Chapter 7: Beyond Bayesian inference: Raiffa, Schlaifer, and rational decision-making
The introduction explains how Raiffa and Schlaifer extended Bayesian inference into a framework for choosing actions and deciding whether more information is worth obtaining.
After this chapter you will be able to
- From posterior belief to choice. Explain why a posterior probability alone does not prescribe an action, and how Raiffa and Schlaifer combined posterior probabilities with utilities to compare actions by expected utility.
- Worth of additional evidence. Explain when obtaining more information is worthwhile in Raiffa and Schlaifer’s framework.
- Posterior probability and utility. Describe utility as a common numerical scale for how desirable outcomes are, including nonmonetary effects, and explain how posterior probabilities are used with those utilities to compare actions by expected utility.
- Probability and Consequence Distinction. Explain why a posterior probability of disease, even after a positive test, does not by itself determine whether to treat. Describe how utilities express the decision maker’s judgments about the desirability of each outcome.
- Expected Utility Action Comparison. Apply the section’s expected-utility comparison: for each action, weight its utilities for disease present and absent by their posterior probabilities, add the products, and prefer the action with the higher total. Explain why changing probabilities or valuations could change that preference.
- Decision tree structure. Describe how decision nodes, chance nodes, terminal nodes, and branches represent a choice under uncertainty.
- Rollback and preferred action. Trace how rollback evaluates a decision tree from its terminal outcomes to the initial choice.
- Expected value of learning. Explain how to compare acting immediately with obtaining information before choosing an action. Describe how possible results, their probabilities, and the preferred action after each result enter the comparison.
- Perfect and sample information. Distinguish the expected value of perfect information from the expected value of sample information, and explain why perfect information sets an upper bound.
- Action changes and acquisition costs. Explain when information has positive decision value and when it is worth acquiring after costs.
- Treatment threshold and actions. Explain how the assigned utilities establish a treatment threshold and why neither second-test result changes the preferred action.
- Second test decision value. Explain how the section compares acting immediately with testing first, and distinguish the test’s information from its decision value.
- Components of a decision. Explain how Raiffa and Schlaifer turn the physician example into a framework for decisions across domains by identifying acts, states, and experiments.
- Acts and feasible alternatives. Describe acts as the actions a decision maker can control, and explain how to define a useful set of alternatives.
- States and current beliefs. Explain what states represent and how probabilities express current beliefs about them.
- Experiments before final action. Distinguish experiments from final acts and explain when gathering information is useful.
- Belief to action cycle. Trace how prior beliefs, possible evidence, posterior probabilities, and expected utility lead to an act.
- Expected Utility Choice Rule. Explain how expected utility combines preferences and posterior beliefs to identify the preferred act: weight each act–state outcome's utility by the state's posterior probability, total the results for each act, and choose the act with the highest value.
- Preposterior value of information. Describe how preposterior analysis evaluates whether to acquire information before its result is known, and distinguish that question from choosing an act after evidence is available.
- State-Specific Opportunity Loss. Explain opportunity loss as the utility sacrificed when an act is not the best one for the state that occurs, using the physician’s treatment choices to illustrate it.
- Expected Loss and Utility. Explain how probabilities turn state-specific opportunity losses into expected opportunity loss, and state how choosing by that measure compares with choosing by expected utility.
- Opportunity loss for an act. Explain how opportunity loss compares a chosen act with the act that would be best if the true state were known. State when that loss is zero or positive.
- Expected loss and utility equivalence. Describe how probabilities turn opportunity losses into expected opportunity loss, and explain why minimizing it gives the same preferred act as maximizing expected utility.
- Expected utility launch choice. Explain why delaying the launch is preferred when demand is uncertain, even though launching is best if demand is high.
- Opportunity loss comparison. Describe how opportunity losses are assigned and probability-weighted in the launch example, and explain why their comparison agrees with expected utility.
- Conjugate priors as reusable updates. Describe how conjugate priors make Bayesian learning practical and repeatable in this section’s account.
- Posterior normalization challenge. Explain how the prior and likelihood determine the posterior’s shape, why their product must be normalized, and why doing so can be difficult for repeated updates.
