- Atlas
- Engineering
- Decision And Risk Analysis
Decision And Risk Analysis
Decision and risk analysis is the part of systems engineering that helps people choose what to build and how to build it when the consequences are uncertain and the stakes are high.
10 frameworks, from 1944 to today · Systems Engineering · Engineering
- Expected utility hypothesis 1944-Present
- Monte Carlo method 1945-Present
- Systems Analysis 1948-1980
- Subjective Expected Utility 1954-Present
- Decision Analysis 1960-Present
- Multiple-criteria decision analysis 1970-Present
- Probabilistic Risk Assessment 1975-Present
- Risk Management 1980-Present
- Structured Expert Judgment 1985-Present
- Robust Decision Making 1995-Present
Overview
Decision and risk analysis is the part of systems engineering that helps people choose what to build and how to build it when the consequences are uncertain and the stakes are high. Its practitioners ask two linked questions: what should we do, given what we value? and how wrong could we be about what will happen? The answers are not guesses but structured arguments—models of alternatives, probabilities, and preferences that can be examined, challenged, and defended. The field matters because large engineering decisions (which reactor design, which aircraft fleet, which defense system) are made once, cost enormously, and cannot be tested by trial and error. Decision and risk analysis grew up around exactly those decisions, and its history is a story of borrowing ideas from economics, statistics, and operations research and forging them into practical tools for engineers and managers.
Expected Utility and the Foundations of Rational Choice
The intellectual root is the expected utility theory of the mid-twentieth century12. In 1944, John von Neumann and Oskar Morgenstern, working on game theory, showed that a decision-maker whose preferences satisfy a few consistency axioms—such as preferring more to less and being indifferent between a gamble and a certainty of equal value—can be represented as maximizing expected utility: the average of the utilities (subjective values) of outcomes, each weighted by its probability13. Utility here means a numerical measure of how much a person values an outcome, not money3. The idea was powerful because it turned vague "what's best" questions into arithmetic.
The theory's weakness was that it assumed probabilities were known4. In 1954, Leonard Savage's The Foundations of Statistics extended the framework to subjective probability: a rational person's beliefs, too, could be represented as probabilities consistent with their choices. This made the theory usable for one-of-a-kind decisions where no frequency data exist—precisely the situation of a new weapons system or a first-of-a-kind plant. The economist Howard Raiffa, in Decision Analysis (1968) and later with Robert Schlaifer, carried these ideas into business and engineering practice, coining much of the vocabulary still used.
Decision Analysis as a Practical Craft
The 1960s turned theory into method. The central tool was the decision tree: a diagram branching at each decision and each uncertain event, with probabilities on the chance branches and values at the ends, rolled back to compute expected values1. Its companion was the influence diagram, a compact graphical model of which variables affect which1. The Stanford Research Institute and later Strategic Decisions Group became centers where these methods were taught and applied to corporate and government problems.
A crucial refinement was value of information5. If a decision hinges on an uncertain quantity, one can compute how much it would be worth to learn that quantity before deciding—the expected improvement in outcome from perfect or imperfect information. This told engineers which tests, surveys, or studies were worth funding and which were not. The same period produced sensitivity analysis: systematically varying assumptions to see which ones actually drive the answer, so that effort goes into the inputs that matter. Decision analysis thus became less a calculator than a discipline of structuring a problem—separating decisions from uncertainties, clarifying objectives, and exposing where the real leverage lies.
Systems Analysis and the Policy Era
In parallel, the postwar policy world produced systems analysis: the application of quantitative modeling to large public choices. The RAND Corporation, founded in 1948, was its home. Analysts there studied strategic bombing, force posture, and later urban and environmental problems, treating a policy question as a system to be modeled, its alternatives costed, and its uncertainties examined. The approach spread into government—the U.S. Department of Defense's Planning-Programming-Budgeting System in the 1960s attempted to tie budgets to explicit objectives and analysis—and into civil agencies.
Systems analysis and decision analysis overlapped but differed in emphasis: systems analysis was broader and often cost-effectiveness oriented, asking which alternative delivers a given capability most cheaply; decision analysis was more explicitly about preferences and uncertainty. Both taught the same lesson: that a well-structured model, even an imperfect one, disciplines debate and reveals hidden assumptions. The era also bred skepticism, as critics noted that models could encode the analyst's assumptions and that quantification could crowd out judgment. That tension—between formal analysis and the limits of formalization—has run through the field ever since.
