An investigation of the coauthor graph

The structure of coauthor graphs and the statistical validity of the associated author partitions are investigated as a function of productivity and collaborative thresholds and show that for “small” databases, statistically preferred partitions occur when all authors and coauthor pairs appear in the graph.

Abstract

The structure of coauthor graphs and the statistical validity of the associated author partitions are investigated as a function of productivity and collaborative thresholds. The productivity threshold determines the number of authors (points) in a coauthor graph, and the collaborative threshold determines the number of coauthor pairs (lines) in the graph. The statistical validity of author partitions is determined by the random-graph hypothesis. The results show that for “small” databases, statistically preferred partitions occur when all authors and coauthor pairs appear in the graph. For “large” databases, statistically preferred partitions occur when authors and coauthor pairs who publish only one article are excluded from the graph. Unlike other bibliometric relationships, the highly selective nature of the collaborative relationship produces a wide range of threshold values for which the associated partitions are statistically valid. It remains to be shown how the statistical validity of partitions is related to the empirical significance of the same partitions. © 1987 John Wiley & Sons, Inc.