementary measures: degree, betweenness, closeness, and eigenvector centrality. The analysis reveals a striking decoupling between traditional performance statistics (batting average) and network centralities. For instance, a player with the highest degree is not necessarily the most prolific scorer, and a player with modest degree may possess the highest betweenness, acting as a crucial conduit for runs flow across the team. This divergence underscores that conventional batting metrics capture individual skill but ignore the collaborative dimension of partnership formation.

Community detection is performed using the Louvain algorithm, which optimizes modularity. The resulting partitions typically contain three to five communities per team, often aligning with batting order segments (openers, middle‑order, finishers) or situational sub‑groups that emerge during an innings. To characterize individual roles within and between communities, the authors adopt the Guimerà‑Nunes framework, calculating within‑module degree Z‑score and participation coefficient P. Nodes are then classified into four archetypes: provincial hubs (high Z, low P), connector hubs (high Z, high P), peripheral nodes (low Z, low P), and kinless nodes (low Z, high P). Notably, Sir Donald “Don” Bradman—widely regarded as the greatest batsman—falls into the peripheral category, indicating that despite his extraordinary personal statistics, he does not serve as a structural bridge in the partnership network.

The paper proceeds to a “node‑removal” experiment, systematically deleting each player from the BPN and measuring the impact on network efficiency (average inverse shortest‑path length) and the size of the giant component. Removal of connector hubs (high betweenness) leads to a pronounced increase in average path length and a sharp decline in efficiency, whereas removal of provincial hubs (high degree but low betweenness) has a comparatively modest effect. This sensitivity analysis quantifies the strategic importance of maintaining certain partnership links for overall team resilience.

Finally, the authors synthesize the multiple centrality scores into a composite “Performance Index” (PI) by assigning weights derived from regression against match outcomes. The PI correlates with, but is not identical to, batting average; it captures a player’s contribution to the team’s structural cohesion and run‑scoring dynamics. Players with modest averages but high PI are identified as undervalued assets whose partnership behavior enhances team performance.

In sum, the study demonstrates that cricket batting partnerships form small‑world, disassortative networks whose structural properties reveal insights beyond traditional statistics. By integrating community detection, role classification, centrality analysis, and robustness testing, the authors provide a multifaceted framework for evaluating player importance, informing selection decisions, and designing batting strategies. The methodology is readily transferable to other team sports where dyadic interactions drive collective outcomes, offering a powerful tool for sports analytics, coaching, and performance optimization.