eduKateSG · GS1 · How X Works
Find your next learning step
A useful decision model shows what can be chosen, what remains uncertain and why one option follows from the stated assumptions. Start with the part of the reasoning you need.
Full chapter index · Work the expected-utility example · Calculate imperfect-information value · Connected article directory
The gold standard of Decision Theory is reasoning that keeps the available actions, uncertain states, consequences, preferences and decision rule connected. A recommendation should remain understandable when a probability changes, information arrives or an action becomes unavailable.
This subject root continues The Gold Standard of Education | Learning That Transfers into Life under How X Works. It develops the formal comparisons behind the existing Decision Making and Decision Quality guides, with routes into Probability, Expected Value, Bayesian Inference, Game Theory and Operations Research.
The exhibition and benefit examples are fictional teaching models. Their probabilities and utilities are stipulated to make the calculations inspectable. They are not measured forecasts, survey ratings or recommendations for an actual event. Each conclusion belongs to the stated model and criterion.
Full chapter index
Chapters 1–4 · Frame the formal problem
Chapters 5–8 · Compare uncertain prospects
Chapters 9–12 · Check beliefs and information
Chapters 13–16 · Evaluate policies and criteria
Chapters 17–20 · Transfer and review
CHAPTER 1 OF 20 · Frame the formal problem
1. Start with a choice whose result is still uncertain
A school exhibition can use a courtyard or a hall. The courtyard is more attractive in dry weather, while the hall performs better in rain. The weather is not yet known, and the booking must be made before it is observed.
Decision Theory studies how choices relate to beliefs, possible consequences and preferences. Formal models make those relationships explicit so that a recommendation can be inspected, recalculated and challenged.
The gold standard is a bounded conclusion: given these feasible actions, probabilities, utilities and decision rule, this option or policy follows. That is different from a guarantee that the chosen action will produce the best realised outcome.
This root develops the formal reasoning behind such conclusions. The existing Decision Making and Decision Quality guides retain their practical roles. Here, original calculations connect them with Probability, Bayesian Inference, Expected Value, Game Theory and the value of information.
Continue into: How Decision-Making Works · The Gold Standard Of Decision Quality.
CHAPTER 2 OF 20 · Frame the formal problem
2. Distinguish normative, descriptive and practical questions
A normative theory asks which relationships between beliefs, preferences and choices should be satisfied under a proposed account of rational choice. It needs an argument for those requirements, rather than only a record of what people happen to choose.
A descriptive theory asks how people actually choose. Evaluating that account requires evidence about behaviour. Observing that someone departs from one mathematical model does not by itself explain the psychological process behind the departure.
Practical decision analysis helps a person or group structure a particular choice. It can use formal models while also investigating objectives, feasible options, implementation and evidence. Its recommendation remains conditional on that work.
Keep these questions separate when reading a claim. A useful calculation is not automatically a universal theory of behaviour or a complete ethical justification. This guide concentrates on inspectable formal comparisons and on the assumptions needed to apply them.
Continue into: The Gold Standard Of Decision Making.
CHAPTER 3 OF 20 · Frame the formal problem
3. Distinguish actions, states, consequences and policies
An action is something the decision-maker chooses. A state describes a relevant uncertain condition. A consequence is what results from an action in a state, including the features the decision model values.
In the exhibition example, the action is the venue and the state is dry or wet weather. These two states are stipulated as mutually exclusive and exhaustive for the fictional calculation. They are not a claim that every real weather question has only two relevant possibilities.
A policy is a rule for choosing after permitted information becomes available. “Use the hall if the forecast signal is wet; otherwise use the courtyard” is a policy. It differs from committing to one venue before any signal arrives.
Timing therefore belongs in the definition. List which observations are available at each decision point and which actions remain possible. A plan that requires knowledge the decision-maker will not yet possess is not an implementable policy.
Continue into: The Gold Standard Of Problem Framing.