Probabilistic Risk Assessment
The most consequential engineering strand began with nuclear power. In 1975, the U.S. Nuclear Regulatory Commission published WASH-1400, the Reactor Safety Study, led by Norman Rasmussen of MIT. It was the first large-scale probabilistic risk assessment (PRA): a systematic estimate of the likelihood and consequences of accidents, built from event trees (branching sequences of what happens after an initiating event) and fault trees (logic diagrams of how component failures combine to cause a failure). The study estimated that the risk of a severe reactor accident was far lower than many had assumed—a conclusion that was controversial and, in the light of later accidents, contested, but the method itself became standard.
PRA's contribution was to make risk a computed quantity with visible structure: which failures matter, how redundant systems reduce risk, where common causes could defeat redundancy. It spread to chemical plants, space systems, and offshore oil. The 1979 Three Mile Island accident and later disasters both tested and drove refinement of the method, and risk management—the organizational process of identifying, assessing, mitigating, and monitoring risk—grew up around it, codified in standards and regulatory frameworks.
Monte Carlo Simulation and Sensitivity Analysis
Underlying much of this is Monte Carlo simulation, named for the casino and developed at Los Alamos in the 1940s by Stanislaw Ulam and John von Neumann. The method estimates the behavior of a complex system by drawing many random samples of its uncertain inputs and computing the outcome for each, so that a distribution of results emerges. Where analytical solutions are impossible—as in most real engineering systems with many interacting uncertainties—simulation provides the answer. Combined with sensitivity analysis, it lets analysts see not just the expected outcome but the spread, the tails, and which inputs dominate the risk. Cheap computing from the 1980s onward made simulation routine, and it is now the workhorse behind PRA, cost-risk models, and schedule-risk analysis.
Structured Expert Judgment
Real systems often lack data, especially for rare events. Structured expert judgment elicitation developed as a disciplined way to convert expert opinion into probability distributions. The approach, associated especially with Roger Cooke's work from the 1980s and 1990s, uses carefully designed questionnaires and calibration: experts answer questions with known answers (so-called seed questions), and their performance is used to weight their judgments. This addresses the obvious objection that expert opinion is subjective and possibly biased, by making the subjectivity measurable and combining multiple experts in a defensible way. It is now standard in risk assessments where historical data are thin.
Multi-Criteria Decision Analysis
Not every decision reduces to a single monetary or risk measure. Multi-criteria decision analysis (MCDA) handles choices among alternatives judged on several incommensurable objectives—cost, safety, environmental impact, schedule. Methods such as the analytic hierarchy process, developed by Thomas Saaty in the 1970s, and multi-attribute utility theory, developed earlier by Keeney and Raiffa in their 1976 Decisions with Multiple Objectives, provide ways to elicit weights and scores and combine them transparently. The value is procedural as much as numerical: the method forces stakeholders to state what they care about and how much, making trade-offs explicit rather than hidden.
Robust Decision Making and the Turn to Deep Uncertainty
By the 1990s and 2000s, a further problem came into focus: deep uncertainty, where the parties cannot agree on the probabilities, the models, or even the objectives. Traditional expected-utility analysis assumes these are known or estimable. Robust decision making, developed largely at RAND by Robert Lempert and colleagues, instead seeks strategies that perform acceptably across a wide range of plausible futures, identified by computer search over many scenarios. Rather than optimizing for a best estimate, it asks which options are least vulnerable to being wrong. This approach has been applied to water resources, climate adaptation, and long-lived infrastructure, and it represents the field's current frontier: how to decide well when the future cannot be pinned down.
The Present
Today the strands coexist and interlock. Expected utility and decision trees remain the conceptual core; PRA and Monte Carlo simulation supply the risk estimates; expert elicitation fills data gaps; MCDA structures multi-objective choices; and robust decision making addresses irreducible uncertainty. The tools have migrated from nuclear and defense into finance, medicine, supply chains, and climate policy, and computing has made simulation and scenario search routine. The open questions are the old ones sharpened: how to validate models of rare events that have never happened, how to combine formal analysis with the judgment of experienced people, and how to keep quantification from masquerading as certainty. The field's enduring contribution is less any single method than a habit of mind—making decisions and their uncertainties explicit enough to be argued about.
