Standard Gamble

Standard Gamble: Concept Architecture

Orientation and learning roadmap

The standard gamble elicits the strength of preference for a health state through a choice under uncertainty. This page explains the choice task, indifference equation, anchors, administration, interpretation, and adaptations for difficult states. It then connects the elicited value to QALYs while examining risk attitude, respondent perspective, bias, validation, and reproducible reporting.

The task compares certainty with a risky alternative

A respondent chooses between living in a specified health state (H) with certainty and accepting a gamble with two possible outcomes. The gamble produces a better anchor (B) with probability (p) and a worse anchor (W) with probability (1-p). The probability is varied until the respondent reports indifference between the certain state and the gamble.

The indifference equation defines the elicited value

Under expected-utility theory, indifference means that the utility of the certain state equals the probability-weighted utility of the gamble outcomes. The fundamental relationship is shown below.

[ U(H)=p^U(B)+(1-p^)U(W), ]

where (p^*) is the indifference probability. The value is conditional on the states, duration, framing, respondent, administration, and assumptions used in the task.

Full health and death create the familiar zero-to-one scale

For a health state considered better than death, the conventional anchors are full health and death. Normalizing (U(\text{full health})=1) and (U(\text{death})=0) simplifies the indifference equation. The utility assigned to (H) is then:

[ U(H)=p^*. ]

A value of 0.80 means the respondent is indifferent between certain (H) and the specified gamble with an 80% chance of full health and a 20% chance of death. It does not mean the respondent believes that (H) contains 80% of full health or predicts an 80% clinical success rate.

A worked valuation example

Suppose a respondent is indifferent between living for the stated duration in (H) and a treatment gamble that gives full health with probability 0.78 or immediate death with probability 0.22. With the conventional anchors, the standard-gamble utility is 0.78. The arithmetic is:

[ U(H)=0.78(1)+0.22(0)=0.78. ]

If the respondent instead prefers the certain state at (p=0.80) but prefers the gamble at (p=0.76), the indifference point is bracketed between 0.76 and 0.80. Reporting 0.78 would require an explicit midpoint or interpolation rule rather than an observed exact indifference.

General anchors require the full weighted value

Some tasks use outcomes other than full health and death. In that case, the certain state's value remains the expected utility of the two anchors at indifference. If anchor utilities are already known on the desired scale, the calculation is:

[ U(H)=p^U(B)+(1-p^)U(W). ]

The shortcut (U(H)=p^*) is valid only when the better and worse anchors are normalized to 1 and 0. Failing to rescale a chained or alternative-anchor gamble can place the result on the wrong utility scale.

For example, if (p^*=0.40), (U(B)=0.90), and (U(W)=0.20), then (U(H)=0.40(0.90)+0.60(0.20)=0.48). The value is 0.48, not the indifference probability 0.40, because neither anchor is normalized to the conventional endpoint pair.

The health-state description defines what is valued

The description should specify symptoms, functioning, pain, emotional effects, treatment burden, and other attributes required to understand (H). It should also state duration, prognosis, adaptation assumptions, and what happens after the described period. Missing or ambiguous information invites respondents to supply different unstated scenarios and weakens comparability.

Duration is part of the elicitation

Preferences can change when the same state is described as temporary, chronic, improving, deteriorating, or lasting for a different number of years. A standard-gamble value should therefore retain the duration and temporal profile used in the task. Treating a value elicited for one duration as universally duration-independent is an additional assumption.

Administration searches for an indifference probability

An interviewer or digital instrument presents a sequence of probabilities and records the respondent's choices. Titration, bisection, ping-pong, or another search procedure narrows the interval containing indifference. The starting probability, step size, stopping rule, and handling of inconsistent responses can affect the reported value.

Probability presentation affects comprehension

Probabilities may be shown as percentages, frequencies, icon arrays, spinners, or verbal descriptions. Equivalent numerical risks are not always perceived equivalently, especially when numeracy is limited or outcomes are emotionally salient. A validated presentation should keep the denominator, time frame, and wording consistent and should check that respondents understand both branches of the gamble.

Standard gamble differs from time trade-off

Time trade-off asks a respondent to exchange length of life for better health, while standard gamble asks the respondent to accept outcome risk. Both are choice-based direct valuation methods, but they engage different attitudes toward time and uncertainty. Their values need not agree, and one should not be substituted for the other without documenting the methodological change.

Standard gamble differs from a rating scale

A rating scale places a health state on a visual or numerical continuum without requiring a risky choice. It is usually easier to administer but does not directly use an expected-utility indifference condition. Rating-scale values may be useful for description or ordering, yet they are not automatically interchangeable with standard-gamble utilities.