- Conjugate parameter updating. Describe what makes a prior conjugate to a likelihood and how that relationship makes repeated Bayesian updating more practical, using a pairing from the section as an example.
- Beta prior as pseudocounts. Explain why a Beta distribution represents the unknown defect probability and how its two parameters encode the manufacturer’s prior belief.
- Conjugate posterior parameter update. Trace how the observed defects and non-defects update the Beta prior, and interpret the resulting posterior mean.
- Computational value of conjugacy. Explain how conjugate priors replace repeated integration or new derivations with exact parameter updates, and why that remains useful even with modern computers.
- Continuity of sequential updating. Describe how a conjugate posterior can serve as the next prior, and what typically happens as observations accumulate; illustrate with one application from the section.
- Disciplined choices under uncertainty. Explain how the framework helps decision-makers structure uncertain choices, weigh probabilities against consequences, assess whether more information is worth its cost, and select the act with the highest expected utility.
- Decision Quality Versus Outcomes. Explain why a decision should be judged by the evidence and expected utility available when it was made, rather than by its eventual outcome. Describe how explicit assumptions and sensitivity analysis make that reasoning open to review.
- Staged R&D Funding. Describe how an organization can fund an R&D project in stages, use each stage’s results to reconsider its options, and pay for another experiment only when its expected information value exceeds its cost.
- Decision framework portability. Explain why the same decision framework can guide choices for companies, patients, and the public, and how government decisions make its trade-offs explicit.
- Prediction Versus Decision. Distinguish predicting what is likely from deciding what to do, and explain how decision analysis brings beliefs, consequences, and preferences together.
- Exploration and Exploitation. Explain how a multi-armed bandit balances acting on the currently best-supported option with testing an uncertain alternative, and relate that balance to the value of information.
- Repeated Automated Choices. Trace how an automated decision system updates probabilities from new evidence, compares acts by expected utility, and revisits its choice as evidence changes.
What it builds on
- From prior to posterior: how Bayes’ Theorem works (chapter 2) · Introduces updating a prior belief to a posterior probability.
- Bayesian inference in real-world scenarios (chapter 2) · Bayesian updating of disease probability after diagnostic evidence.
- Information as the reduction of uncertainty (chapter 5) · Introduces information as a reduction of uncertainty.
- The mathematical structure of Bayes’ Theorem (chapter 2) · Assumes the prior, likelihood, and posterior structure of Bayes’ Theorem.
- Maximum likelihood estimation (chapter 3) · The idea that a likelihood can favor a particular parameter value.
Section by section
Introduction
The introduction explains how Raiffa and Schlaifer extended Bayesian inference into a framework for choosing actions and deciding whether more information is worth obtaining.
- From posterior belief to choice: Explain why a posterior probability alone does not prescribe an action, and how Raiffa and Schlaifer combined posterior probabilities with utilities to compare actions by expected utility.
Why it matters: A likely outcome is not necessarily the outcome that makes an action worth taking. - Worth of additional evidence: Explain when obtaining more information is worthwhile in Raiffa and Schlaifer’s framework.
Why it matters: More evidence is useful only if its potential benefit to the decision warrants obtaining it. - Beliefs and preferences kept distinct: Describe why the framework separates beliefs about possible outcomes from preferences about their consequences.
Why it matters: Separating what you think will happen from what you value makes a consequential choice easier to examine.
Key terms: posterior probabilities, utilities, expected utility, additional information, tests, experiments, beliefs, preferences, consequences
7.1 Revisiting Bayes: from diagnosis to decision
The section uses a medical diagnosis to show why choosing an action requires weighing its possible consequences alongside posterior probabilities.
- Consequences of treatment choices: Explain why a positive test result does not by itself settle whether to treat: treatment and non-treatment have different consequences depending on whether disease is present.
Why it matters: The same probability can lead to different choices when the consequences of acting differ. - Posterior probability and utility: Describe utility as a common numerical scale for how desirable outcomes are, including nonmonetary effects, and explain how posterior probabilities are used with those utilities to compare actions by expected utility.
