Find your next learning step
A robust action set can reveal that two capacities are both defensible, but it does not say whether collecting another week of queue data is worth the delay and cost. A value-of-information trigger compares the expected improvement in the final capacity decision with the cost of measurement and waiting. Sequential radius review then watches whether new evidence narrows the action set, shifts its boundaries or exposes a changed operating regime. The result is a governed recalibration cycle, not continuous tuning whenever a new observation arrives.
Open the chapter contents · Queueing Theory root · Gold Standard of Education · How X Works
Chapter contents
Choose the route closest to the decision, or read from Chapter 1. The chapters remain open; these grouped routes are a navigable index.
Route 1 · Define the decision
Route 2 · Work the ledgers
Route 3 · Transfer the method
Route 4 · Check limits and evidence
Route 5 · Ask and continue
CHAPTER 1 OF 12 · DEFINE THE DECISION
1. Ask whether information can change the action
More precise arrival and service estimates have little operational value when every plausible update keeps the same capacity. Information becomes valuable when it can remove a disputed action, prevent a costly switch or reveal that the current ambiguity set no longer covers the relevant regime.
Define the decision first: capacity, routing rule, staffing band or contingency trigger. Information value is measured through that decision and its loss, not through statistical precision alone.
CHAPTER 2 OF 12 · DEFINE THE DECISION
2. Separate expected value from guaranteed value
Expected value of sample information averages the improvement from a proposed sample over its possible outcomes. It is an upper-level planning quantity and depends on the predictive model, action set and loss.
A positive point estimate is not a promise. Monte Carlo error, model misspecification and collection delay can consume the apparent gain, so report uncertainty and compare a conservative value with the total acquisition cost.
CHAPTER 3 OF 12 · WORK THE LEDGERS
3. A value-of-information ledger
A service centre can collect one, two or four additional operating weeks before fixing next month’s roster. Each option is evaluated through nested predictive simulation.
| Extra data | Expected regret reduction | Collection and delay cost | Conservative net value | Decision |
|---|---|---|---|---|
| 0 weeks | 0.0 | 0.0 | 0.0 | Implement current robust set |
| 1 week | 2.8 | 1.1 | 1.2 | Collect |
| 2 weeks | 4.1 | 2.3 | 1.0 | Collect only if switch risk stays high |
| 4 weeks | 5.0 | 5.4 | −1.1 | Do not wait |
The best sample is not necessarily the largest. Diminishing decision value and the cost of postponing capacity can make a short targeted sample preferable.
CHAPTER 4 OF 12 · WORK THE LEDGERS
4. Worked trigger from the robust action set
Suppose capacities 19 and 20 are admitted because each has worst-band regret below five. Simulated one-week samples remove one of them in 62 per cent of futures and reduce expected implementation regret by 2.8 units.
| Quantity | Estimate | Conservative adjustment | Used value |
|---|---|---|---|
| Regret reduction | 2.8 | −0.5 Monte Carlo margin | 2.3 |
| Measurement cost | 0.7 | +0.1 contingency | 0.8 |
| Delay cost | 0.3 | No adjustment | 0.3 |
| Net information value | 1.8 | −0.6 model-risk reserve | 1.2 |
If the model-risk reserve made the net value negative, implement from the present action set rather than collecting data merely because uncertainty exists.
CHAPTER 5 OF 12 · WORK THE LEDGERS
5. Review the radius on a schedule and on events
Use a routine review cadence aligned with staffing or procurement, plus event triggers for structural change: new service policy, altered abandonment, a demand shock, changed measurement resolution or sustained coverage failure.
Do not re-estimate the radius after every ordinary fluctuation. Continuous reactive tuning can turn noise into operational churn and invalidate a holdout that is repeatedly reused.
CHAPTER 6 OF 12 · TRANSFER THE METHOD
6. Use sequential stopping with a declared precision target
Sequential sampling methods can stop when an optimality-gap interval is sufficiently narrow for the decision. Choose the tolerance before sampling and account for repeated looks rather than applying a fixed-sample interval at every checkpoint.
A narrow objective gap can still hide an action switch at a discrete boundary. Require both a precision gate and an action-stability gate across the accepted radius band.
CHAPTER 7 OF 12 · TRANSFER THE METHOD
7. Changed-context transfer: ambulance intake
Transfer the method to ambulance bays by valuing a new week of shift-level data through changes in safe bay staffing, diversion and delay. Severe-event clusters remain intact in resampling.
The transfer check raises the cost of delay and sets clinical safety gates. It does not reuse the service-centre net-value threshold or assume that an average regret reduction offsets a tail-risk failure.
CHAPTER 8 OF 12 · TRANSFER THE METHOD
8. Connect action sets, bands and calibration
Start with the robust action-set guide. The radius-uncertainty guide supplies the calibration distribution used in predictive sampling.
Return to the coverage–conservatism guide when the review reveals coverage drift or a reused holdout.
CHAPTER 9 OF 12 · CHECK LIMITS AND EVIDENCE
9. Limits and common misconceptions
Information gain is not the same as decision value. Data can sharply improve parameter estimates while leaving the preferred action unchanged, or weakly change estimates while resolving a capacity breakpoint.
A positive expected net value does not guarantee a better realised decision. It is conditional on the predictive model, candidate samples, collection cost and loss used in the calculation.
CHAPTER 10 OF 12 · CHECK LIMITS AND EVIDENCE
10. Practical evidence checklist
Archive the decision, loss, current action set, predictive model, sample options, nested simulation, random seeds, collection and delay costs, uncertainty adjustment, trigger threshold and stopping rule.
At each review, record why it was routine or event-driven, whether the radius changed, which action boundaries moved, what holdout remained untouched and when the next review becomes eligible.
CHAPTER 11 OF 12 · ASK AND CONTINUE
11. Frequently asked questions
Must recalibration happen whenever the radius changes?
No. Recalibrate operational capacity when the change affects the declared action or risk gate, not merely a reported decimal.
Can information value be negative?
The gross value of optional information is non-negative in an ideal coherent model, but net value becomes negative after collection, delay and implementation costs.
What if the sample plan was chosen after seeing results?
Treat it as adaptive selection and evaluate it with an appropriate sequential design or fresh data; do not report a fixed-plan calculation.
CHAPTER 12 OF 12 · ASK AND CONTINUE
12. Sources and next reading
Pasupathy and Schmeiser develop a sequential sampling procedure for stochastic programming. Bayraksan and Morton study fixed-width sequential stopping rules based on optimality-gap intervals. Esfahani and Kuhn provide data-driven Wasserstein robust optimisation. The trigger ledgers here are original teaching examples.
Continue through the Queueing Theory root, the Gold Standard of Education and How X Works.
Top · Contents · Previous · How X Works