CHAPTER 4 OF 20 · Frame the formal problem
4. Remove infeasible options and inspect dominance
Before calculating an average, establish the feasible set. A venue unavailable on the event date is not an option. A required accessibility condition is part of feasibility rather than something automatically traded away through a high numerical score.
For the fictional model, assume both the courtyard and hall meet the required conditions. Their supplied utility values are 100 and 20 for the courtyard in dry and wet weather, and 70 and 60 for the hall respectively. These are stipulated utilities, not measured satisfaction scores.
A third proposed venue has utilities 60 when dry and 10 when wet. The hall gives a higher utility in both states, so it strictly dominates that third venue under this model.
Dominance can simplify a comparison without choosing probabilities. It does not establish that the hall dominates the courtyard: the courtyard is better in the dry state, while the hall is better in the wet state. That remaining trade-off needs a further criterion.
| Action | Utility if dry | Utility if wet |
|---|---|---|
| Courtyard | 100 | 20 |
| Hall | 70 | 60 |
| Third venue | 60 | 10 |
Continue into: The Gold Standard of Operations Research.
CHAPTER 5 OF 20 · Compare uncertain prospects
5. State what the probabilities represent
Assume the model assigns probability 0.3 to wet weather and 0.7 to dry weather. Those numbers express the uncertainty used for this teaching decision. They are supplied assumptions, not a forecast for an actual place or date.
Probabilities can be informed by observations, a forecasting model, comparable cases and expert assessment. Their quality depends on that basis and on whether the events were defined consistently. A precise decimal does not supply its own evidence.
Distinguish a probability from a utility. Probability concerns how likely a state or consequence is; utility represents how an outcome is evaluated within the preference model. Raising a probability because an outcome is desirable mixes those different jobs.
The exhibition’s weather state is treated as independent of the venue choice. In another problem, an action might change the outcome distribution. Then the analysis needs the appropriate action-dependent consequences or probabilities, rather than reusing the same table without justification.
Continue into: The Gold Standard of Probability.
CHAPTER 6 OF 20 · Compare uncertain prospects
6. Calculate expected utility without predicting one outcome
For a finite state model, expected utility for action a is the sum over states of probability times the utility of that action’s consequence. Every relevant state must be included once with its assigned weight.
For the courtyard, the calculation is 0.7 × 100 + 0.3 × 20 = 76. For the hall, it is 0.7 × 70 + 0.3 × 60 = 67. Maximising expected utility selects the courtyard under the supplied assumptions.
Neither venue produces a utility of 76 in the two-state table. The courtyard produces 100 if dry and 20 if wet. The expected value summarises the distribution and preference model; it is not an intermediate outcome promised for the event.
A wet result does not retroactively alter the probabilities and options that existed at the booking time. Review both the decision model and what actually happened, rather than treating one realised outcome as proof of the whole method.
CHAPTER 7 OF 20 · Compare uncertain prospects
7. Find the probability threshold that changes the choice
Let p be the probability of wet weather while keeping the utilities fixed. The courtyard’s expected utility is 100(1 − p) + 20p = 100 − 80p. The hall’s is 70(1 − p) + 60p = 70 − 10p.
The courtyard is preferred by expected utility when 100 − 80p > 70 − 10p. Rearranging gives p < 3/7, approximately 0.429. Above that threshold the hall is preferred; at the threshold the expected utilities are equal.
This is a sensitivity result, not a new forecast. It tells us how much a particular assumption could move before the model’s recommendation changes. If the plausible probability range straddles the threshold, that uncertainty matters to the action comparison.
Keep the utility assumptions visible too. If the consequences of wet weather were reassessed, the threshold would change. Sensitivity analysis should examine the quantities capable of changing the choice, not only whichever input is easiest to vary.
Continue into: How Intelligence Works | Sensitivity Analysis — How a Mind Finds Which Assumptions Matter Most to the Outcome.
CHAPTER 8 OF 20 · Compare uncertain prospects
8. Do not average rankings as if they were utility intervals
An ordering says one outcome is preferred to another. Expected-utility calculations require a richer representation of preferences over uncertain prospects. Arbitrary star ratings or rank numbers do not automatically provide that representation.
