Nova Academy — Lesson 028: Propagation of Measurement Uncertainty

Nova Academy · Foundations of Inquiry · November 17, 2026

Lesson 028 builds on Lesson 027: Uncertainty Budgets and Measurement Error.

Learning objectives

By the end, you should be able to (1) distinguish input uncertainty from output uncertainty, (2) propagate uncertainty through addition and subtraction under stated independence assumptions, (3) explain why correlated errors need special treatment, and (4) test a result with a simple simulation.

From measurement to calculated result

Suppose you measure two lengths and calculate their sum. Each length is uncertain. The sum is not magically exact simply because you used a calculator. The uncertainty of a calculated result depends on the relationship between the inputs and the mathematical operation. An uncertainty budget lists the sources; propagation estimates how they affect the result.

Worked example: combining independent uncertainties

Two independently measured lengths are A = 12.0 ± 0.2 cm and B = 8.0 ± 0.3 cm. Treat the ± values as standard uncertainties (approximately one standard deviation), with independent, approximately unbiased measurement errors. The total length is L = A + B = 20.0 cm. For independent inputs, the combined standard uncertainty is u(L) = √[u(A)² + u(B)²] = √(0.2² + 0.3²) = √0.13 ≈ 0.3606 cm. Report L = (20.0 ± 0.36) cm as an illustrative standard-uncertainty result. Do not add 0.2 + 0.3 and call 0.5 the standard uncertainty: that is a different, conservative bound under different assumptions.

When independence fails

If both rulers have a common calibration offset, their errors can move together. For a sum, Var(A+B) = Var(A) + Var(B) + 2Cov(A,B). Positive covariance raises uncertainty; negative covariance can reduce it. Simply using the square-root-of-squares rule without checking correlation can understate or overstate uncertainty. For a difference, the covariance term has the opposite sign.

A second example: scaling

Suppose a measured width W = (5.0 ± 0.1) cm is multiplied by an exact factor 3. Then P = 3W = 15.0 cm and u(P) = 3u(W) = 0.3 cm. The multiplier scales both the estimate and its absolute standard uncertainty. If the factor is also uncertain, its contribution must be included.

Activity: paper-strip experiment

Measure two paper strips three times each using the same ruler. Record all readings, calculate each strip’s average, and estimate the spread of repeated measurements. Add the averages and compare your result with a direct measurement of the joined strips. Identify at least two uncertainties: ruler resolution, alignment, reading angle, and shared zero offset are possibilities. State whether the measurements are independent; if not, explain what common source connects them. Do not treat three repeated readings as a complete calibration study.

Reflection

What assumptions made the square-root-of-squares calculation appropriate? How might a shared zero error change the result? When would an interval describing possible extremes be more useful than a standard uncertainty?

Knowledge check

1. If two independent standard uncertainties are 0.4 and 0.3 units, what is their combined standard uncertainty for a sum? 2. Does subtracting two measurements always eliminate uncertainty? 3. If X is multiplied by an exact factor of 4, what happens to its standard uncertainty? 4. What extra quantity matters if two measurement errors are correlated?

Answer key

1. √(0.4² + 0.3²) = 0.5 units. 2. No; the uncertainty depends on both measurements and any covariance. 3. It is multiplied by 4. 4. Their covariance (or correlation coefficient together with their standard uncertainties).

Limitations and responsible interpretation

These examples assume simple approximately linear calculations, appropriately characterized standard uncertainties, and, where stated, independent inputs. Nonlinear models, strongly non-Gaussian distributions, large relative uncertainties, unknown bias, and correlated sources may require sensitivity coefficients, Monte Carlo propagation, or specialized analysis. A standard uncertainty is not automatically a 95% confidence interval. Numerical inputs here are invented for teaching, not experimental data.

Sources and further reading

JCGM 100:2008, Guide to the Expression of Uncertainty in Measurement (GUM); NIST Technical Note 1297, Guidelines for Evaluating and Expressing the Uncertainty of NIST Measurement Results; NIST/SEMATECH e-Handbook of Statistical Methods. These references provide the formal context beyond this beginner exercise.

Nova Academy index