How Reference Class Forecasting Works | Estimate From Comparable Outcomes Before Trusting the Inside View

Reference class forecasting works by asking a question that ordinary planning often avoids: what happened to similar cases in reality?

Instead of beginning only with the details of the current task, project or study plan, the method starts with a comparison group.

That outside view is powerful because the current case always feels special from the inside. We know its intentions, hopes and planned sequence. Comparable outcomes remind us that many earlier cases also had detailed plans.

This article is the first pillar of How Estimates Fail.

The Core Idea: Start With Distribution Before Story

The inside view tells a story about this case.

The outside view asks where cases like this usually ended up.

Neither view should automatically replace the other. The outside view creates a prior expectation. The inside view then adjusts for features that are genuinely different.

Homes England describes reference class forecasting as using actual performance from past comparable projects to generate an external view of cost and schedule risk and a realistic range of outcomes.

Step 1: Define the Thing Being Estimated

A reference class is only useful when the target is clear.

Time to complete one homework assignment.
Duration of a website rebuild.
Cost of a renovation.
Time needed to revise one A-Math chapter to independent mixed-question performance.
Number of editing passes before an essay is submission-ready.

Different targets require different comparison sets.

Step 2: Choose Comparable Cases

Similarity should be decision-relevant, not superficial.

Two projects may both be called ‘website redesign’ while differing dramatically in number of pages, migration complexity, integrations and approval structure.

Two students may both be in Secondary 3 while one is repairing prerequisite algebra and the other is refining timing.

The class should match the factors that dominate the outcome.

Step 3: Prefer Real Outcomes to Planned Outcomes

A reference class built from earlier estimates simply recycles earlier optimism.

Use actual completion times, final costs, observed scores, realised workloads or other outcome data where possible.

The method gains strength by comparing what people expected with what actually happened.

Step 4: Build the Outcome Distribution

Do not reduce the class immediately to one average.

Look at spread.

What was the median?
How wide was the middle of the distribution?
How bad were the overruns?
Were there a few extreme cases?
Was the error mainly in one direction?

The distribution shows the risk hidden by a single point.

Step 5: Locate the Current Case

Ask whether the present case is genuinely easier, typical or harder than the reference set.

The word genuinely matters.

‘Our team is better’ is weak without evidence. ‘This project has half the page count and no data migration’ is stronger.

Adjustments should correspond to observable differences.

Step 6: Keep the Reference Class Broad Enough

A class of three cherry-picked successes is not an outside view.

Homes England’s guidance notes that more comparable projects generally improve robustness and suggests that roughly 20–30 projects can be enough to get started where such data exist.

In small personal decisions, you may have fewer cases. The principle still applies: use the broadest honest evidence set available.

The Reference Class Selection Trap

The method can be manipulated by choosing the class after seeing the desired answer.

A planner who wants a shorter estimate can select only fast cases.

A better system defines the comparison criteria before reading the final distribution.

Reference Classes and Novel Work

What if nothing is truly comparable?

Then use nested classes.

A broad class for the whole category.
A narrower class for similar scale.
A component-level class for familiar sub-tasks.
An explicit uncertainty premium for genuinely novel parts.

Novelty should widen uncertainty rather than magically justify confidence.

Reference Classes for Student Time Planning

A student can build a personal reference class from previous tasks.

Instead of saying ‘I think this worksheet will take thirty minutes’, compare the last five worksheets of similar length and difficulty.

eduKateSingapore’s homework-estimation article uses exactly this feedback logic: estimate, record focused and elapsed time, compare similar work and update future plans.

Read Why Students Underestimate How Long Homework Takes.

Reference Classes for Revision

A revision plan can use evidence from earlier chapters and previous assessment cycles.

How many sessions did it take before a repaired topic survived mixed questions?

How long did a full paper plus marking plus repair actually take?

This is more useful than treating every new chapter as a blank planning problem.

eduKate Sengkang’s How Study Planning Works treats time estimates as revisable assumptions and encourages using experience from similar work where possible.

Reference Classes in Mathematics

Mathematics uses a related idea in estimation: known magnitudes provide a neighbourhood for a new answer.

Bukit Timah Tutor’s Primary Mathematics estimation guide asks learners to predict rough magnitude before trusting exact arithmetic.

The reference class is not identical to project forecasting, but the cognitive discipline is similar: compare the new result with known structure before accepting it.

Reference Classes in English Writing

SETC exam-time planning can also use outside-view evidence.

A student may believe a body paragraph ‘should take ten minutes’. Timed practice can show the actual distribution across different prompts.

The schedule should be built from those observed durations, not from one optimistic memory.

When the Outside View Should Change the Plan

The outside view is most valuable when it contradicts confidence.

If every comparable case overran, the current estimate needs strong evidence to justify being the first one that will not.

If similar students needed six weeks to stabilise a skill, a two-week plan may still work, but it should be treated as an aggressive case rather than the default.

When the Inside View Still Matters

Reference classes are not fate.

The current case may contain real differences.

A better tool, smaller scope, stronger prerequisite, more experienced team or lower quality requirement can legitimately shift the estimate.

The discipline is to adjust from evidence, not from hope.

A Simple Outside-View Worksheet

Target — what outcome are we estimating?
Reference class — which past cases are genuinely comparable?
Observed distribution — what were their actual outcomes?
Baseline — what central estimate does the class imply?
Differences — what is materially different about this case?
Adjustment — how should those differences move the estimate?
Range — what plausible uncertainty remains?
Decision — what changes because of this estimate?

The Deeper Principle: Similar Cases Are Data About the Future

Planning often begins by imagining the future from scratch.

Reference class forecasting reminds us that many futures have already partially happened in comparable form.

The outside view does not remove uncertainty.

It makes optimism pay rent to evidence.

Across the eduKate Ecosystem

eduKateSG owns the general estimation mechanism. eduKateSingapore carries project estimation and deeper planning methods. eduKate Sengkang applies estimation to study planning. eduKate Punggol applies it to examination-year workload and family capacity. Bukit Timah Tutor develops mathematical estimation and verification. SETC applies time estimation to English exam execution. eduKate Yishun applies evidence-first estimation to recovery and rebuild decisions.