A simple illustration compares a certain benefit index of 6 with equal chances of indices 0 and 10. Averaging the indices favours the certain 6 over the uncertain average of 5. Squaring all three numbers preserves their order but makes the uncertain average 50 exceed the certain value of 36.
The reversal shows why ordering alone does not justify averaging. For an expected-utility representation, positive affine changes u′ = a u + b, with a positive, preserve expected-utility rankings. Arbitrary increasing transformations need not.
The exhibition values are supplied as utilities for the teaching model precisely so the calculation is defined. Obtaining a defensible utility representation for a real decision is an additional task, not something created by putting numbers in a spreadsheet.
Continue into: What Is Value? | Why Some Outcomes Matter More Than Others.
CHAPTER 9 OF 20 · Check beliefs and information
9. Treat representation assumptions as part of the theory
Expected-utility representation theorems connect patterns of preference with a numerical form. In a standard lottery setting, familiar conditions include completeness, transitivity, continuity and independence, with the formal domain and hypotheses specified.
Completeness concerns whether prospects can be compared; transitivity concerns consistency across comparisons. Independence constrains how a comparison behaves when the same additional prospect is mixed into both alternatives. These are substantive assumptions about preference relationships.
A representation theorem does not establish the right objective for a school, prove a forecast accurate or supply missing consequences. It shows a mathematical relationship when its stated conditions hold.
The assumptions can also be questioned or fail to describe a particular person’s choices. When evaluating a model, identify whether the concern is a calculation error, an unsupported input, an unsuitable representation or disagreement with the proposed norm. These are different criticisms and call for different responses.
Continue into: The Gold Standard Of Judgment.
CHAPTER 10 OF 20 · Check beliefs and information
10. Separate a consequence measure from preferences about uncertainty
Expected utility is not necessarily the expected amount of a resource or other consequence measure. The preference model determines how each consequence enters the calculation before probabilities are applied.
For example, consider a fictional nonnegative benefit amount X. Under u(X) = √X, a certain amount of 9 gives utility 3. Equal chances of amounts 0 and 20 have expected amount 10, but expected utility is √20/2, approximately 2.236.
This preference representation favours the certain amount even though its expected amount is lower. The example illustrates curvature in the utility function; it does not prescribe how every person should evaluate uncertainty.
Constraints and omitted consequences also matter. A model that records only one benefit amount can miss effort, timing, distribution or a required condition. Specify what an outcome contains before interpreting its utility or calling an action optimal.
CHAPTER 11 OF 20 · Check beliefs and information
11. Update beliefs using the signal’s actual reliability
Suppose a forecast gives a wet signal S. In the fictional model, P(S|wet) = 0.8 and P(S|dry) = 0.2. The prior wet probability remains 0.3.
The probability of a wet signal is 0.8 × 0.3 + 0.2 × 0.7 = 0.38. Bayes’ rule gives P(wet|S) = 0.24/0.38 = 12/19, approximately 0.632.
If no wet signal appears, its probability is 0.62 and the posterior wet probability is 0.06/0.62 = 3/31, approximately 0.097. Both posterior probabilities come from the joint model, including the prior.
The signal’s 0.8 sensitivity is therefore not the probability of wet weather after a wet signal. Confusing P(S|wet) with P(wet|S) skips the base-rate calculation and can change the action chosen. The Bayesian Inference guide provides the deeper route into that distinction.
| Observed signal | Signal probability | Posterior wet probability | Expected-utility venue choice |
|---|---|---|---|
| Wet signal S | 0.38 | 12/19 ≈ 0.632 | Hall |
| No wet signal | 0.62 | 3/31 ≈ 0.097 | Courtyard |
Continue into: How Bayesian Inference Works.
CHAPTER 12 OF 20 · Check beliefs and information
12. Calculate the value of perfect information
Imagine a hypothetical free observation that reveals the weather state before booking, while leaving both venues available. Choose the courtyard if dry and the hall if wet. The expected utility of that informed policy is 0.7 × 100 + 0.3 × 60 = 88.
