| This dissertation addresses the problem of discovering and characterizing unknown elements in network systems. Klir (1985) provides a general definition of a system as "...a set of some things and a relation among the things" (p. 4). A system, where the 'things', i.e. nodes, are related through links is a network system (Klir, 1985). The nodes can represent a range of entities such as machines or people (Pearl, 2001; Wasserman & Faust, 1994). Likewise, links can represent abstract relationships such as causal influence or more visible ties such as roads (Pearl, 1988, pp. 50-51; Wasserman & Faust, 1994; Winston, 1994, p. 394). It is not uncommon to have incomplete knowledge of network systems due to either passive circumstances, e.g. limited resources to observe a network, active circumstances, e.g. intentional acts of concealment, or some combination of active and passive influences (McCormick & Owen, 2000, p. 175; National Research Council, 2005, pp. 7, 11). This research provides statistical and graph theoretic approaches for such situations, including those in which nodes are causally related (Geiger & Pearl, 1990, pp. 3, 10; Glymour, Scheines, Spirtes, & Kelly, 1987, pp. 75-86, 178-183; Murphy, 1998; Verma & Pearl, 1991, pp. 257, 260, 264-265). A related aspect of this research is accuracy assessment. It is possible an analyst could fail to detect a network element, or be aware of network elements, but incorrectly conclude the associated network system structure (Borgatti, Carley, & Krackhardt, 2006). The possibilities require assessment of the accuracy of the observed and conjectured network systems, and this research provides a means to do so (Cavallo & Klir, 1979, p. 143; Kelly, 1957, p. 968). |
