System Dynamics Applications / 2 / 2.1 / 2.1.1

2.1.1 ‘Hard’ And ‘Soft’ Variables

A hard variable is one which has attributes and relationships with other variables in a problem space to which physical laws apply. In the case of hard variables the governing business rules we build into models are readily formulated using numerical values and algebraic operators. Hard variables are readily quantifiable, and quantification can be verified. Soft variables are a class of variables which includes a sub-class known as intangibles. We generally associate soft variables with attributes of humans in human activity systems. They assist us in describing the complexity of human affairs.

Soft variables are often an inescapable part of our dynamic hypotheses. We build dynamic hypotheses that include soft variables in an attempt to explain (in part, at least) how systemic behaviour of human activity systems is produced. Most frequently we can measure the effects produced by variations in the state of soft variables, without directly measuring the soft variables themselves. The effects produced by the extant state (or changes in state) of soft variables are generally measurable, but measuring soft variables themselves with the intent of including them in models can be particularly problematic.

Motivation would be considered as a soft variable. When building models we would be very interested in the effects of motivation. We might choose to create in our model a measure of the change in the level of production in a factory (or reduction in amount of rework) stemming from the perceived (or actual) level of motivation of the workforce.

Corporate knowledge or human intellectual capital (what people know and how they apply their knowledge to be creative) are examples of intangibles.

The only thing we can say with certainty about models of real-world problems—that is, human activity systems that do not include the effects (influences) of soft variables—is that they will be wrong (Forrester, 1961: 57; Sterman, 2002: 523)

In this book, the term model refers to a device to represent a complete problem or larger portions of a problem. Therefore, when the term model is used it includes sector, being a compartment of a problem space, and module being a small (albeit specifically focused) model.

These modules, sectors and models are intellectual devices, or transient conceptual objects, which we use to describe, communicate ideas about, and represent parts of the real world around us. They are necessary simplifications of the real world produced for a specific purpose, such as providing the basis of analysis and informing our understanding of past behaviour and what future behaviour might look like.

The main reason for building these models (in a general sense, the term model includes model, sector, and module) is to provide, through simulation in a defined synthetic environment, useful sets of data describing past or expected future behaviour and exciting opportunities to learn. The modelling and simulation activities, and the knowledge we gain from them, present new opportunities to explore alternative futures and for developing the futures we desire.

Even though they may include consideration of the influences of (or effects produced by) soft variables, each model is ultimately a hard (quantitative) representation of a particular problem. This can be both an advantage and a disadvantage. The advantage is that the model can be closely and publicly scrutinised and its performance should be reliable and repeatable. The disadvantage is that we might be tempted to interpret this quantitative model as the only possible representation of the real world.

There may well be other perspectives and alternative representations. This is an important point which we should never forget. Models are built using component parts glued together with specific business rules, governing rules developed as partial interpretations of our dynamic hypotheses. In turn, these are developed from interpretation of real-world problems and (ultimately) expressed as precise sets of mathematical relationships.

Williams (2002: 33) explains that when we build such models we are not limiting ourselves to mathematical models, and indeed many of our quantitative mathematical models will be developed from qualitative models expressed in “English” terms, but in formalised formats so that the concepts are made as consistent, unambiguous and precise as we can make them—with particular attention being paid to causal relationships and relationship structures, which define the problem faced.