Uncertainty propagation turns a measured number into a defensible physics result. This practical guide shows how to calculate absolute, relative, and percentage uncertainty; combine measurements; draw meaningful error bars; round with appropriate significant figures; and check a lab calculation before submitting it.
Overview
Every measurement has a limit to its precision. If you measure a length as 25.0 cm, the final digit communicates something about the instrument and method. The result is not simply a number; it is better written as a value with an uncertainty, such as (25.0 ± 0.1) cm.
Uncertainty is not the same as a mistake. A random error causes repeated measurements to scatter, while a systematic error shifts results consistently in one direction. Repeating trials can help estimate random uncertainty, but repetition alone does not remove a zero offset, poor calibration, parallax, or an unsuitable experimental model.
Three useful ways to express uncertainty
| Type | Formula | Meaning | Example |
|---|---|---|---|
| Absolute uncertainty | Δx | Uncertainty in the same unit as the measured quantity | 0.1 cm |
| Relative uncertainty | Δx / x | Uncertainty compared with the measured value | 0.1 / 25.0 = 0.004 |
| Percentage uncertainty | 100(Δx / x)% | Relative uncertainty written as a percentage | 0.40% |
Use the form that matches the calculation. Absolute uncertainties are convenient for addition and subtraction. Relative or percentage uncertainties are usually more convenient for multiplication, division, and powers.
A reusable propagation workflow
- Define the measured inputs. Record each value, unit, instrument resolution, and uncertainty before calculating.
- Write the equation. Do not substitute numbers until the relationship is clear.
- Identify the propagation rule. Addition and multiplication require different treatments.
- Calculate with extra digits. Round only at the end.
- Report value and uncertainty together. Include units and, where useful, the percentage uncertainty.
- Check plausibility. Confirm units, scale, signs, and limiting behavior. A useful companion is the plausibility checklist for physics results.
Checklist by scenario
Addition and subtraction
For quantities that are added or subtracted, combine independent absolute uncertainties in quadrature:
z = x ± y
Δz = √[(Δx)2 + (Δy)2]
For a conservative introductory lab estimate, some courses use the maximum-uncertainty method:
Δz = Δx + Δy
Follow the convention specified by your course or laboratory manual. For example, if a pendulum length is L = (0.800 ± 0.002) m and a second measured length is d = (0.120 ± 0.001) m, then the difference is L − d = 0.680 m. Quadrature gives an uncertainty of approximately 0.0022 m, so the result can be reported as (0.680 ± 0.002) m after sensible rounding.
Multiplication and division
For a product or quotient, combine relative uncertainties in quadrature:
z = xy or z = x/y
Δz/z = √[(Δx/x)2 + (Δy/y)2]
Example: a cart travels s = (1.20 ± 0.01) m in t = (2.50 ± 0.02) s. Its speed is v = s/t = 0.480 m/s. The relative uncertainty is approximately √[(0.01/1.20)2 + (0.02/2.50)2] = 0.011, or 1.1%. Therefore, v = (0.480 ± 0.005) m/s is a suitable presentation.
Powers and roots
If z = xn, then, for small independent uncertainties:
Δz/z = |n|(Δx/x)
For a sphere, volume is V = (4/3)πr3. A 2% uncertainty in radius produces approximately a 6% uncertainty in volume. This is why measuring a diameter carefully matters in density experiments: the radius enters cubed.
Mechanics: determining gravitational acceleration
Suppose an experiment uses g = 2h/t2. The height has uncertainty Δh and the time has uncertainty Δt. Treating the factor 2 as exact:
Δg/g = √[(Δh/h)2 + (2Δt/t)2]
The time contribution is doubled because time is squared in the denominator. Before collecting data, this expression helps identify which measurement deserves improvement.
Circuits: resistance from Ohm’s law
For R = V/I, use:
ΔR/R = √[(ΔV/V)2 + (ΔI/I)2]
If V = (6.00 ± 0.05) V and I = (0.240 ± 0.005) A, then R = 25.0 Ω. The relative uncertainty is about 2.2%, giving approximately R = (25.0 ± 0.6) Ω. When working with series and parallel circuits, first check that the measured circuit matches the assumed model; see the Ohm’s law and circuit analysis guide.
