Simple egocentric measures are often used to give an indication of node importance in networks, for example counting the number of arcs at a node. However these (and even less obviously egocentric measures like centrality and the identification of cliques) are potentially fallible since any such measure is only a proxy for the structural properties of the network as a whole. The effectiveness of such proxies relies on assumptions about the generative processes which give rise to particular networks and these assumptions are not (and possibly cannot be) tested within the normal framework of Social Network Analysis (SNA). In case studies of real social networks it is often difficult to get enough data to evaluate how faithfully the standard network measures do characterise the network. Given these empirical and methodological difficulties, therefore, we use a set of agent-based simulations as case studies to investigate some of the circumstances in which simple measures fail to characterise key aspects of social networks. Using this approach we start to map out situations where simple measures might not be trusted. In such circumstances, we suggest that an alternative approach (in which agent-based simulations are used as a sort of dynamic description capturing qualitative facts about the target domain) might be more productive.