Why it matters: It connects what the evidence suggests with what each possible decision would mean for the patient.
Key terms: actions, states of the world, consequence, utility, posterior probabilities, expected utility
7.1.1 From posterior probability to expected utility
This section shows how posterior probabilities and judgments about consequences are combined through expected utility to choose between treating and not treating a patient.
- Probability and Consequence Distinction: Explain why a posterior probability of disease, even after a positive test, does not by itself determine whether to treat. Describe how utilities express the decision maker’s judgments about the desirability of each outcome.
Why it matters: A test result becomes useful for a decision only when you also consider what each action could lead to. - Expected Utility Action Comparison: Apply the section’s expected-utility comparison: for each action, weight its utilities for disease present and absent by their posterior probabilities, add the products, and prefer the action with the higher total. Explain why changing probabilities or valuations could change that preference.
Why it matters: The comparison turns updated beliefs and valued consequences into an explicit choice.
Key terms: posterior probability, utilities, consequences, expected utility, posterior probabilities, preferred action
7.1.2 Decision trees: visualizing choice under uncertainty
This section shows how a decision tree displays choices, uncertainty, and consequences, then uses rollback to identify the action with the highest expected utility.
- Decision tree structure: Describe how decision nodes, chance nodes, terminal nodes, and branches represent a choice under uncertainty.
Why it matters: The structure makes clear which parts of a problem are choices, which are uncertainties, and what each outcome is worth. - Rollback and preferred action: Trace how rollback evaluates a decision tree from its terminal outcomes to the initial choice.
Why it matters: Rollback turns the visible paths in a tree into a defensible choice. - Decision versus predictive trees: Distinguish a decision-analysis tree from a predictive tree used in machine learning.
Why it matters: The distinction prevents a diagram for choosing an action from being mistaken for a model trained to predict outcomes. - Information gathering branches: Explain how a decision tree can represent learning before acting and the comparison that makes the information valuable.
Why it matters: The tree can help decide not only what to do, but whether to learn more before doing it.
Key terms: decision node, chance node, terminal node, branches, utility, rollback, expected utility, root, decision-analysis trees, predictive trees
7.1.3 Understanding the value of information
This section explains how to decide whether additional information is worth acquiring by comparing its effect on expected utility with its cost.
- Expected value of learning: Explain how to compare acting immediately with obtaining information before choosing an action. Describe how possible results, their probabilities, and the preferred action after each result enter the comparison.
Why it matters: It makes the choice to learn comparable to the choice to act immediately. - Perfect and sample information: Distinguish the expected value of perfect information from the expected value of sample information, and explain why perfect information sets an upper bound.
Why it matters: The distinction separates the value of knowing the truth from the value of a fallible test, survey, or experiment. - Action changes and acquisition costs: Explain when information has positive decision value and when it is worth acquiring after costs.
Why it matters: A useful result can still be too costly to obtain, while a result that cannot change the decision offers no benefit before costs. - Information content versus decision value: Distinguish Shannon's measure of information from Raiffa and Schlaifer's measure of its decision value, using the effect of a probability revision on the preferred action.
Why it matters: It prevents confusing a large change in beliefs with an improvement in the decision.
Key terms: expected value of information, expected utility, Expected Value of Perfect Information, EVPI, Expected Value of Sample Information, EVSI, posterior beliefs, decision value, acquisition cost, net value of sample information
7.1.4 Applying the value of information
The section applies Bayesian updating, a treatment threshold, and expected utility to decide whether a second diagnostic test is worth its cost.
- Second test belief updates: Describe how the second test’s accuracy and possible results change the estimated probability of disease.
Why it matters: The test’s possible outcomes must be evaluated separately before deciding whether testing helps. - Treatment threshold and actions: Explain how the assigned utilities establish a treatment threshold and why neither second-test result changes the preferred action.
Why it matters: A change in probability matters to this choice only if it changes which action has higher expected utility. - Second test decision value: Explain how the section compares acting immediately with testing first, and distinguish the test’s information from its decision value.
