Utilities in Influence diagrams

[nikkne]

Hi all,
I have a naive question

I have an influence diagram matching a real-world experiment. This diagram has only one utility node, and in the experiment I can observe the value corresponding to this utility node. Since the utility node maps utility value to the values in the parents node in the odometer-like style, what to do when the same combination of values in the parents have two different utilities? Should I use this node as utility node at all?

Cheers,
Nikola

[marek [BayesFusion]]

Hi all,
I have a naive question

I have an influence diagram matching a real-world experiment. This diagram has only one utility node, and in the experiment I can observe the value corresponding to this utility node. Since the utility node maps utility value to the values in the parents node in the odometer-like style, what to do when the same combination of values in the parents have two different utilities? Should I use this node as utility node at all?

Cheers,
Nikola

I believe that this is impossible (i.e., that the same combination of values of parents has two different utility values). It would mean that you value the same thing at different levels, while everything else is equal. Theoretically quite shaky, although I admit that possible with a human. It is possible that I misunderstood you. If you send me your diagram, I will be glad to help.
Cheers,

Marek