Optics: a fitted focal length
When focal length is obtained from a graph rather than a single reading, the uncertainty should reflect the fit. For the thin-lens relationship, plotting 1/v against 1/u can produce a straight line whose intercept is related to focal length. Do not report only the calculator’s digits. Use the fit uncertainty supplied by the regression, or estimate it from an appropriate confidence interval or extreme acceptable lines if that is the method taught in your course. Record the number of data points, fit equation, units, and residual pattern.
Thermal experiments: temperature differences
If heat transfer is calculated from Q = mcΔT, and ΔT = Tfinal − Tinitial, first propagate the two temperature readings into ΔT. For independent readings with equal uncertainty ΔTreading, the difference has uncertainty √2 ΔTreading. Then combine the relative uncertainties of mass, specific heat capacity, and temperature change. If the temperature change is small compared with the thermometer uncertainty, the final heat estimate may be much less precise than the raw temperature values suggest.
What to double-check
Significant figures and rounding
Keep guard digits during intermediate calculations. Round the uncertainty to one significant figure, or sometimes two when the leading digit is 1 or 2 and your reporting convention allows it. Round the measured value to the same decimal place as the uncertainty. For example, write (12.36 ± 0.08) cm, not (12.36471 ± 0.08) cm and not (12.4 ± 0.08) cm.
Random versus systematic effects
Use repeated measurements, a mean, and a spread such as standard deviation to describe random variation. A systematic effect requires a different response: check zero readings, calibrate the instrument, reverse or reposition equipment where appropriate, and discuss the likely direction of the bias. Do not hide a suspected systematic error inside a small statistical uncertainty.
Error bars and fitted quantities
Label error bars clearly. State whether they represent instrument uncertainty, standard deviation, standard error, a confidence interval, or another defined quantity. The visual size of an error bar is not automatically evidence that two results agree or disagree; compare the stated uncertainty model and the physical question. For a best-fit line, inspect residuals rather than relying only on the slope and its displayed digits. The physics data analysis tutorial provides a broader workflow for graphs and conclusions.
Spreadsheet and Python checks
In a spreadsheet, keep separate columns for each measured value, its uncertainty, the calculated result, and the propagated uncertainty. Use formulas rather than manually retyping rounded values. In Python, a simple independent-product calculation can be written as:
import math
x, dx = 1.20, 0.01
t, dt = 2.50, 0.02
v = x / t
relative = math.sqrt((dx / x)**2 + (dt / t)**2)
dv = v * relative
print(v, dv)For correlated variables, nonlinear fits, or large uncertainties, the simple formulas may be inadequate. Use a covariance-aware fit, Monte Carlo sampling, or a method required by your course. State the method in the report.
Common mistakes
- Adding percentage uncertainties for a sum. Use absolute uncertainties for addition and subtraction.
- Forgetting powers. If a variable is squared, its relative uncertainty contribution is multiplied by 2.
- Using instrument resolution as the whole uncertainty without explanation. Resolution may be a reasonable component, but calibration, repeatability, and reading technique can also matter.
- Rounding every intermediate value. This can create avoidable disagreement between equivalent calculations.
- Calling every discrepancy “random error.” A consistent offset is evidence to investigate systematic error.
- Reporting an uncertainty with no units. Absolute uncertainty must use the same unit as the result.
- Confusing standard deviation with uncertainty in the mean. The spread of individual readings and the precision of their average answer different questions.
- Presenting a fit without describing the data. Include axes, units, fit parameters, uncertainty method, and enough information to reproduce the analysis.
For a complete report structure, pair this checklist with the physics lab report format guide.
When to revisit
Return to this checklist whenever you change an instrument, measurement range, data-processing tool, fit model, or laboratory procedure. Recheck the uncertainty model before a new practical investigation, after replacing sensors, or when a result appears unexpectedly precise or inconsistent. Course conventions also vary: confirm whether your instructor expects quadrature, maximum uncertainty, standard deviation, standard error, or a particular confidence level.
Before submitting a lab report, run this final sequence:
- Write every final result as value, uncertainty, and unit.
- Confirm that the uncertainty method matches the equation and the course convention.
- Round the uncertainty and value to matching decimal places.
- Separate random variation from possible systematic bias.
- Check graphs, error-bar definitions, residuals, and fitted-parameter units.
- Compare the result with a plausible scale or reference expectation without forcing agreement.
- State what measurement or assumption most limits the conclusion.
Used consistently, uncertainty propagation is more than a final formatting step. It guides experimental design, shows which measurements matter most, and makes a physics conclusion easier to evaluate and reproduce.