Article Preview
Top1. Introduction
System designers and analysts often begin their work by developing and using graphical representations of relevant features of the domain under examination (Burton-Jones and Meso, 2006). These representations are called conceptual models (Wand and Weber, 2002). They play a significant role in the early detection and correction of systems development errors and help analysts to better communicate with stakeholders (Moody, 2005; Saghafi and Wand, 2014).
Much research on conceptual models and the grammars with which they are created has been conducted over the past decades, often using ontological analysis (Wand and Weber, 1990, 1993, 1995). Ontological analysis allows researchers to suggest how grammars for conceptual modeling might be modified to be ontologically sound and how well developed models that correspond to ontological guidelines are understood. There is a strong track record of studies showing empirical support for these guidelines (Saghafi and Wand, 2014). Yet noticeably, this rich research tradition has focused almost entirely on the evaluation of single grammars (like UML, ERD, BPMN and so forth) or single models (such as those that are ontologically clear versus unclear).
One important insight from this stream of research is that no single available grammar is ontologically complete (e.g., Irwin and Turk, 2005; Opdahl and Henderson-Sellers, 2002; Recker, Rosemann, Indulska, and Green, 2009; Wand and Weber, 1993; Weber, 1996). This situation implies that users will never be able to create a single model that fully represents all relevant aspects of the real-world phenomena they wish to have represented.
This situation is not necessarily problematic; in fact, incompleteness seems to be a design choice for many popular grammars: UML, for example, provides fourteen different grammars to describe structure, behavior, and interactions of a system from a variety of perspectives, each of which is, by necessity, incomplete (Rumbaugh, Jacobson, and Booch, 2004). Other longstanding methodologies, such as Multiview (Avison and Wood-Harper, 1986), have promoted the use of multiple models with different perspectives for close to thirty years.
In this paper, we explore two questions that follow from this situation: how do analysts and designers deal with the fact that any model they have available is not a complete representation? How do they select from a variety of possibly available models given that each of them will offer some representation but never a complete one? The proposition that we put forward is that they use multiple models in combination such that the completeness of the representation of their relevant real-world phenomena can be maximized.
We are not the first to make this proposition. Theoretically, this proposition has been explored, firstly, by Weber (1997) and Green (1996) who suggested two principles, maximum ontological completeness and minimal ontological overlap, to explain why designers might select different grammars for conceptual modeling. More recently, Recker (2014) suggested theoretical arguments in a theory of faithful use of conceptual model combinations. Yet, what remains notably absent is empirical knowledge about how practitioners work with multiple models. We take this step in this paper and explore two broad research questions: