Towards Validating Social Network Simulations - Advances in Social Simulation - page 1

Towards Validating Social Network Simulations

Syed Muhammad Ali Abbas 1 , Shah Jamal Alam 2 , and Bruce Edmonds 1

1 Centre for Policy Modelling, Manchester Metropolitan University
[e-mail], [e-mail]
2 School of Geosciences, University of Edinburgh
[e-mail]

Abstract. We consider the problem of finding suitable measures to validate si-
mulated networks as outcome of (agent-based) social simulations. A number of
techniques from computer science and social sciences are reviewed in this pa-
per, which tries to compare and 'fit' various simulated networks to the available
data by using network measures. We look at several social network analysis
measures but then turn our focus to techniques that not only consider the posi-
tion of the nodes but also their characteristics and their tendency to cluster with
other nodes in the network - subgroup identification. We discuss how static and
dynamic nature of networks may be compared. We conclude by urging a more
comprehensive, transparent and rigorous approach to comparing simulation-
generated networks against the available data.

Keywords: networks, validation, measures, ABM, underdetermination.

1 Introduction

Agent-based models (ABM) provide a method of representing local interactions using
the interactions between software agents [16]. Agents' interactions can lead to the
formation of ties with other agents and thus the generation of simulated networks.
Often, the social processes governing the agents' interactions influence the evolution
of such networks. In such cases, identifying the generative mechanisms is non-trivial,
but ABM can help do this, leading to an understanding of emerging network struc-
tures [14, 30]. Moreover, where the agents form several types of relations, multiple
overlapping networks result.
Validating agent-based simulations is hard. Their relative complexity means that it
is easier to 'fit' simple sets of empirical data so that a simple comparison of one or
two measures with empirical data is not sufficient to ensure that a simulation is an
adequate representation of what is being modeled. Different agent-based simulations
might well end up producing the same outputs if only compared in a few ways. The
more ways, in which a simulation is measured, especially if these are of very different
aspects, the greater the chance that any differences in simulation composition will be
revealed in a significant difference in the simulation outputs.
Many agent-based simulations directly generate explicit social networks, e.g., [1, 2,
4, 10, 14, 19, 20, 24, 35, 38]. There are many other simulations where relatively stable