Lesson 027 — Uncertainty Budgets: Where Measurement Error Comes From
Foundations of Inquiry | November 16, 2026. A measurement is an estimate, not an exact copy of reality. An uncertainty budget lists plausible sources of error, quantifies their effects when possible, and makes the limits of a result visible. This lesson builds on Lesson 026’s calibration and validation.
Learning objectives
- Distinguish random variation from systematic bias
- identify instrument resolution, calibration, environment and repeatability as possible uncertainty sources
- calculate a mean and sample standard deviation
- distinguish an uncertainty estimate from a confidence interval
- explain why correlated errors cannot always be combined by a simple square-root formula.
Key concepts
Random error changes from reading to reading and can sometimes be reduced by averaging repeated independent measurements. Systematic error shifts readings consistently, such as a scale that reads 0.3 g high. More repetitions alone do not remove that bias. Accuracy describes closeness to an accepted reference; precision describes repeatability. Calibration compares an instrument against a reference, but the reference itself also has uncertainty.
Worked example: a laboratory balance
Suppose three readings of the same stable object are 10.1 g, 10.3 g, and 10.2 g. Their mean is (10.1 + 10.3 + 10.2)/3 = 10.2 g. The sample standard deviation is sqrt(((−0.1)^2 + (0.1)^2 + 0^2)/(3−1)) = sqrt(0.01) = 0.1 g. The estimated standard error of the mean is 0.1/sqrt(3) ≈ 0.058 g, if the readings are independent and the conditions are stable.
Now suppose a reference check indicates the balance reads 0.3 g too high. Correcting the mean gives 10.2 − 0.3 = 9.9 g. This correction is not the same as the 0.058 g repeatability uncertainty. We must separately account for uncertainty in the calibration correction and any other important effects. The example is invented and does not constitute a certified measurement.
Build an uncertainty budget
- Write down what quantity you are measuring and why
- list instrument resolution, repeated-reading variation, calibration, environmental effects, and operator effects
- estimate each source in common units, stating assumptions
- identify which sources are correlated
- document corrections separately from uncertainty estimates
- report the result with a sensible number of digits and the method used.
Activity
Measure the same object five times using a ruler. Record each reading to the ruler’s appropriate resolution. Compute the average and range. Rotate the ruler or change the observer and repeat. Compare the spread between groups and discuss whether a shared zero-point error could remain even if every reading agreed. Do not assume the range alone is a formal uncertainty interval.
Reflection
If ten measurements agree almost perfectly, does that prove the instrument is accurate? What independent reference would help you test a consistent bias? When might environmental changes matter more than extra decimal places?
Knowledge check
- 1. A balance reads 0.2 g too high every time. Is that primarily random or systematic error? 2. What is the mean of 4, 5, and 6? 3. Why might averaging fail to remove a shared calibration bias? 4. Is the standard deviation of readings automatically a 95% confidence interval?
Answer key
- 1. Systematic error. 2. (4+5+6)/3 = 5. 3. The same offset affects all readings, so averaging retains it. 4. No
- a confidence interval requires a specified statistical model and sampling assumptions, and a standard deviation measures spread rather than directly specifying a 95% interval.
Limitations
Real uncertainty budgets may require sensitivity coefficients, probability distributions, correlations, nonlinearity and traceable calibration. The square-root-of-sums-of-squares method applies only under suitable assumptions; correlated contributions require covariance terms. A small sample gives a fragile estimate of repeatability. Report uncertainty and confidence intervals according to the relevant measurement standard rather than treating them as interchangeable.
Sources and further reading
NIST Technical Note 1297 — Guidelines for Evaluating and Expressing the Uncertainty of NIST Measurement Results; JCGM — Guide to the Expression of Uncertainty in Measurement; NIST/SEMATECH e-Handbook of Statistical Methods.
Previous: Lesson 026 — Calibration and Model Validation · Academy index