References
- Decision analysis (History, Methodology), Wikipedia
- Decision theory (History), Wikipedia
- Value tree analysis (History), Wikipedia
- Subjective expected utility, Wikipedia
- Value of information, Wikipedia
The frameworks of Decision And Risk Analysis
Each framework is a way of working the field has used: when it held, and the question it answers.
- 1944-Present
Expected utility hypothesis
How should a rational agent choose among gambles?
- 1945-Present
Monte Carlo method
How can random sampling estimate a complex system's behavior?
- 1948-1980
Systems Analysis
How should large public choices be modeled and compared?
- 1954-Present
Subjective Expected Utility
How can beliefs be represented as probabilities from choices?
- 1960-Present
Decision Analysis
How should a one-off high-stakes decision be structured?
- 1970-Present
Multiple-criteria decision analysis
How should alternatives be compared on many objectives?
- 1975-Present
Probabilistic Risk Assessment
How likely and how bad are accidents in a complex system?
About Probabilistic Risk Assessment
Probabilistic Risk Assessment (PRA) is the framework that systems engineers use to answer a question that other decision methods often sidestep: How safe is safe enough? When a failure could be catastrophic—a nuclear meltdown, a satellite collision, a dam breach—engineers cannot rely on simple cost-benefit trade-offs or single-point safety margins. They need a systematic way to identify what can go wrong, how likely each scenario is, and what the consequences would be. PRA provides that discipline. It is the engineering framework for managing low-probability, high-consequence risks by combining event logic models of failure sequences with probabilistic models of component behavior. Unlike Decision Analysis, which optimizes a single decision under uncertainty, or Multi-Attribute Utility Theory, which trades off multiple objectives, PRA focuses on the risk profile of a system: the full set of accident sequences, their frequencies, and their consequences. Its core curriculum teaches engineers to build and analyze models that reveal where a system is most vulnerable and what design changes or operational procedures reduce risk most effectively.Core CurriculumA first course in Probabilistic Risk Assessment typically proceeds through these concepts in logical order:- Failure Modes and Effects Analysis (FMEA) and Event Tree Analysis and Fault Tree Analysis: These are the three foundational logic-modeling techniques. FMEA is a bottom-up, inductive method that asks, for each component, “How can it fail, and what happens next?” Fault Tree Analysis is a top-down, deductive method that starts with a top event (e.g., “loss of coolant”) and decomposes it into combinations of basic component failures using AND/OR logic. Event Tree Analysis is a forward, inductive method that traces the possible sequences of events (successes or failures of safety functions) following an initiating event. Together, they form the backbone of accident sequence modeling.
- Accident Sequence Analysis and Consequence Analysis: Accident Sequence Analysis links initiating events, through event trees and fault trees, to specific accident outcomes (e.g., core damage, release of radioactive material). Consequence Analysis then estimates the physical effects of those outcomes—how far a toxic plume travels, how many people are exposed, what the economic damage is. These two steps transform a logic model into a risk model.
- Common Cause Failure Analysis and Human Reliability Analysis: These address two critical sources of dependency that naive models miss. Common Cause Failure Analysis identifies failures that disable multiple redundant components simultaneously (e.g., a single maintenance error that affects all pumps). Human Reliability Analysis models the probability that operators will fail to perform a required action under accident conditions, accounting for stress, procedure quality, and time available.
- Uncertainty Propagation and Sensitivity Analysis and Importance Measures: Because component failure rates and human error probabilities are uncertain, PRA propagates these uncertainties through the logic models to produce a probability distribution over accident frequencies, not just a point estimate. Sensitivity Analysis identifies which input uncertainties most affect the output risk. Importance Measures (e.g., Risk Achievement Worth, Fussell-Vesely) rank components or human actions by their contribution to overall risk, guiding where to focus design improvements or inspection resources.
