DPL 9 Help

Expected Value of Perfect Information/Control Expected Value of Perfect Information/Control

DPL can automatically calculate the expected value of perfect information and control (VOIC) for each chance node in a decision tree (except those which have Don't Gamble specifications or are controlled, as explained later in this topic).

Generating an Expected Value of Perfect Information/Control Chart

In the Home | Run group, check the VOIC box and click the Decision Analysis button. DPL will calculate the expected value of perfect information and the expected value of control for all chance nodes and display the results graphically. The value of perfect information for each variable is displayed as a red bar, and the value of control is displayed as a yellow bar.

Note: if endpoints are available you can also request Expected Value of Perfect Information Control when playing endpoints (Run | Play Endpoints).

Interpreting Expected Value of Perfect Information

A positive value of information means that knowing which state of the chance event will occur before making a decision changes the Policy Tree in some way to improve expected value. A zero value means that knowing the state has no impact on the Policy Tree. A negative value usually means that the decision was a minimize decision, and that there is value to knowing the state of the event.

Interpreting Expected Value of Control

A positive value of control means that you can improve the expected value if you can decide which branch of the chance node occurs. A zero value means that controlling the state has no impact on the Policy Tree. A negative value usually means that the decision is minimizing expected value, and that changing the chance node into a decision further reduces (improves) the expected value.

How DPL Calculates the Value of Perfect Information

In effect, DPL rearranges the decision tree and moves each chance node one at a time as far up the tree as feasible. For most nodes, this will be to the root of the tree. For some nodes, the node can only move part of the way to the root. For example, a node whose probabilities depend on a decision can only be moved up to just after the first instance of the conditioning decision event. After rearranging the tree, DPL re-runs the analysis. The difference between the original expected value (and certain equivalent) and the new expected value (and certain equivalent) is the expected value of information for the chance node. This information will also be written to the Session Log.

Nodes for which DPL can not calculate the value of information are written to the Session Log.

If your original tree has instances of the chance node marked Don't Gamble or which are controlled, DPL will leave them as is during the perfect information run. This means that it is possible for DPL to construct a value of information tree that you cannot, since you might need to have a node appear twice on the same path.

How DPL Calculates the Value of Control

In effect, DPL changes each chance node into a decision node and moves it to the head of the tree and re-runs the analysis. The difference between the original expected value (and certain equivalent) and the new expected value (and certain equivalent) is the value of control. As with the value of information, the value of control is written to the Session Log.

Nodes for which DPL can not calculate the value of control are written to the Session Log.

Note: DPL's proprietary algorithms allow it to perform all of the above calculations without actually re-arranging and re-running the tree multiple times.

Policy Levels to Save

In order to calculate the value of perfect information and control, the maximum setting must be chosen for the Number of policy levels to save within the Home | Run group on the ribbon. If this setting is anything but the maximum, DPL automatically change it to the maximum level if the expected value of perfect information and control option is enable.

When Can't DPL Calculate the Value of Perfect Information/Control?

Chance Node Probabilities are Conditioned by Decisions

If the probability distribution of a chance node depends on the state of a decision, DPL cannot move the chance node in front of the decision, and the therefore cannot calculate the value of information for the chance node (it can, however, calculate the value of control).

Can't Move the Chance Node In Front of a Decision

If there are no decision nodes between the chance node and the root of the tree, the node has no value of information.

Don't Gamble and Controlled Specifications

If all instances of a chance node have a Don't Gamble specification or are controlled in the decision tree, DPL cannot calculate a value of information or control for the node.

Versions: DPL Professional, DPL Enterprise, DPL Portfolio