Independent Estimation¶
Elicitation method — instantiates Bias-Specific Decision Audit
Collects judgments from several people separately, before any of them see the others' answers, then aggregates — so the estimate reflects genuinely independent information instead of the first number or the loudest voice.
Independent Estimation collects a judgment from each of several people separately — before anyone sees anyone else's answer — and then aggregates the results, so the final estimate rests on genuinely independent information rather than on the first number spoken or the most senior voice in the room. Its defining idea is the independence of the estimators: when people estimate in parallel, their individual errors are uncorrelated and tend to cancel in the aggregate, whereas the moment one estimate is heard the rest anchor to it and the group converges on a shared, confident, and possibly wrong number. It fights anchoring and groupthink not by hiding information about the subject but by preventing the estimators from contaminating each other.
Example¶
A consumer-goods company is forecasting next quarter's demand for a new product, a number that will size a factory order. In the old process, the VP of sales opened the meeting with a bullish figure and everyone else's "estimate" drifted toward it. The team switches to independent estimation. Each of seven people — sales, operations, finance, two regional managers, and two analysts — submits a private point estimate and a range before the meeting, with no visibility into the others.
The submissions arrive spread widely: several near the VP's optimistic number, but three far lower, one citing a supply constraint the VP hadn't weighed. Only then does the group convene, now arguing over a distribution of independent views rather than ratifying an anchor. The aggregate — and the visible disagreement — reframes the order from a confident bet on the high number to a hedged commitment around the median with an option to reorder. The estimators were never masked; they were simply kept from hearing one another until each had committed a view.
How it works¶
- Elicit in parallel, in private. Every estimator produces a judgment with no access to the others', which is the whole source of the method's power; a single leaked "anchor" number collapses it.
- Commit before exchange. Estimates are locked before any discussion, so the subsequent debate revises independent views rather than manufacturing a false consensus.
- Aggregate, then discuss. Combine the estimates (a median, a trimmed mean, a full distribution) and treat the spread as signal — wide disagreement flags a real unknown. Structured variants iterate: reveal the anonymized distribution, let people revise privately, and repeat.
- Preserve dissent. The outlier who saw the supply constraint is the payoff; a process that averages the disagreement away without examining it throws out the method's most valuable output.
Tuning parameters¶
- Panel size and diversity — more and more-varied estimators widen the independent information pool and improve cancellation, but cost coordination and can dilute genuine expertise.
- Anonymity — whether estimates and revisions are attributed. Anonymity protects against seniority and conformity pressure; attribution improves accountability. Structured Delphi keeps it anonymous for exactly this reason.
- Aggregation rule — median, mean, trimmed, or weighted by track record. Robust rules resist a single wild estimate; weighting rewards demonstrated calibration but requires a history to weight on.
- Rounds — one-shot versus iterated revision. Iteration lets estimates converge on shared information without re-importing the anchor, but each round costs time and risks quiet herding.
When it helps, and when it misleads¶
Its strength is turning a single anchored guess into a distribution of independent views, which both improves the central estimate and — often more valuably — surfaces the disagreement a consensus meeting would have buried. Under genuine independence, aggregation is a remarkably cheap accuracy gain.[n1]
Its failure mode is that independence is easily counterfeit. If the estimators share a training, a data source, or a boss's known preference, their errors are correlated and the aggregate is confidently biased in the same direction — averaging does nothing about a bias everyone shares. And the smallest procedural leak — a number mentioned in the hallway, an email that goes to the whole group — re-anchors everyone, and the parallel structure becomes theater. The guarding discipline is to protect the independence literally: elicit before any exchange, keep early estimates private, recruit estimators who don't share the same blind spot, and treat suspiciously tight agreement as a warning that independence has already been lost.
How it implements the components¶
independence_requirement— its core: the estimators must judge separately, with no access to each other's answers, so their errors stay uncorrelated.targeted_bias_check— the parallel elicitation is a concrete check aimed at anchoring and groupthink, the biases a sequential group discussion reliably produces.
It does not mask the subject's identity or gate on concealment-before-exposure (review_timing_gate — that is Blind or Masked Review). Blind or Masked Review hides information about the thing being judged from one reviewer; Independent Estimation hides the reviewers from each other.
Related¶
- Instantiates: Bias-Specific Decision Audit — Independent Estimation is the check an audit deploys when its map flags anchoring, seniority pressure, or groupthink in a judgment.
- Sibling mechanisms: Bias Audit · Blind or Masked Review · Hiring Review Rubric · Diagnostic Debiasing Check · Decision Checklist · Structured Review Form · Decision Log · Reference-Class Forecasting
Editorial Notes¶
Form Classification¶
Form family: Communication, Facilitation & Learning
Rationale: Several people are separately prompted for judgments before exposure to peers, then their genuinely independent information is aggregated.
Nearest alternative: Protocol, Workflow & Routine — The method follows a repeatable order, but the operative mechanism is designed human elicitation rather than workflow coordination.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Convergent development
Present-day reach: Universal
Rationale: Independent judgments followed by aggregation arise centrally from statistical reasoning about uncorrelated errors and aggregate estimation.
Related originating lineages:
- Futurism & Strategic Foresight — The Delphi tradition operationalized anonymous independent estimates for expert forecasting.
- Operations Research — Delphi methods professionalized independent expert elicitation for forecasting.
- Organizational & Management Science — Group-decision practice materially developed private elicitation as a defense against hierarchy and anchoring.
- Psychology — Research on anchoring, conformity, and correlated judgment explains why estimates must be isolated before aggregation.
Review resolution: Both independent reviews place the primary lineage in statistics_experimental_design. The queued differences (alternate_origin_disagreement) concern secondary metadata rather than primary provenance. The final retains futurism_foresight, organizational_management, operations_research, psychology only where a reviewer supplied a formative-lineage rationale; this does not convert downstream applicability into origin. origin_mode=convergent because the reviewers document independently established or materially co-developing traditions. domain_reach=universal records application breadth separately from provenance.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] The parallel-then-aggregate idea underlies both the "wisdom of crowds" (Francis Galton's observation that the average of many independent guesses can beat the experts) and the RAND Corporation's Delphi method, which iterates anonymous, independent estimates precisely to keep panelists from anchoring on one another. Both depend entirely on the estimates being independent to begin with. ↩