- Level 1, 2, 3 PRA and Probabilistic Safety Assessment (PSA): These terms define the scope of a PRA study. Level 1 PRA estimates the frequency of core damage (or equivalent plant damage states). Level 2 PRA models the progression of an accident from core damage to release of radioactive material (or other hazardous material) to the environment. Level 3 PRA estimates off-site consequences—health effects, economic costs, land contamination. The broader term Probabilistic Safety Assessment (PSA) is often used synonymously with PRA, especially in the nuclear industry, and emphasizes the safety application.
- Risk Aggregation and Risk Acceptance Criteria and Risk-Informed Decision Making: Risk Aggregation combines the frequencies and consequences of all accident sequences into a single risk profile (e.g., a complementary cumulative distribution function of consequences). Risk Acceptance Criteria are quantitative thresholds (e.g., “core damage frequency less than per reactor-year”) that define tolerable risk. Risk-Informed Decision Making uses the PRA results, together with deterministic safety requirements and other considerations, to make decisions about design changes, operational limits, inspection schedules, and regulatory actions.
Intellectual HistoryProbabilistic Risk Assessment emerged in the 1970s from the confluence of two earlier traditions: Reliability Engineering, which had developed fault tree analysis in the 1960s for the aerospace and missile industries, and Decision Analysis, which had formalized the use of subjective probability and expected utility for engineering decisions. The immediate catalyst was the need to assess the safety of commercial nuclear power plants, which were being built at a scale where a catastrophic accident, however unlikely, could have enormous consequences. The U.S. Nuclear Regulatory Commission (NRC) sponsored the landmark Reactor Safety Study (WASH-1400), published in 1975 under the direction of Norman Rasmussen. This study was the first comprehensive PRA of a nuclear power plant. It used event trees and fault trees to estimate the frequency of core damage and release scenarios, and it introduced the concept of a “risk curve” showing the probability of exceeding various consequence levels. WASH-1400 was controversial—its methods and conclusions were debated for years—but it established PRA as a distinct framework within systems engineering.Key figures in the early development of PRA include Norman Rasmussen (MIT), who directed WASH-1400; Stanley Kaplan, who later, with John Garrick, formalized the definition of risk as a triplet of scenarios, likelihoods, and consequences; and George Apostolakis, who advanced the treatment of uncertainty and Bayesian methods in PRA. The seminal works of this period are the Reactor Safety Study (1975) and the PRA Procedures Guide (NUREG/CR-2300, 1983), which codified the methodology for the nuclear industry. The framework was soon adopted by other high-hazard industries: aerospace (NASA’s use of PRA after the Challenger accident), chemical process safety, and, later, offshore oil and gas.PRA coexists with and has been supplemented by other frameworks in decision and risk analysis. Multi-Attribute Utility Theory (1970s) provides a way to trade off multiple objectives (e.g., cost, safety, environmental impact) that PRA does not directly address. Real Options Analysis (1990s) adds a focus on flexibility and the value of delaying decisions under uncertainty, which is less central to PRA’s emphasis on fixed designs and operational procedures. Robust Decision Making (2000s) addresses deep uncertainty where probability distributions are unknown, a situation that PRA’s probabilistic framework handles less comfortably. Within its core PRA, later developments include dynamic PRA (modeling time-dependent accident sequences) and the integration of PRA with risk-informed regulation, but the core curriculum taught today remains remarkably close to the methods established in the 1970s and 1980s.The 15 concepts its tower teaches
- Accident Sequence Analysis
- Common Cause Failure Analysis
- Consequence Analysis
- Event Tree Analysis
- Failure Modes and Effects Analysis
- Fault Tree Analysis
- Human Reliability Analysis
- Importance Measures
- Level 1, 2, 3 PRA
- Probabilistic Safety Assessment
- Risk Acceptance Criteria
- Risk Aggregation
- Risk-Informed Decision Making
- Sensitivity Analysis
- Uncertainty Propagation
- 1980-Present
- 1985-Present
Structured Expert Judgment
How can expert opinion become defensible probabilities?
- 1995-Present
Robust Decision Making
Which strategies work well across many uncertain futures?
How the ideas connect
Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.
Subjective Expected Utility grew out of Expected utility hypothesis
Savage's Foundations of Statistics extended von Neumann and Morgenstern's expected utility axioms to subjective probability, so that beliefs as well as values could be represented numerically. This made the expected-utility framework usable for one-of-a-kind decisions without frequency data.
