Sensitivity analysis not always working when uncertainty distribution is defined as complement

[Miguel Ángel Artaso Landa]

When a variable has an uncertainty with the distribution defined as the complement of the other, the values obtained in the sampled network during the sensitivity analysis under certain scenarios are not the correct ones. With the network attached to the issue this behaviour is reproducible.

The steps to do so are:

1. Open the attached network in Open Markov

2. Open the sensitivity analysis dialog (Tools -> Sensitivity analysis)

3. Configure the analysis desired as follows:

• Analysis type: Plot (one-way analysis)
• Points per parameter: 2 (the value can be any, but a lower value is better in this case)
• Parameter: prevalence (the only possible one in this network).
• Horizontal axis variation: user defined interval. from 0 to 1.
• Scope: global

It seems that the configuration that may be causing the issue in the confluence of the definition of the 'distribution' of the uncertainty in a variable as 'complement' and the 'horizontal axis variation' as 'user defined interval'.