Why it matters: A more precise diagnosis is not worth buying when it cannot improve the choice enough to cover its cost.
Key terms: sensitivity, specificity, false positive, false negative, posterior probability, treatment threshold, expected utility, Expected Value of Sample Information, EVSI, net value of sample information
7.2 The decision framework of Raiffa and Schlaifer
This section introduces Raiffa and Schlaifer’s reusable decision framework by identifying the basic elements of a choice under uncertainty.
- Components of a decision: Explain how Raiffa and Schlaifer turn the physician example into a framework for decisions across domains by identifying acts, states, and experiments.
Why it matters: These distinctions make the logic of one decision reusable in other settings.
Key terms: acts, states, experiments
7.2.1 Acts, states, and experiments
This section defines acts, states, and experiments and shows how they connect prior beliefs, new evidence, and a choice of action.
- Acts and feasible alternatives: Describe acts as the actions a decision maker can control, and explain how to define a useful set of alternatives.
Why it matters: A decision analysis cannot select a good action if that action was left out of the choices considered. - States and current beliefs: Explain what states represent and how probabilities express current beliefs about them.
Why it matters: States distinguish what a person can choose from the uncertainty their choice must account for. - Experiments before final action: Distinguish experiments from final acts and explain when gathering information is useful.
Why it matters: More information is worthwhile only if its potential effect on the decision justifies obtaining it. - Belief to action cycle: Trace how prior beliefs, possible evidence, posterior probabilities, and expected utility lead to an act.
Why it matters: The cycle connects uncertainty and learning to an actionable choice.
Key terms: acts, set of acts, states, mutually exclusive, collectively exhaustive, experiments, evidence, expected decision value, decision-analytic cycle, prior beliefs
7.2.2 The expected utility formulation
This section shows how to choose an act under uncertainty by combining the utility of each possible outcome with the posterior probability of its state.
- Expected Utility Choice Rule: Explain how expected utility combines preferences and posterior beliefs to identify the preferred act: weight each act–state outcome's utility by the state's posterior probability, total the results for each act, and choose the act with the highest value.
Why it matters: It connects what the decision maker values with what the evidence suggests is likely.
Key terms: act, state, utility, posterior probability, expected utility, preferred act
7.2.3 Terminal versus preposterior analysis
This section distinguishes choosing an act after evidence is known from deciding whether to obtain information before its result is known.
- Terminal choice after evidence: Explain how terminal analysis uses known evidence to choose an act.
Why it matters: It identifies the decision that can be made once a test result is in hand. - Preposterior value of information: Describe how preposterior analysis evaluates whether to acquire information before its result is known, and distinguish that question from choosing an act after evidence is available.
Why it matters: It makes the value of learning assessable before a test result arrives.
Key terms: terminal analysis, posterior probabilities, expected utility, act, preposterior analysis, posterior beliefs, value of learning
7.3 Opportunity loss, regret, and expected utility
The section reframes a physician’s treatment decision in terms of opportunity loss and explains why minimizing expected loss selects the same act as maximizing expected utility.
- State-Specific Opportunity Loss: Explain opportunity loss as the utility sacrificed when an act is not the best one for the state that occurs, using the physician’s treatment choices to illustrate it.
Why it matters: It gives you another way to compare acts under uncertainty: by what each choice gives up. - Expected Loss and Utility: Explain how probabilities turn state-specific opportunity losses into expected opportunity loss, and state how choosing by that measure compares with choosing by expected utility.
Why it matters: The loss perspective changes how you describe a decision without changing which act the framework recommends.
Key terms: opportunity losses, utility, act, state, expected opportunity loss, expected regret, expected utility
7.3.1 From utility to opportunity loss
Defines opportunity loss for an act in a given state and shows why minimizing its expected value selects the same act as maximizing expected utility.
- Opportunity loss for an act: Explain how opportunity loss compares a chosen act with the act that would be best if the true state were known. State when that loss is zero or positive.
