Error checking: Comparing results of alternative models

There are often several ways that one could construct a Monte Carlo model to tackle the same problem. Each method should give you the same answer, of course. So, if you are unsure about one way of manipulating distributions then try it another (perhaps less efficient way) and see if the answers are the same. In the examples we give in this help file we often show two or more ways of arriving at the same answer.

The more difficult area is where you may feel that there are two or more completely different stochastic processes that could explain the problem a hand. Ideally, one would like to be able to construct both models and see whether they come up with similar answers. But what do we mean by similar? In fact, from a decision analysis point of view we don't actually mean that they come up with the same numbers or distributions: we mean that if presented with either result, the decision-maker would make the same decision. This topic is discussed in decision-sensitivity.

We nearly always will not have the luxury of being able to model two or more different approaches to the same problem because of time and resource constraints. If you are going to have to put all your efforts into one model, try to make sure that your peers agree with your approach, and that the decision-maker will be comfortable with making a decision based on the model's assumptions. The decision-maker could prefer you to construct a model that may not be the most likely explanation for your problem, but that offers the most conservative guidance for managing it.

Error checking: Comparing results of alternative models

Navigation