The best uninformed action had expected utility 76. The expected value of perfect information is therefore 88 − 76 = 12 utility units in this model.
This is an upper benchmark for information about the same state under the same feasible actions, consequence model and decision criterion. It is not a claim that a perfect forecast exists or that acquiring it has no operational consequences.
Notice the order of operations. With information, choose the better action inside each known state and then average. Without information, average each fixed action first and then choose the better average. Reversing that order silently grants knowledge before it is available.
Continue into: How Intelligence Works | Value of Information — How a Mind Decides Which Missing Fact Is Worth Obtaining Next.
CHAPTER 13 OF 20 · Evaluate policies and criteria
13. Calculate the value of an imperfect forecast
Use the earlier signal with wet-signal probability 0.38. After S, the posterior wet probability is approximately 0.632, above the 3/7 threshold, so the model selects the hall. Without S, the posterior is approximately 0.097, so it selects the courtyard.
Evaluate that policy before the signal arrives. The four joint branches are wet with S, dry with S, wet without S and dry without S. Their probabilities are 0.24, 0.14, 0.06 and 0.56.
The expected utility is 0.24 × 60 + 0.14 × 70 + 0.06 × 20 + 0.56 × 100 = 81.4. The expected value of this sample information is 81.4 − 76 = 5.4 utility units.
The signal is useful because it can support a different action at the booking time. Its value is below the perfect-information benchmark of 12. These are values of the specified information-and-response policies, not measures of whether a forecast sounds confident.
| Joint event | Probability | Policy action | Utility contribution |
|---|---|---|---|
| Wet, with wet signal | 0.24 | Hall | 0.24 × 60 = 14.4 |
| Dry, with wet signal | 0.14 | Hall | 0.14 × 70 = 9.8 |
| Wet, without wet signal | 0.06 | Courtyard | 0.06 × 20 = 1.2 |
| Dry, without wet signal | 0.56 | Courtyard | 0.56 × 100 = 56 |
Continue into: How Intelligence Works | Value of Information — How a Mind Decides Which Missing Fact Is Worth Obtaining Next.
CHAPTER 14 OF 20 · Evaluate policies and criteria
14. Include acquisition cost, delay and the ability to act
The 5.4-unit improvement is a gross value before information-acquisition consequences. If obtaining the signal adds a fixed disutility of 3 in the same utility scale to every final branch, the net value is 2.4 and the informed policy’s expected utility is 78.4.
If that fixed disutility is 6, the informed policy’s net expected utility is 75.4, below the uninformed 76. This calculation assumes the cost is represented additively in the supplied utility model. A money price cannot simply be subtracted from utility points without a justified conversion.
Timing can change the comparison more fundamentally. If the useful signal arrives only after the venue is irrevocably booked, it cannot enable this switching policy. Its value for that venue decision can be zero even if it predicts weather well.
Free information that may be ignored cannot reduce the best attainable expected utility under an unchanged model and feasible policy set. Costs, changed circumstances, compulsory responses or mistaken use of information can invalidate those premises.
Continue into: How The World Works | Opportunity Cost — The Future You Give Up When You Choose · How Information Works | Decision-Relevance Threshold — When Should You Stop Gathering More Information?.
CHAPTER 15 OF 20 · Evaluate policies and criteria
15. Draw the decision tree in the order events can happen
A decision tree distinguishes choice nodes from chance nodes. At a chance node, compute the probability-weighted value of its branches. At a choice node, select the available branch according to the stipulated criterion.
For the exhibition, the first choice is whether to obtain the signal. If obtained, the signal is observed before the venue choice. The actual weather outcome then determines the consequence. A branch must use probabilities conditional on what is already known at that point.
A tree that chooses the courtyard in dry weather and the hall in rain without an earlier observation describes perfect information, not an uninformed booking. Drawing that tree and calling it a realistic plan would overstate available capability.
Check that branch probabilities form the intended partition and that acquisition costs are included exactly once. A large diagram is not inherently a better model; a small tree with correct timing is more useful than a detailed one whose decisions use future knowledge.
