As noted in Chapter 6, the degree of model complexity employed in analysis is a key aspect of designing effective RMPs and other management science intervention processes. An interesting survey of failures and successes of quantitative methods in management by Tilanus (1985) supports the widely held view that successful modelling requires approaches that are 'simple', flexible, easily understood, appropriate to the situation, and able to cope with low-quality data. A detailed discussion of the effectiveness of 'simple' analysis is provided by Ward (1989), employing a 'constructive simplicity' concept that describes the form and level of detail in a model. Chapman and Ward (2002) further develop this 'constructive simplicity' concept and its application to various aspects of project uncertainty. Usual arguments for constructive simplicity focus on model-building considerations such as model clarity, flexibility, and convenience, but constructively simple models can also provide an efficient way of learning about decision situations. The basic idea is to start with effective, simple analysis that is then elaborated in useful directions as understanding develops. A key theme here is that additional model complexity should be introduced only if it is useful. In this respect, a constructively simple approach is fundamentally different from a simplistic approach that involves adopting simplicity naively and precludes any further model development.
Note that this approach to modelling in a particular instance is the reverse of the overall approach just discussed. This reversing of directions is not inconsistent. The craft skills required to use the process effectively in a particular instance are developed within the overall approach outlined earlier in this chapter. Usually additional model complexity proves useful (constructive) because it:
• makes estimation easier;
• allows the integration of estimation expertise involving different people or databases;
• clarifies what estimates measure and what they do not measure;
• provides richer insights about decision alternatives;
• provides more confidence that issues are properly understood.
As indicated earlier, the simplest formal quantitative model of uncertainty for project duration analysis is the basic PERT (Program Evaluation and Review Technique) model; the most complex the authors are aware of in a source-response-impact dimension is the basic SCERT model (Chapman, 1979). An earlier publication (Chapman et al., 1985b) addresses making choices along the PERT-SCERT axis, and subsequent publications have discussed these choices in more detail (e.g., Chapman, 1990). Other modelling complexity dimensions include systems dynamics models to capture feedback and feedforward loops (Forrester, 1961; Richardson and Pugh, 1981; Senge, 1990), cognitive mapping to capture other interdependencies in a qualitative manner (Eden, 1988), and more general 'soft' methods (Rosenhead, 1989; Checkland and Scholes, 1990), as mentioned in Chapter 8. Such modelling complexity dimensions are worth exploring by the reader with a view to more effective modelling of uncertainty, and further approaches may prove worthy of development.
The more modelling choices become available the more difficult it is to make the most appropriate choices, unless we clearly understand what each model feature costs and what benefits it is likely to yield. Only some very general guidelines can be offered here in terms of where to start with basic model development:
• make sure all key issues are identified and associated with appropriate responses, whether or not formal quantitative modelling is feasible;
• don't attempt implementation or interpretation of quantitative analysis unless you understand prior, underlying qualitative analysis;
• if project activities involve repetitive component processes, consider the use of Markov process models to show the combined effect of these processes;
• if feedback or feedforward loops are important, consider the use of system dynamics models.
A 'constructively simple' approach to estimating parameters for any given model structure will facilitate starting with a very simple but 'conservative' (safe) approach to filtering out what matters and what does not, in order to spend the available analysis effort as effectively as possible. Two extended examples illustrate what is involved in the next two sections. Both are heavily based on recent papers: Chapman and Ward (2000) in the case of the next section and Chapman and Ward (2003) in the following section. Both examples are treated jointly and somewhat differently in Chapman and Ward (2002, chap. 4).
Was this article helpful?