Expert Window | Vose Software

Expert Window

See also: Modeling expert opinion

A common problem in risk analysis is to elicit reliable, realistic estimates of uncertain quantities in your model. ModelRisk offers two interfaces to help achieve this: the Expert window described here for single uncertain variables; and the SME Time Series window for uncertain time series forecasts.

The Expert Window is accessed by clicking the Expert icon under the Tools group in the ModelRisk toolbar. It has two tabs:

Selector – for defining a distribution from a number of statistical properties; and

Shaper – for ‘drawing’ your own distribution

Selector tab

The Selector tab allows you to select a number of statistical parameters with which you wish to define a distribution. As you select each statistical parameter using the check box, the parameters that are no longer valid gray out. The process goes as follows:

1. Select the parameters you wish to use by checking each applicable box;

2. Enter the values for these parameters in the adjacent fields; and

3. Click the ‘Calculate list’ button to see the options that are available.

In the following image, the minimum, mode and maximum have been used, and ModelRisk has provided a list of potential candidates.

On the left a ‘Distribution list’ provides all available options. Un-ticking any distribution will remove it from the graphs and statistics tables..

In the middle are relative and cumulative plots of each distribution within the list. Clicking on any item in the ‘Distribution list’ will bring that distribution to the front of the chart.

On the right is a table of statistics to allow you to compare, for example, the means or percentiles of the available options.

In certain cases, for example here where one has defined the minimum, mode and maximum, some distributions have an extra degree of control. For example, the following image shows the Kumaraswamy4 distribution selected with a red dot that can be slid up and down within the bar range to assign more or less peakedness whilst respecting the defined parameters:

Once an appropriate distribution has been selected, the user can click the ‘Insert in Worksheet’ button and select to enter a distribution random sample or object function for the selected distribution.

ModelRisk will then enter the function into Excel, and add an Excel Comment to display just how this distribution has been defined.

This is useful in situations where, for example, one has defined a mean, standard deviation and skewness and found the most appealing option to be an unusual distribution like a Bradford or Fatigue.

Shaper tab

The Shaper tab allows you to draw your own distribution:

To use this tool, proceed as follows:

1. Define the minimum and maximum values to give a range for ModelRisk to plot. Here we’ll set it to 5 and 30:

2. Choose whether to edit the relative or cumulative plots. In general, it is more intuitive to edit the relative plot.

3. Double-click within the selected graph to create some extra points, and then use the mouse to move them around and create the shape you are looking for.

4. Take note of the statistics in the right pane and the graph you are not editing to make sure that they continue to make sense.

Once you are satisfied with the distribution, select the ‘Insert in Worksheet’ button and select to enter a distribution random sample or object function.

 

ModelRisk will then enter the distribution you drew as a Relative distribution if you used the relative plot, or a CumulA distribution if you edited the cumulative plot.

Note: ModelRisk interpolates between the points you place in the plot with straight lines. Thus you should place more points in areas of greatest curvature to get a more accurate reflection of your estimate – but you don’t need to add too many points because, after all, subjective estimates are not that precise by their nature. One result of using straight line interpolation is that if you edit the cumulative plot, the relative plot will become a histogram, and the relationship may not be that intuitive to a subject matter expert, possibly providing a distraction from the elicitation exercise:

 

 

Navigation