Skip to content

MODULE 5 · LESSON 29 · Analytical thinking

Understanding uncertainty

Identify distinct sources of uncertainty and avoid reporting more precision than the evidence supports.

By Laurence Svekis · Self-paced · All worked data is synthetic unless you supply your own source.

Try the practice ↓

PredictInspectCalculateCheckExplainTransfer

Your starting point

Prerequisites: Use the ideas and vocabulary from lesson 28; revisit its worked examples if needed.

Definition: Uncertainty is limited knowledge about a quantity, process or conclusion. It can arise from sampling variation, measurement error, missingness, selection, model assumptions or future change. A point estimate is a single numerical summary, not a complete description of uncertainty.

Separate uncertainty from a mistake

A correct calculation can still be uncertain as an estimate of a wider population. A sample mean exactly summarizes the observed sample under its rules, but generalizing beyond that sample requires assumptions. Measurement error concerns the observations themselves; sampling variation concerns which observations were selected. Improving arithmetic does not automatically solve either problem.

Use intervals only with a justified method

A confidence interval has a method-specific interpretation tied to a sampling or statistical model. It is not a blanket probability that every individual lies within the interval, and it cannot repair selection bias. This beginner workbench provides descriptive statistics rather than automatic inferential intervals. Do not invent a confidence band by adding arbitrary percentages to a point estimate.

Communicate precision and decision relevance

Round results to a useful level while retaining the underlying calculation for checking. A difference of 0.0001 may be immaterial to the decision and unsupported by measurement precision. Ask what range of plausible outcomes would change the recommended action. A good uncertainty statement names the source, its likely consequence and an additional check rather than using vague language to avoid taking responsibility.

Two worked examples

A small observed sample

Three recorded waits are 4,5,6 minutes, giving a mean of 5. This is the sample mean, not a guarantee that the next customer waits five minutes or that the population mean is exactly five. A larger and appropriately collected sample could improve some estimates, while systematic missing long waits would remain a concern.

Rounding with context

A calculated campaign rate is 12.345678%. Reporting 12.3% with its counts is often more understandable than six decimal places. The longer display does not make the measurement more accurate. Keep the unrounded value for internal calculations and explain the denominator and collection limitations.

Where this becomes useful

A manager must decide whether to pilot a new process. The relevant question may be whether plausible benefits justify a small reversible test, not whether a point estimate can be presented as certain. State what evidence the pilot should collect.

Make an attempt

Use Dataset Comparison Studio. Inspect group counts and spread, then write an uncertainty note without inventing a confidence interval. Identify which uncertainty could be reduced by more records and which requires a better collection process.

Open the accompanying interactive lab →

Misconception check

Claim: Uncertainty means the analysis is useless.

Inspect the correction

Useful decisions can be made with bounded uncertainty when its sources, consequences and next checks are explicit.

Independent application

Write a finding from a small synthetic sample, then attach an uncertainty statement naming sampling, measurement or selection limits. Explain one additional observation or design improvement that would reduce the most decision-relevant uncertainty.

Completion records practice and self-review. It is not certification of statistical competence.