Continue into: How Information Works | How Signals Become Meaning, Knowledge and Action.
CHAPTER 16 OF 20 · Evaluate policies and criteria
16. Compare expected utility with maximin and regret
Different criteria can select different actions. Under the exhibition utility table, the courtyard’s worst outcome is 20 and the hall’s is 60. A maximin criterion, which chooses the action with the highest minimum utility, selects the hall.
Regret compares an action with the best available utility in the same state. The best dry value is 100 and the best wet value is 60. Courtyard regret is therefore 0 when dry and 40 when wet; hall regret is 30 when dry and 0 when wet.
A minimax-regret criterion selects the hall because its maximum regret is 30 rather than 40. Expected utility at wet probability 0.3 selects the courtyard. This disagreement reflects different evaluation rules, not an arithmetic contradiction.
With fixed state probabilities, minimising expected regret over the same action set is equivalent to maximising expected utility: the expected statewise-best benchmark is the same for every action. Minimax regret is a different criterion and should not be confused with that probability-weighted calculation.
| Criterion | Courtyard | Hall | Selected action |
|---|---|---|---|
| Expected utility, wet probability 0.3 | 76 | 67 | Courtyard |
| Minimum state utility | 20 | 60 | Hall under maximin |
| Maximum statewise regret | 40 | 30 | Hall under minimax regret |
| Expected regret, wet probability 0.3 | 12 | 21 | Courtyard when minimising expected regret |
Continue into: The Gold Standard Of Risk Management.
CHAPTER 17 OF 20 · Transfer and review
17. Represent ambiguity instead of inventing a confident probability
Sometimes the evidence does not support one well-established probability distribution. A range of plausible models or probabilities can be more informative than presenting a single value without a defensible basis.
If the exhibition’s wet probability is believed to lie between 0.2 and 0.5, the range crosses the action threshold of 3/7. The expected-utility preference is therefore not constant over that interval.
One possible robust comparison evaluates each action at its worst expected utility within the specified probability range. The courtyard’s lowest value there is 60, while the hall’s is 65. Under that stated worst-case-over-models criterion, the hall is preferred.
The probability range does not select that criterion by itself. Different attitudes towards ambiguity and different constraints can lead to other approaches. Report the model set and decision rule separately so that a reader can see which assumption drives the conclusion.
Continue into: How Uncertainty Works | From Known Limits and Unknowns to Ranges, Scenarios, Sensitivity, Decisions and Correction.
CHAPTER 18 OF 20 · Transfer and review
18. Distinguish learning, flexibility and strategic interaction
A reversible action can preserve later choices, but flexibility has value only through the consequences and future decisions it enables. A staged plan should represent what can be learned, what can change and what becomes unavailable after commitment.
Exploration and exploitation introduce another distinction. An action can produce an immediate benefit and also information useful for later decisions. The appropriate comparison includes the relevant future horizon rather than treating information as valuable without a possible use.
Game Theory becomes important when consequences depend on other decision-makers whose actions respond to incentives and expectations. Treating such a response as fixed weather-like uncertainty can omit the strategic relationship.
Operations Research connects decisions with feasibility, scheduling and resource allocation. Decision Theory supplies formal ways to compare uncertain prospects, but the action set and consequence model still depend on the actual problem. The neighbouring roots keep those different tasks connected.
Continue into: The Gold Standard of Game Theory · How Intelligence Works | Exploration–Exploitation Arbitration — How Intelligence Decides When to Use What Already Works and When to Spend Opportunity Discovering Something Better · How Intelligence Works | Reversibility Reasoning — How Intelligence Chooses Actions That Preserve a Way Back When Uncertainty Is High.
CHAPTER 19 OF 20 · Transfer and review
19. Keep group values and constraints visible
A utility function representing one decision-maker does not automatically measure everyone’s wellbeing on a common scale. Two people can rank outcomes similarly while their numerical utility representations use different origins and units.
