In general, there are two major variables that contribute to emotion intensity. There are correlates for both principle-based emotions (e.g., pride), and goal-based emotions (e.g., satisfaction).
The first of these, simulation-event variables, measure those factors which are external to any particular agent and might simultaneously, and differentially, affect more than one agent at the same time. An example of this might be the number of parts damaged when a compressor overheats. For example, consider that Steve has the following goals:
High Level goal: Do not allow users to damage the (virtual) equipment. Goal: Do not fail to teach the user to check the oil level before starting the compressor. Simulation event meets these constraints: (a) compressor overheats and parts are damaged, and (b) history does not show an instance of the user checking the oil level.
In this case the number of compressor parts which are damaged might affect the modeled experience of failure on Steve's part. The greater the number of damaged parts, the greater the failure. Now consider that the user has the following goals, and principles (see Affective User Modeling section, below):
Goal: do not screw up any tasks that the tutor gives me to do. Principle: It is wrong to break the virtual equipment. Principle: Steve should teach me well enough that I do not end up breaking the equipment.
Simultaneously, the number of compressor parts which are damaged might also affect a user, Sarah, with respect to the above goal: she might be disappointed that she damaged the compressor; she might be ashamed of having damaged the compressor; she might be angry at Steve for failing to teach her how to avoid damaging the compressor. In all cases, the extent of damage to the compressor is likely to be directly proportional to the degree of intensity in the negatively valenced emotions.
By contrast, the stable disposition variables are those which determine the importance of a particular goal, or principle, to an agent. These values are internal, and changes in them do not affect any other agents. For example, one Steve might be very concerned about safety, and damage to the equipment. Whereas another Steve might be more concerned with exposing the user to explanations. For Safety-Steve the importance of the equipment being damaged might be quite high, whereas for Explanation-Steve the importance might be quite low (or even help him to achieve a goal through affording him a chance to give an explanation). Since these are internal variables, and help to give the agents their own dispositional personalities, changes in the importance values for one agent will not affect another.
Example ``errors-of-omission'' and their relative treatment with respect to an importance variable in relevant goals of Steve:
Dipstick is removed, not placed back in. Certainty is high, importance is low. User turns the alarm lights turned on, but does not turn them off, for N minutes. Thwarts a goal of not confusing other operators. Importance of the latter is medium high. User fails to open one of the valves, which is closed. Goal of not damaging the equipment is thwarted. Importance is high.