Standard gamble differs from willingness to pay

Willingness to pay values a change in monetary terms and is constrained by income and ability to pay. Standard gamble varies probability between health or survival outcomes and returns a utility anchored by those outcomes. Both involve choices, but they answer different valuation questions and use different scales.

Expected-utility theory provides the formal foundation

The standard gamble follows directly from expected-utility representation when preferences over lotteries satisfy the relevant axioms. Under that model, outcomes receive utilities and lottery values are probability-weighted averages. The task's theoretical appeal does not establish that every respondent behaves according to those axioms in practice.

Risk attitude can influence the measured value

The choice involves risk, so aversion to or attraction toward the gamble can affect the indifference probability. A risk-averse respondent may require a higher chance of the better outcome before accepting the gamble, which can raise the elicited value for the certain state. Differences between methods cannot be attributed solely to risk attitude because framing, time preference, probability weighting, loss aversion, and task comprehension may also contribute.

Probability weighting can violate linear weighting

Respondents may overweight small probabilities or underweight moderate and large probabilities rather than treating (p) as a linear decision weight. If so, the observed indifference does not map cleanly to expected utility by (U(H)=p^*). The standard calculation can still be reported as the protocol's elicited value, but its behavioral interpretation requires caution.

The worse outcome makes the task ethically and emotionally demanding

Immediate death is a strong anchor that can cause distress, refusal, lexicographic responses, or unwillingness to consider any risk. Some respondents may always choose the certain state or always reject death regardless of probability. Ethical review, advance explanation, the ability to pause or withdraw, and a prespecified rule for non-trading responses are essential.

States worse than death need an adapted task

The conventional better-than-death gamble cannot produce a negative value because (p^) lies between zero and one. States judged worse than death require an alternative arrangement or chaining procedure that places the assessed state and death differently in the choices. Formulas vary across protocols, so the exact task and transformation must be reported rather than applying (U(H)=p^) mechanically.

Temporary states need explicit end conditions

A temporary-state standard gamble must specify what happens after the temporary period, such as return to full health, continuation in another state, or death. Otherwise respondents may value different imagined trajectories rather than the stated temporary condition. The resulting value should not be treated as equivalent to a chronic-state utility without a justified transformation.

Chaining expands the range but adds error

Chaining values one state relative to intermediate anchors and then rescales it using separately valued anchor utilities. It can make extreme choices easier, but uncertainty and inconsistency from each stage propagate into the final estimate. The chain structure, direction, anchor values, and rescaling equation should remain auditable.

Respondent perspective changes the estimand

Patients, caregivers, clinicians, and members of the general public can value the same description differently because of experience, adaptation, knowledge, and perspective. The relevant respondent group depends on the decision framework and research question. Mixing perspectives within one value set without design or adjustment produces an unclear estimand.

Experience and adaptation can shift preferences

People living with a condition may anticipate or experience adaptation that members of the public do not incorporate when imagining the state. Conversely, patients may know burdens or risks that brief descriptions omit. Neither perspective is universally correct, so the source population and information given to respondents should be reported and interpreted against the policy context.

Framing and ordering can change responses

Presenting the gamble as survival versus mortality, gain versus loss, treatment versus no treatment, or cure versus failure can influence choice even when probabilities are equivalent. The order of states, initial probability, visual emphasis, and interviewer prompts can also create anchoring or sequence effects. Standardized scripts and randomized design elements can reduce but not eliminate these influences.

Inconsistent choices need a prespecified response

A respondent may switch back and forth as probabilities change, violate dominance, or report indifference across a wide range. These patterns can reflect misunderstanding, imprecision, genuine preference instability, or search design. The study should define comprehension checks, repetition rules, exclusion criteria, and whether interval-censored preferences will be retained.

Precision should not exceed the elicitation grid

If probabilities are offered in increments of 0.05, the task does not directly support a value precise to three decimal places. Midpoint interpolation can summarize a bracket, but it adds an assumption about the location of indifference. Reported decimal places should reflect the search resolution, response consistency, and statistical estimation method.

A direct utility can contribute to QALY calculation

If utility is assumed constant at (u) for (T) years and QALYs are linear in time, the contribution is (uT). An illustrative utility of 0.78 maintained for five years gives (0.78\times5=3.90) undiscounted QALYs. This calculation adds assumptions about constant quality, survival, duration, and QALY structure that the standard-gamble task alone does not test.

[ \operatorname{QALYs}=uT. ]

For a changing trajectory, utilities should be matched to the applicable time intervals and integrated or summed using a documented timing convention. Mortality, transitions, discounting, and treatment effects must be handled in the economic model rather than inferred from one elicited value.

