Causal Models

MODIST uses causal models to describe software engineering processes and the key factors that influence their successful implementation.

These models are represented by graphical networks:

• the nodes represent the uncertain variables in the processes
• the arcs represent causal or relevance relationships among these variables.

At any one time, the value of a node is represented by the set of probabilities for each of its possible states. Many nodes have parent nodes, so they have associated with them conditional probabilities for all combinations of parent states. The probabilities are held in Node Probability Tables (NPTs).

Other models have focused on risk, or on quality, or on estimating effort and time. Typically they have been regression-based models. They help in identifying an optimum approach, but not in reasoning about alternatives as is often needed for managerial decision making. MODIST tackles these issues using an integrated and configurable set of causal models that lend far greater visibility to cause and effect.

MODIST recognises that project success often depends on 'soft' factors - notably human influences - that are difficult to measure objectively. Causal models cope with such factors far more readily than do statistical models.

The calculations for executing the causal models make extensive use of Bayes' Theorem of probabilities, so we call the models Bayesian Network (or BN) models.