Adding their utility numbers therefore requires further assumptions or an agreed social decision model. It is not justified merely because both columns contain numbers between zero and one hundred.
Group decisions also include responsibilities, participation and requirements that may belong outside an unrestricted trade-off calculation. In the exhibition example, required accessibility was included in feasibility before comparing the remaining outcomes.
Use a model to expose disagreement: a disputed probability, a different evaluation of consequences, an omitted action or an unacceptable constraint violation. An apparently precise total can conceal those disagreements if they were never made explicit.
Continue into: The Gold Standard Of Decision Quality.
CHAPTER 20 OF 20 · Transfer and review
20. Check the reasoning when the outcome disappoints
Suppose the courtyard is booked under the original expected-utility model and rain occurs. The realised utility is 20, below the hall’s wet-weather value of 60. That single outcome does not establish that choosing the courtyard was an arithmetic error.
Review what was knowable when the decision was made. Was the wet probability defensible? Were both venues actually feasible? Did the utilities represent the intended preferences? Was a useful forecast available in time, and had its acquisition consequences been considered?
Now change the evidence. If wet probability was actually assessed as 0.6 before booking, expected utility would have favoured the hall. If the signal arrived after booking, using its posterior to justify an earlier switch would be a timing error.
The gold standard is independent use of the model with its limits intact. Recalculate when evidence or feasible actions change, distinguish a probability disagreement from a value disagreement and retain the original reasoning for a fair review after the outcome.
Continue into: The Gold Standard of Education | Learning That Transfers into Life · The Gold Standard Of Forecasting.
Contents · Previous chapter · Continue to the article directory
Connected specialist article directory
Choose the branch matching the part of your model that needs attention. A probability question, a disagreement about values and a feasibility problem require different kinds of work, even when they appear in the same decision.
Begin with decisions and formal probability
| Existing article | Use this route to |
|---|---|
| The Gold Standard Of Decision Making | Connect a formal recommendation with practical framing, action and review. |
| The Gold Standard Of Decision Quality | Evaluate the decision process separately from realised luck. |
| How Decision-Making Works | Follow the general choosing-under-uncertainty mechanism. |
| The Gold Standard of Probability | Keep event definitions, uncertainty and probability models precise. |
| How Expected Value Works | From Probability-Weighted Possibilities to Long-Run Averages, Decisions, Risk, Utility, Games and Better Statistical Thinking | Distinguish weighted expectations from modes, medians and promised outcomes. |
| How Bayesian Inference Works | Update beliefs through priors, likelihoods and posterior evidence. |
| How Statistical Inference Works | From Samples and Estimands to Models, Estimates, Tests, Uncertainty, Assumptions and Defensible Conclusions | Connect estimates and uncertainty with the assumptions supporting them. |
Check the frame, evidence and values
| Existing article | Use this route to |
|---|---|
| The Gold Standard Of Problem Framing | Establish which choice is actually being compared. |
| The Gold Standard Of Evidence-Based Thinking | Ask which observations support or challenge the model’s inputs. |
| The Gold Standard Of Forecasting | Evaluate statements about uncertain future outcomes. |
| The Gold Standard Of Risk Management | Connect consequence distributions with exposure and required protections. |
| The Gold Standard Of Judgment | Interpret evidence and trade-offs in their actual context. |
| What Is Value? | Why Some Outcomes Matter More Than Others | Examine what it means for one consequence to matter more than another. |
| How Uncertainty Works | From Known Limits and Unknowns to Ranges, Scenarios, Sensitivity, Decisions and Correction | Distinguish uncertainty descriptions, scenarios and model limits. |
Investigate information and changing assumptions
| Existing article | Use this route to |
|---|---|
| How Intelligence Works | Value of Information — How a Mind Decides Which Missing Fact Is Worth Obtaining Next | Ask which missing fact can improve an available decision. |
| How Information Works | Decision-Relevance Threshold — When Should You Stop Gathering More Information? | Distinguish additional facts from decision-changing information. |
| How Information Works | How Signals Become Meaning, Knowledge and Action | Follow the relationship between signals, meaning and action. |