Direct elicitation and preference-based instruments serve different roles

Standard gamble can value a particular described state directly, whereas instruments such as EQ-5D classify health and apply values estimated from a separate valuation study. Direct elicitation may capture condition-specific details but can reduce comparability across studies. A preference-based measure offers standardization, yet may omit dimensions important to a particular condition.

Individual values and population tariffs are not interchangeable

One respondent's indifference probability represents that person's elicited preference under the task. Economic evaluation commonly needs a mean or modeled value for a defined population, estimated with sampling, quality control, and uncertainty analysis. Applying an individual value as though it were a representative tariff can create false precision and poor generalizability.

Sampling uncertainty should accompany the estimate

A valuation study should report sample design, response and completion rates, exclusions, missing data, distributional shape, and uncertainty around summary values. Standard-gamble observations can cluster at anchors or search-grid points and may not follow a normal distribution. Statistical methods should match the bounded, censored, repeated, or hierarchical structure of the data.

Reliability and validity require several checks

Test-retest agreement assesses stability, while dominance and monotonicity checks assess internal ordering. Construct validity can compare expected patterns across severity, related measures, or known groups without treating another method as an error-free gold standard. Debriefing and qualitative evidence help determine whether respondents understood the intended choice.

Digital administration needs functional validation

An electronic task should reproduce the approved state text, probabilities, branching logic, visuals, and stopping rule on supported devices. The system should prevent impossible probabilities, log every offer and response, preserve partial completions, and distinguish a true indifference from an algorithmic midpoint. Accessibility, keyboard navigation, screen-reader meaning, language versions, and data privacy also require testing.

A disciplined standard-gamble workflow

Valid elicitation begins with a clear estimand and ends with transparent scoring and uncertainty. The procedure should not be treated as a simple probability question detached from its anchors and behavioral demands. The following sequence provides a reproducible structure.

1. Define the estimand. Specify the health state, duration, perspective, respondent population, and intended use.
2. Write the states. Describe the certain state and both gamble outcomes with balanced, comprehensible information.
3. Choose the anchors. Justify full health, death, or alternative outcomes and define their scale values.
4. Design the search. Prespecify starting points, probability steps, stopping rules, interpolation, and repetition.
5. Test comprehension. Confirm understanding of probabilities, exclusivity, timing, and consequences before scoring.
6. Administer consistently. Standardize wording, visuals, ordering, interviewer behavior, and device presentation.
7. Record the path. Preserve every probability offered, choice, reversal, indifference response, and task duration.
8. Apply the equation. Use the correct anchor utilities and transformation for the exact task.
9. Assess quality. Examine dominance, consistency, non-trading, missingness, reliability, and qualitative feedback.
10. Estimate uncertainty. Report the distribution, sampling method, model, interval estimates, and sensitivity to exclusions.
11. Connect cautiously. State the additional assumptions required when the value enters QALY or decision models.
12. Report reproducibly. Publish the protocol, materials, scoring algorithm, sample, deviations, and limitations.

Common errors and safeguards

Standard-gamble values can look simple even when the underlying task is poorly specified or understood. Strong safeguards retain the anchors, duration, respondent perspective, and search path alongside the final number. They also prevent behavioral responses to risk from being described as pure health-state preference.

• Reporting (p^*) without naming the better and worse outcomes leaves the utility scale undefined.
• Using (U(H)=p^*) with anchors not valued at 1 and 0 produces the wrong score.
• Omitting duration allows respondents to value different imagined health trajectories.
• Describing 0.80 as “80% health” misinterprets an indifference probability.
• Treating the gamble probability as a clinical event probability confuses valuation with prognosis.
• Ignoring risk attitude and probability weighting overstates the purity of the health-state utility.
• Forcing a death gamble on an unwilling respondent can create distress and invalid responses.
• Applying a better-than-death formula to a state judged worse than death prevents valid negative values.
• Averaging patient and public values without a defined estimand obscures perspective.
• Reporting more precision than the probability grid supports creates false accuracy.
• Excluding inconsistent or non-trading respondents after seeing results can bias the sample.
• Multiplying utility by time without stating survival and linearity assumptions overstates what the elicitation established.
• Treating a single direct value as a population tariff ignores sampling and generalizability.

Reproducible reporting preserves the meaning of the value

A complete record states the certain state, better and worse anchors, duration, probabilities, framing, visuals, respondent perspective, sample, administration mode, search algorithm, stopping rule, interpolation, scoring transformation, exclusions, and uncertainty. It also reports states worse than death, non-trading responses, inconsistent paths, missing data, adverse reactions, and protocol deviations. When the value enters an economic model, the report should identify the exact parameter, time period, aggregation method, uncertainty distribution, and effect on QALYs or decisions.