Don't Gamble

Don't Gamble

Occasionally, you may wish to temporarily remove a lottery from the sequence section of your model. This could be to shorten the run-time or to make it possible to add new decisions or chance events while keeping run-times approximately the same. It could also be a result of event sensitivity analysis identifying chance events that have little impact on the model.

One way to remove a lottery is to employ branch control to set the chance event to a particular outcome (See Control/Block Dialog). Variables that depend on the state of this event will be assigned the value associated with that state.

A different approach is to make the chance nodes for this event into don't gamble nodes in the decision tree. You do this by clicking on the Don't Gamble checkbox on the Tree Instance tab of the Node Definition dialog. For more on specifying Don't Gamble see Specifying Always / Don't Gamble. Any subsequent variables that depend on the state of this event will be assigned values calculated by taking the expected value across the lottery, where feasible. This calculation is performed each time the don't gamble specification occurs in the tree -- thus, the resulting value depends on the tree location and may differ between occurrences of the lottery.

If the don't gamble specification is applied, the tree's structure is changed. In the Policy Tree, an asterisk shows that the states have been replaced with a single state.

Changing all the lotteries in a tree to don't gamble nodes will not necessarily reduce the tree to a single path. You cannot reduce the number of paths to less than the number of decision alternatives. In addition, don't gamble specifications may not reduce a lottery to a single path. Asymmetries in the tree structure after the lottery can increase the number of branches remaining.

There are two things to remember about don't gamble nodes. First, using the don't gamble specification changes the structure of the tree during compile time. The decision analysis roll forward and rollback are then done normally. Second, a model with no don't gamble nodes and a model with all the lotteries changed to don't gamble will sometimes have the same expected value. The expected value will change only if there are non-linearities in your model.

Versions: DPL Professional, DPL Enterprise, DPL Portfolio

See Also

Specifying Always / Don't Gamble

Controlling the State of an Event

Event Tornado Diagram

Policy Trees