The Order of Stages

The order of stages is an important issue in structuring the decision problem. This sets the sequence of events or a time horizon and determines when a decision has to be made and when a chance stage has its uncertainty resolved. If a decision stage precedes another decision stage in the stages order, the decision to the right is made after the decision to the left. Moreover, the choice made in the first decision is remembered by the decision maker when he or she makes the second decision. Any chance stages that occur to the left of a decision stage have their uncertainty resolved before the decision is made. In other words, the decision maker knows what actually happened when he or she makes the decision. However, the order of two chance stages is fairly arbitrary if there are no other decision stages between them. For example, you can change the order of stages 'Cost' and 'Oil_Deposit' in the oil wildcatter’s problem without affecting the results.

PROC DTREE determines the order (from left to right) of all stages specified in the STAGEIN= data set. The ordering is based on the rule that if stage A is the successor of an outcome of stage B, then stage A should occur to the right of (or after) stage B. With the MOVE statement, you can change this order. The MOVE statement is very useful in determining the value (benefit or penalty) of postponing or hurrying a decision. In particular, the value of perfect information about an uncertainty can be determined by moving the corresponding chance stage to the beginning. However, as mentioned in early sections, the results may be misleading if you use the MOVE statement without care. See the section Input Data Sets for an example.

Suggestions for preventing misleading results are as follows:

  • Using the SAVE statement, always save the original structure before making any changes.

  • Use the TREEPLOT statement to display the complete decision tree and check all details after you change the order.