BIASED NET THEORY - APPROXIMATIONS, SIMULATIONS AND OBSERVATIONS

Rapoport's biased net theory is one of the earliest stochastic approaches to the representation of networks. It remains an attractive view because of its constructionist image that networks are built up from local events of connection that have both biased and random components. Difficulties in analysis abound, however, and derivations must appeal to approximation arguments. Research reported here examines two algorithms for generating networks in line with biased net formulations. Both attempt to generate nets subject to specific bias factors. The first attempts to match a target outdegree distribution while the second only seeks to match a target total number of relations. The second approach is more successful on statistical grounds. It also fits better data from two networks presented by Coleman (1964). Results call into question the soundness of the approximation arguments that have been central to previous derivations from biased net theory. © 1990.