| How Intelligence Works | Sensitivity Analysis — How a Mind Finds Which Assumptions Matter Most to the Outcome | Find the assumptions capable of changing the preferred option. |
| How The World Works | Opportunity Cost — The Future You Give Up When You Choose | Include what is given up when a route or investigation is chosen. |
| How Intelligence Works | Stopping Rules — How a Mind Knows When to Stop Searching, Testing or Thinking and Commit | Examine when further searching or testing should end. |
Extend from one choice to policies and interactions
Transfer the reasoning into independent learning
| Existing article | Use this route to |
|---|---|
| How to Grow Up Properly | Learn to Live With Uncertainty Without Freezing | Build practical agency while acknowledging incomplete knowledge. |
| How to Grow Up Properly | Learn to Choose When Every Choice Costs Something | Keep trade-offs visible in everyday decisions. |
| The Gold Standard of Education | Learning That Transfers into Life | Test whether the reasoning remains useful after the case changes. |
| How X Works Hub | Return to the wider mechanism and subject collection. |
A practical quality check
A reviewable decision model states the action set, relevant uncertain events, consequence mapping, probability basis, preference representation and criterion. It identifies when observations arrive and which decisions remain available after them.
Recalculate the recommendation under the assumptions that matter. If a plausible probability range crosses the decision threshold, report that sensitivity. If a preferred action is infeasible, return to the available options rather than treating the infeasible maximum as a plan.
Keep the original decision record when the outcome arrives. Compare forecasts with evidence and examine implementation, but do not silently replace the earlier information set with hindsight. The objective is a better model and a better next choice, not a story that makes every past action look inevitable.
Questions readers ask
Is Decision Theory the same as Decision Making?
Decision Theory studies formal relationships between beliefs, preferences and choices. Decision Making is the broader practical process of choosing and acting. A useful analysis connects them while retaining the assumptions of the mathematical model.
Does highest expected utility guarantee the best result?
No. It selects a prospect under a specified probability and preference model. A lower-utility outcome may still occur. Evaluate the reasoning and the realised outcome separately.
Can I use any scoring scale in an expected-utility calculation?
A scale that only preserves ranking is insufficient by itself. Expected-utility representation requires appropriate structure in preferences over uncertain prospects. Arbitrary rescaling can change the result, as the worked index example shows.
Is more accurate information always worth acquiring?
Its value depends on the decision, available responses, timing and acquisition consequences. A signal can predict well yet arrive too late to change the relevant action. Compare implementable policies and include the consequences of obtaining the signal.
Why can maximin and expected utility disagree?
They apply different criteria. Maximin compares worst state utilities. Expected utility weights all included states by probabilities. State the rule being used before describing one recommendation as superior.
What should I check first when two analyses disagree?
Compare the action sets, event definitions, probabilities, outcome descriptions, utilities, timing and criterion. A difference in any of those can explain the recommendations without an arithmetic error.
Sources and further learning
Stanford Encyclopedia of Philosophy · Decision Theory — Scholarly discussion of preferences, utility representation, formal models and their challenges.
Stanford Encyclopedia of Philosophy · Normative Theories of Rational Choice: Expected Utility — The expected-utility account, its assumptions and philosophical questions.
MIT OpenCourseWare · Decision Analysis and Information Value — A university study route into decision trees and the perfect-information benchmark.
MIT OpenCourseWare · Expected Value of Sample Information — Further study of imperfect information and decision improvement.
MIT OpenCourseWare · Expected Value of Perfect Information — Further study of the information benchmark and its interpretation.
All exhibition probabilities, utility values, forecast properties and acquisition disutilities in this guide are fictional teaching inputs. The benefit-index and square-root examples are also original illustrations. Their calculations demonstrate the stated models; the references provide routes into the surrounding theory.
Continue into The Gold Standard of Education | Learning That Transfers into Life or The Gold Standard of Decision Quality. For the next uncertain choice, write down what can be done, what is unknown and which observation could change an available action.