Carried by Leonard Savage; The Foundations of Statistics (1954).
Decision Analysis influenced by Expected utility hypothesis
Decision analysis took the expected-utility maximization criterion from von Neumann and Morgenstern and turned it into a practical craft of decision trees and rolled-back expected values. Raiffa and Schlaifer carried the theory into business and engineering practice.
Carried by Howard Raiffa; Robert Schlaifer.
Decision Analysis influenced by Subjective Expected Utility
Raiffa and Schlaifer's applied decision analysis built directly on Savage's subjective expected utility, using subjective probabilities on chance branches of decision trees. Their textbooks codified the vocabulary of applied subjective expected utility.
Carried by Howard Raiffa; Robert Schlaifer; Applied Statistical Decision Theory (1961).
Multiple-criteria decision analysis influenced by Expected utility hypothesis
Multi-attribute utility theory extended expected-utility ideas to choices among alternatives judged on several objectives, as in Keeney and Raiffa's Decisions with Multiple Objectives. The utility axioms supplied the basis for eliciting weights and combining scores.
Carried by Ralph Keeney; Howard Raiffa; Decisions with Multiple Objectives (1976).
Robust Decision Making reacted against Expected utility hypothesis
Robust decision making arose as a response to the expected-utility framework's assumption that probabilities and models are known or estimable. It instead seeks strategies performing acceptably across many plausible futures under deep uncertainty.
Carried by Robert Lempert.
Probabilistic Risk Assessment influenced by Monte Carlo method
Probabilistic risk assessment used Monte Carlo sampling to propagate uncertainties through fault and event trees where analytical combination was intractable. The WASH-1400 study and later PRAs relied on simulation to estimate accident frequencies and their uncertainty ranges.
Carried by WASH-1400 (1975).
Robust Decision Making influenced by Monte Carlo method
Robust decision making uses computer search over many simulated futures, drawing on Monte Carlo sampling to generate the scenario ensembles against which strategies are tested. Simulation supplies the plausible futures over which robustness is evaluated.
Carried by Robert Lempert.
Probabilistic Risk Assessment influenced by Systems Analysis
Systems analysis's practice of modeling a large technical system, costing alternatives and examining uncertainties was carried into the Reactor Safety Study's systematic treatment of reactor accidents. The RAND-style systems approach shaped how PRA structured the problem of a whole plant.
Carried by Norman Rasmussen; WASH-1400 (1975).
Multiple-criteria decision analysis grew out of Decision Analysis
Multi-criteria decision analysis grew out of decision analysis when the single-attribute expected-utility framework was generalized to multiple incommensurable objectives. Keeney and Raiffa's multi-attribute utility work carried the decision-analysis craft of elicitation into multi-objective settings.
Carried by Ralph Keeney; Howard Raiffa; Decisions with Multiple Objectives (1976).
Robust Decision Making reacted against Decision Analysis
Robust decision making defined itself against the standard decision-analysis practice of assigning probabilities and maximizing expected utility. Lempert and colleagues at RAND replaced the single optimal choice with strategies robust across a wide range of futures.
Carried by Robert Lempert.
Structured Expert Judgment influenced by Probabilistic Risk Assessment
Probabilistic risk assessment's need for probabilities of rare events lacking data drove the development of formal expert elicitation. Structured expert judgment supplied defensible input distributions for fault and event trees in PRA.
Carried by Roger Cooke.
Risk Management influenced by Monte Carlo method
Risk management practice adopted Monte Carlo simulation to characterize the spread and tails of risk in cost, schedule and safety models. Simulation let managers see distributions of outcomes rather than single point estimates.
Beyond the field
Where Decision And Risk Analysis's ideas came from in other fields, and what they led to there: the strongest 12 of 80 links.
Expected utility hypothesis provoked Bounded Rationality in Reasoning Decision Making
Simon's bounded rationality was framed explicitly against the expected-utility model of fully rational choice, arguing that real agents satisfice rather than maximize expected utility.
Carried by Herbert A. Simon.
Expected utility hypothesis provoked Prospect Theory in Choice Under Uncertainty
Prospect theory defined itself against expected utility as used in decision and risk analysis, showing systematic violations of its axioms.
Carried by Daniel Kahneman; Amos Tversky.
