An action is a choice available to a decision-maker. Unlike an uncertain state of the world, it represents something the decision-maker can control. In a Markov decision process, an action can affect both immediate reward and the probabilities of subsequent states. section 6.5.1 section 7.2.1
In decision analysis, acts are evaluated against possible states and their consequences. When probabilities and utilities are available, expected utility compares the desirability of alternatives; probability alone does not determine the choice. section 7.1 section 7.2
A useful set of actions should:
- Include every feasible alternative that could reasonably be preferred, including the status quo or taking no action. section 7.2.1
- Define alternatives distinctly and at comparable levels of detail, without adding immaterial variations that make analysis needlessly complex. section 7.2.1
- Distinguish final actions from experiments: an experiment gathers evidence that may change beliefs and inform a later choice. section 7.2.1
Treatment under diagnostic uncertainty
A physician may treat or not treat a patient. Each choice has different consequences depending on whether disease is present; posterior probabilities and utilities help compare the actions. section 7.1
Figure 7.3 Raiffa and Schlaifer’s decision-analytic cycle. Prior beliefs describe uncertainty about the possible states of the world. An experiment generates data that produces posterior beliefs, which inform the selection of an act. The resulting outcome is evaluated through utility and may provide the starting point for another decision.

The decision cycle shows how evidence from an experiment updates beliefs before an act is selected and its outcome evaluated. section 7.2
