Why it matters: It lets you describe a choice in terms of satisfaction sacrificed rather than satisfaction gained. - Expected loss and utility equivalence: Describe how probabilities turn opportunity losses into expected opportunity loss, and explain why minimizing it gives the same preferred act as maximizing expected utility.
Why it matters: It shows that a regret-based description changes the perspective on a decision, not the preferred act.
Key terms: opportunity loss, utility, optimal act, state of nature, expected opportunity loss, expected utility, probability-weighted average
7.3.2 A simple illustration
A product-launch example shows that maximizing expected utility and minimizing expected opportunity loss select the same action under uncertain demand.
- Expected utility launch choice: Explain why delaying the launch is preferred when demand is uncertain, even though launching is best if demand is high.
Why it matters: The best action under one possible outcome need not be the best action before the outcome is known. - Opportunity loss comparison: Describe how opportunity losses are assigned and probability-weighted in the launch example, and explain why their comparison agrees with expected utility.
Why it matters: Opportunity loss gives a second perspective on the same decision without changing which act is preferred.
Key terms: acts, states, expected utility, opportunity loss, expected opportunity loss
7.4 Conjugate priors and computable Bayesian learning
The section explains why repeated decisions need practical Bayesian updating and how conjugate priors make that updating reusable.
- Updating beliefs for decisions: Explain why Raiffa and Schlaifer’s decision framework needed a practical way to update probabilities when experiments produced new evidence.
Why it matters: A decision framework is useful in practice only if new evidence can be incorporated repeatedly. - Conjugate priors as reusable updates: Describe how conjugate priors make Bayesian learning practical and repeatable in this section’s account.
Why it matters: Conjugacy makes repeated belief updates manageable enough to support decisions as evidence arrives.
Key terms: posterior probabilities, expected utility analysis, conjugate priors, likelihood functions, posterior distributions, computable Bayesian learning
7.4.1 From integration to parameter updating
The section explains why posterior normalization can make repeated Bayesian updating difficult and how conjugate priors turn it into parameter updating.
- Posterior normalization challenge: Explain how the prior and likelihood determine the posterior’s shape, why their product must be normalized, and why doing so can be difficult for repeated updates.
Why it matters: This difficulty explains the practical problem that conjugate priors solve. - Conjugate parameter updating: Describe what makes a prior conjugate to a likelihood and how that relationship makes repeated Bayesian updating more practical, using a pairing from the section as an example.
Why it matters: Reusable parameter updates made ongoing belief revision practical for experimentation and decisions.
Key terms: prior, likelihood, posterior distribution, normalized, closed-form solution, conjugate priors, distribution family, parameter updating, Beta, Binomial
7.4.2 Beta-Binomial updating
This section uses a machine-defect example to show how a Beta prior and Binomial data produce an exact posterior through simple parameter addition.
- Beta prior as pseudocounts: Explain why a Beta distribution represents the unknown defect probability and how its two parameters encode the manufacturer’s prior belief.
Why it matters: The prior counts make it clear how historical evidence enters the updated estimate. - Binomial evidence and likelihood: Describe the inspection as a Binomial experiment and distinguish the defect rate most supported by the sample from the rate obtained after updating the prior.
Why it matters: The sample’s 20% rate is evidence for the update, not the final estimate by itself. - Conjugate posterior parameter update: Trace how the observed defects and non-defects update the Beta prior, and interpret the resulting posterior mean.
Why it matters: Parameter addition turns the new inspection results into an updated estimate without discarding the earlier evidence. - Conjugacy benefits and limits: Explain why the conjugate update is computationally convenient without being an approximation, and identify its stated limitation.
Why it matters: The shortcut saves computation, but the prior still needs to be a suitable way to express the available beliefs.
Key terms: Beta distribution, prior, pseudocounts, defect rate, Binomial experiment, Binomial likelihood, defect probability, sample proportion, conjugate pair, posterior
7.4.3 Why conjugacy matters
This section explains why conjugate priors make repeated Bayesian updating practical, both historically and in modern systems.
- Computational value of conjugacy: Explain how conjugate priors replace repeated integration or new derivations with exact parameter updates, and why that remains useful even with modern computers.
