First, the key terms.

Chance node
A point where uncertainty determines which outcome follows.
Rollback
Evaluating a decision tree from terminal outcomes back toward the start.
Break-even threshold
An input value at which modeled alternatives have equal value.

The idea, explained

Decision nodes identify actions under the decision-maker’s control; chance nodes represent uncertain events with appropriate probabilities. Terminal branches contain consequences such as net payoffs. Working backward averages chance outcomes and selects among feasible decisions under the chosen objective. Expected monetary value is useful under its assumptions but does not represent every decision-maker’s risk preferences or ethical constraints. Sensitivity and break-even analysis show when a recommendation changes as uncertain inputs vary.

STEP 1

Respect the information sequence

Place a test or observation before a later action only if it will actually be known then. Allowing an early decision to depend on a future outcome gives the model unrealistic foresight. A prediction is not perfect information.

STEP 2

Calculate branch values

At a chance node, use probabilities that sum to one across its possible outcomes. Include all relevant costs and avoid double counting them along paths. At a decision node, apply the stated preference or objective rather than averaging actions together.

STEP 3

Test the recommendation

Vary influential probabilities and payoffs, calculate switching thresholds and communicate downside outcomes. A positive expected value can hide a severe loss that the decision-maker cannot afford or is not entitled to impose on others.

Follow a worked example.

A fictional pilot costs $20. If successful, its benefit before that cost is $100; otherwise the benefit is zero. With success probability p, net expected value is 100p − 20. Doing nothing has a modeled payoff of zero.

Walk it through, one step at a time

  1. At p = 0.4, expected net value is $20. Individual net outcomes remain $80 or −$20, not a guaranteed $20.
  2. Set 100p − 20 = 0 to find the 0.2 break-even probability under these simplified assumptions.
  3. A proposed test is valuable only insofar as its information can improve the later choice enough to justify its cost and limitations. Recompute conditional decisions rather than assuming every extra test pays.

Where people get stuck

A tree with correct arithmetic can still be misleading if probabilities, feasible actions, timing or risk preferences are wrong.

Now try it yourself.

Pause before opening each answer. Explain your reasoning in your own words.

1. Check the core idea

Are a prediction tree and a decision-analysis tree the same method?

Reveal the answer & reasoning

No. One learns predictive splits from data; the other represents choices, chance outcomes and consequences for a decision.

2. Apply what you learned

If success probability is 0.1 in the example, what is expected net value?

Reveal the answer & reasoning

100 × 0.1 − 20 = −$10. Under an expected-money objective and these assumptions, doing nothing is preferable, but other objectives require explicit analysis.