Decision Analysis Under Uncertainty

Definition

Decision analysis formalizes a choice made before an uncertain event resolves by separating three elements: the decision alternatives under the decision-maker's control, the states of nature (mutually exclusive possible outcomes of the uncertain event), and the payoff associated with each alternative/state combination. These are laid out in a payoff table or, for sequential decisions, a decision tree — "square" decision nodes where a choice is made, "circle" chance nodes where nature decides — and the standard resolution rule is expected value: at each chance node, weight each payoff by its probability and fold the tree back to find the alternative with the best expected outcome.

In the Book

Chapter 13 builds the framework around the Pittsburgh Development Corporation (PDC), a company deciding whether to build a small (30-unit), medium (60-unit), or large (90-unit) luxury condominium complex before knowing whether demand will be strong or weak. The book constructs the payoff table in millions of dollars for all three sizes against both demand states (e.g., a large complex nets $20 million under strong demand but loses $9 million under weak demand), draws the corresponding decision tree, and computes the expected value at each chance node to recommend a specific complex size and expected profit. It then extends the same PDC problem to show how the recommendation changes once PDC can pay for sample information (e.g., a market research report) before committing, folding the value of that information into the decision tree as well.

Why It Matters

Most consequential decisions are made before the relevant uncertainty resolves — decision analysis gives a repeatable structure for making that bet explicit and comparable rather than left to gut feel about which scenario "feels" more likely. Laying alternatives against states of nature in a payoff table forces the decision-maker to enumerate what could actually happen and what it's worth, and the expected-value calculation gives a defensible way to rank options whose outcomes only differ once uncertainty is resolved — the same structure underlying any go/no-go call made under incomplete information.