Why it matters: It explains how updating beliefs can be practical rather than just a theoretical prescription. - Continuity of sequential updating: Describe how a conjugate posterior can serve as the next prior, and what typically happens as observations accumulate; illustrate with one application from the section.
Why it matters: The same update structure lets a system keep revising its beliefs as new evidence arrives.
Key terms: conjugate priors, parameter updates, closed-form updates, computable Bayesian learning, posterior, prior, distributional form, sequential Bayesian learning
7.5 The enduring value of a rational decision-making framework
This section explains why Raiffa and Schlaifer’s framework remains useful for making uncertain choices, both in management and beyond it.
- Disciplined choices under uncertainty: Explain how the framework helps decision-makers structure uncertain choices, weigh probabilities against consequences, assess whether more information is worth its cost, and select the act with the highest expected utility.
Why it matters: It gives you a way to explain why a choice is justified, rather than relying on intuition alone. - Broader reach and AI connection: Describe how the same decision logic applies beyond business and explain the qualified connection to machine learning and AI.
Why it matters: It helps you recognize the same decision logic in different fields without overstating its influence on AI.
Key terms: expected utility, additional information, uncertain choices, incomplete information, expected consequences, learning with acting
7.5.1 Decision analysis as a managerial discipline
This section shows how decision analysis helps organizations judge choices by their reasoning and make staged investment, risk, and strategic decisions under uncertainty.
- Decision Quality Versus Outcomes: Explain why a decision should be judged by the evidence and expected utility available when it was made, rather than by its eventual outcome. Describe how explicit assumptions and sensitivity analysis make that reasoning open to review.
Why it matters: It lets an organization learn from how it made a choice without confusing luck with sound judgment. - Staged R&D Funding: Describe how an organization can fund an R&D project in stages, use each stage’s results to reconsider its options, and pay for another experiment only when its expected information value exceeds its cost.
Why it matters: Staging preserves the option to pursue a promising project while limiting how much is committed before key uncertainties are resolved. - Low Probability Severe Risks: Explain how expected utility brings both the probability and consequences of action and inaction into a decision involving an unlikely but severe risk.
Why it matters: It keeps a foreseeable severe loss in the comparison without treating that loss as certain. - Scenario Based Conditional Plans: Describe how scenario planning compares strategies across possible futures using current probabilities, and why waiting or gathering evidence can change the preferred strategy.
Why it matters: A conditional plan helps an organization respond to new evidence rather than treating an uncertain forecast as fixed.
Key terms: decision quality, expected utility, sensitivity analysis, R&D, staged approach, expected value of the information, gray swan events, fat-tailed distributions, scenario planning, value of waiting
7.5.2 Beyond managerial decision-making
The section shows how Raiffa and Schlaifer’s decision framework extends beyond management to public, scientific, and medical choices.
- Decision framework portability: Explain why the same decision framework can guide choices for companies, patients, and the public, and how government decisions make its trade-offs explicit.
Why it matters: It lets you recognize the same decision problem in settings with very different stakes. - Evidence acquisition trade-offs: Explain how the value-of-information framework helps assess whether to conduct a scientific or medical experiment before making a decision.
Why it matters: More evidence is worth seeking only when its expected help with the decision justifies what it takes to obtain it.
Key terms: acts, uncertain states, consequences, probabilities, utilities or losses, value-of-information framework, expected improvement, experimentation
7.5.3 From decision analysis to machine learning and AI
The section shows how Raiffa and Schlaifer’s decision logic appears in systems that learn while acting or automate choices under uncertainty, while distinguishing decisions from predictions.
- Prediction Versus Decision: Distinguish predicting what is likely from deciding what to do, and explain how decision analysis brings beliefs, consequences, and preferences together.
Why it matters: An accurate forecast cannot by itself identify the best action. - Exploration and Exploitation: Explain how a multi-armed bandit balances acting on the currently best-supported option with testing an uncertain alternative, and relate that balance to the value of information.
Why it matters: The best immediate choice may differ from the choice that helps make better decisions later. - Repeated Automated Choices: Trace how an automated decision system updates probabilities from new evidence, compares acts by expected utility, and revisits its choice as evidence changes.