Expected utility hypothesis led to Bayesian decision theory in Formal Epistemology
Von Neumann and Morgenstern's expected utility hypothesis (1944) supplied the utility axioms that Bayesian decision theory combines with subjective probability; Savage's 1954 system unified the two.
Carried by John von Neumann; Oskar Morgenstern; Theory of Games and Economic Behavior (1944).
Expected utility hypothesis replaced Expected Value Theory in Choice Under Uncertainty
Expected value theory, the Pascal-Fermat criterion of maximising monetary expectation, was replaced by Bernoulli's expected utility as the working account of choice under risk.
Carried by Daniel Bernoulli.
Expected utility hypothesis led to Subjective Expected Utility in Choice Under Uncertainty
Subjective expected utility is a generalization of expected utility: Savage replaced objective probabilities with subjective ones while keeping the expected-utility maximization structure.
Carried by Leonard J. Savage; The Foundations of Statistics (1954).
Expected utility hypothesis led to Smooth Ambiguity Model in Choice Under Uncertainty
The smooth ambiguity model of Klibanoff, Marinacci and Mukerji is formulated as a generalisation of expected utility in which a second-order prior over probabilities enters a concave ambiguity function. It retains the expected-utility representation as the linear limiting case.
Carried by Peter Klibanoff; Massimo Marinacci; Fabio Mukerji.
Expected utility hypothesis led to Rank-dependent expected utility in Choice Under Uncertainty
Rank-dependent expected utility is a generalization of expected utility that replaces linear probability weighting with a rank-dependent transformation while retaining the expected-utility structure.
Carried by John Quiggin.
Expected utility hypothesis provoked Ambiguity Aversion Models in Choice Under Uncertainty
Ambiguity aversion models were developed as alternatives to expected utility, explaining Ellsberg-type choices that violate its sure-thing principle.
Carried by Daniel Ellsberg.
Expected utility hypothesis led to Maxmin Expected Utility in Choice Under Uncertainty
Gilboa and Schmeidler's maxmin expected utility keeps the expected-utility structure of Anscombe-Aumann acts and replaces the single prior with a set of priors, so the Savage/Anscombe-Aumann expected utility framework is the base it generalises. The 1989 paper is explicitly a modification of expected utility under ambiguity.
Carried by Itzhak Gilboa; David Schmeidler.
Expected utility hypothesis led to Subjective expected utility in Reasoning Decision Making
Savage's 1954 Foundations of Statistics generalised von Neumann-Morgenstern expected utility to subjective probabilities, producing subjective expected utility directly out of the earlier framework.
Carried by Leonard J. Savage; The Foundations of Statistics (1954).
Expected utility hypothesis provoked Bounded Rationality in Behavioral Decision Theory
Bounded rationality arose as a critique of the expected utility hypothesis and its assumption of consistent maximisation. Simon argued that real agents lack the computational capacity the expected utility model presumes.
Carried by Herbert A. Simon.
Monte Carlo method influenced Monte Carlo Probabilistic Exposure and Risk Analysis in Environmental Risk Assessment
The Monte Carlo method, developed at Los Alamos in the 1940s, supplied the sampling technique that probabilistic exposure and risk analysis applied to distributions of exposure parameters. The later field is essentially the Monte Carlo method turned on environmental exposure and dose models.

Decision And Risk Analysis is a planet in Nousfarer
Its city has a tower for each of these 10 frameworks. Land, walk in, and climb one concept at a time.
Land on Decision And Risk AnalysisRelated fields
Fields its ideas are linked to
- Choice Under Uncertainty 13 links
- Managerial Decision Making 7 links
- Operations Research 6 links
- Reasoning Decision Making 3 links
- Agricultural Economics 2 links
- Behavioral Decision Theory 2 links
- Impact Evaluation 2 links
- Environmental Risk Assessment 2 links
- Environmental Science 2 links
- Health Technology Assessment 2 links
- Business Analytics 2 links
- Systems Engineering 2 links
More in Systems Engineering
- Model Based Systems Engineering
- System Safety
- Systems Architecture
- Systems Engineering
- Systems Integration
- Systems Lifecycle
- Systems Verification
See where Decision And Risk Analysis sits on the Map of Human Knowledge