MS&E 252 Handout #20

Decision Analysis I November 16
. 2004

Page 1 oI 2 Decision Diagrams

Decision Diagrams
Decision Diagram: A conceptual representation oI a decision process used to provide the basic structure Ior a
decision problem. Includes uncertainties (like a relevance diagram). a value node. and at least one decision node.

Arrows in decision and relevance diagrams: The presence oI an arrow Irom uncertainty A to B or Irom B to A in a
relevance or decision diagram indicates that A and B might` be mutually relevant. given the state oI inIormation
represented by the other arrows into A and B. The absence oI an arrow between A and B asserts that A and B are
not relevant to each other. given the state oI inIormation represented by the other arrows into A and B. The
direction oI the arrow (whether Irom A to B or Irom B to A) indicates the order we would like to think oI the tree
as being in (might be diIIerent Ior assessment than Ior inIerence). The stronger statement is made by the exclusion
oI arrows.

Nodes in Decision Diagrams: the Iollowing table provides the diIIerent types oI nodes and their meanings


A decision node may have an underlying structure oI a continuous decision variable
or a discrete set oI alternatives.


An uncertainty node represents a distinction in the mind oI the author that is
uncertain. An uncertainty node can represent a continuous or discrete random
variable or a set oI events. In general. we can think oI an uncertainty node as
characterizing a kind oI distinction within which there are two or more degrees or a
continuous distribution. We characterize the author`s uncertainty with a probability
assessed by the author or an expert designated by the author.


A deterministic node means that iI the inputs to this distinction are speciIied. then
the value oI this distinction can be determined |i.e. deterministically|. We can think
oI a deterministic node as a special case oI an uncertain node where we have
asserted graphically that the conditional probability oI its value has been set to one.
Deterministic nodes will usually represent mathematical relationships. such as proIit
equals revenues minus costs.


A value node is designated by the author to be the quantity whose certain
equivalent is to be optimized by the decisions. Only one such node can be
designated. The value node is thus a special deterministic node; it is usually drawn
as a hexagon. octagon. or diamond.

MS&E 252 Handout #20
Decision Analysis I November 16
. 2004

Page 2 oI 2 Decision Diagrams

Arrows in Decision Diagrams: (also called arcs)
The set oI nodes with arrows leading into a particular node is that node`s parents. The Iollowing table shows the
diIIerent types oI arcs that can be Iound and their meanings.

Figure 1:
D1 D2

Figure 2:

An information arrow leads into a decision node Irom a decision
node (Figure 1) or Irom an uncertainty node (Figure 2).
In Figure 1. the inIormation arrow indicates that decision D2
is made with the knowledge oI the decision made in D1.
Since a decision-maker is assumed not to Iorget previous
decisions she has made. every decision node has an
inIormation arrow Irom each previously made decision.
ThereIore. inIormation arrows between decision nodes have
been called 'no-Iorgetting arrows.¨
In Figure 2. the inIormation arrow indicates that the outcome oI
uncertainty A is known by the time the decision-maker makes
decision D.

An influence arrow is an arrow Irom a decision node to an
uncertainty node. it indicates that the probability distribution
across the degrees oI uncertainty B is conditional on which
alternative is selected in decision D. Deterministic nodes can be
used (as report nodes) to avoid decisions inIluencing uncertainty.
especially when modeling inIormation gathering

An arrow between two uncertainty nodes is called a relevance
arrow. Recall that a relevance arrow is a placeholder Ior
relevance. It indicates that the probability distribution across the
degrees oI uncertainty B is conditioned on each oI the degrees oI
uncertainty A.
We can reverse the direction oI a relevance arrow between two
uncertainties only iI they have the same set oI parents (i.e.
conditioned on the same state oI inIormation).


An arrow Irom an uncertainty node to a deterministic node is
called a functional arrow. Here. Ior instance. once we know the
value oI uncertainty C and that oI uncertainty D. we can determine
the value oI P; in this example. P is simply the product oI C and

An arrow Irom a decision node or an uncertainty node to the value
node is called a direct value arrow. Note that a decision-maker
seeks to optimize the value node by selecting an alternative in
each decision node. So. in general. we should have an arrow Irom
each decision node to the value node.