Why it matters: Repeated decisions can adapt to new evidence without confusing what is likely with what is desirable.
Key terms: prediction, decision-making, utility, opportunity loss, states of the world, multi-armed bandit, exploitation, exploration, value-of-information analysis, inference
7.6 Summary
Where this is used later
- Raiffa and Schalifer: From Bayesian inference to rational action (chapter 10) · Representing choices under uncertainty with decision trees.
- The strengths and limitations of single decision trees (chapter 9) · Discusses predictive decision trees as machine-learning models.
Concepts explained in this chapter
- conjugacyA prior distribution that yields a posterior in the same distribution family when updated with data under a particular likelihood.
- expected value of sample informationThe expected value of sample information is the expected improvement in decision outcome from imperfect information before its acquisition cost is deducted.
- expected utilityExpected utility is the probability-weighted average desirability of the possible outcomes of an action.
- value of informationThe expected improvement in a decision outcome attributable to acquiring additional information before acting.
- net value of sample informationThe net value of sample information is the expected value of sample information after subtracting the cost of acquiring the information.
- expected value of perfect informationExpected value of perfect information is the expected improvement in utility from knowing the true state before choosing an action.
- chance nodeA node in a decision tree representing an uncertain event that the decision-maker cannot control.
- decision nodeA node in a decision tree representing a point at which the decision-maker chooses among available actions.
- leaf nodeA terminal node is a node marking the end of a decision-tree path and recording its resulting outcome.
- rollbackA procedure for evaluating a decision tree in reverse, combining probabilities and utilities at chance nodes and selecting the best branch at decision nodes.
- treatment thresholdA treatment threshold is the probability of a condition at which treating and not treating have equal expected utility.
- experimentationAn experiment is an information-gathering procedure that can reveal something about uncertain states before a decision.
- beta distributionA continuous probability distribution on values from zero to one, parameterized by two positive shape parameters.
- likelihoodA likelihood function measures how compatible observed data are with different parameter values of a statistical model.
- posterior probabilityA posterior is a probability or probability distribution updated after incorporating observed evidence.
- actionA choice available to a decision-maker in a decision problem.
- decision treeA decision tree is a hierarchical predictive model that recursively partitions data using feature-based rules.
- preposterior analysisPreposterior analysis is decision analysis conducted before information is obtained, evaluating possible results and their consequences for later choices.
- terminal analysisTerminal analysis is decision analysis conducted after information is observed, using the resulting beliefs to choose an act.
- opportunity lossOpportunity loss is the loss incurred by choosing an action instead of the best action that would have been selected with the relevant information.
- decision frameworkAn organized structure of methods and assumptions for evaluating choices and selecting actions.
- state of natureA possible condition outside a decision-maker’s control that affects the outcome of an act.
- utility functionUtility is a numerical representation of the desirability or preference associated with an outcome.
- decision analysisDecision analysis formally evaluates choices by representing uncertain states, possible actions, consequences, and preferences.
- Bayesian inferenceBayesian inference updates prior probabilities or distributions about hypotheses or parameters with observed data.
- decision makingDecision-making is the process of selecting an action or outcome under available information, constraints, and uncertainty.
- stateA representation of a problem configuration that captures information needed to determine subsequent choices and outcomes.
- recallThe true positive rate is the proportion of actually positive cases correctly identified.
- specificitySpecificity is the probability that a test gives a negative result when the condition being tested for is absent.
- entropyEntropy is a quantitative measure of uncertainty across the possible outcomes of an information source and their probabilities.
- informationInformation is a quantifiable reduction in uncertainty about the state or outcome of a variable or system.
- beta-binomial distributionThe beta-binomial model combines a Beta prior for an unknown success probability with a binomial likelihood for observed trial outcomes.
- binomial distributionThe binomial distribution gives the probabilities of numbers of successes in a fixed number of independent trials with a constant success probability.
- objective functionA training objective is a quantity that a learning procedure seeks to optimize while fitting a model.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.
