Explaining a models assumptions | Vose Software

# Explaining a models assumptions

The key to gaining acceptance to a model's results is very often the acceptance of the model's structure and assumptions. We recommend that you are very explicit about your assumptions, and make a summary of them in a prominent place in the report, rather than just have them scattered through the report in the explanation of each model component.

A risk analysis model will often have a fairly complex structure and the analyst needs to find ways of explaining the model that can quickly be checked. The first step is usually to draw up a schematic diagram of the structure of the model. The type of schematic diagram will obviously depend on the problem being modelled: GANTT charts, site plans with phases, work breakdown structure, flow diagrams, event trees, etc. - any pictorial representation that conveys the required information.

The next step is to show the key quantitative assumptions that are made for the model's variables.

### Distribution parameters

Using the parameters of a distribution to explain how a model variable has been characterised will often be the most informative when explaining a model's logic. We tend to use tables of formulae for more technical models where there are a lot of parametric distributions and probability equations, because the logic is apparent from the relationship between a distribution's parameters and other variables. For non-parametric distributions, which are generally used to model expert opinion, or to represent a data set, a thumbnail sketch helps the reader most.

Influence diagrams are excellent for showing the flow of the logic and inter-relationships between model components, but not the mathematics underlying the links:

### Graphical illustrations of quantitative assumptions

These are particularly useful when non-parametric distributions have been used. For example, a sketch of a VoseRelative or VoseCumulA distribution will be a lot more informative than noting its parameters values. Sketches are also very good when you want to explain partial model results. For example, summary plots are useful for demonstrating the numbers that come out of what might be a quite complex time series model. Scatter plots are useful for giving an overview of what might be a very complicated correlation structure between two or more variables.