Die Druckversion wird nicht mehr unterstützt und kann Darstellungsfehler aufweisen. Bitte aktualisiere deine Browser-Lesezeichen und verwende stattdessen die Standard-Druckfunktion des Browsers.

Measurement and Experimental Uncertainty

Introduction

Every experimental result is shaped by the way it was measured. A ruler has finite scale divisions, a stopwatch has finite timing resolution, sensors need calibration, people react at slightly different times, and environmental conditions can vary. For this reason, a scientifically useful result is more than a number: it is a measured value together with enough information to judge how reliable that value is.

In this aiMOOC for Grades 11–13, you learn how to plan, perform, analyse, and communicate measurements with measurement uncertainty. You will distinguish accuracy and precision, identify random and systematic effects, estimate uncertainty from instruments and repeated measurements, propagate uncertainties through calculations, use error bars, and evaluate whether conclusions are supported by the data.

The International System of Units provides a common language for reporting physical quantities. Correct units are essential, but a unit alone does not show how well a quantity is known.

Learning goals

By the end of the course, you should be able to:

Measurement: Report a measured quantity with a value, unit, and justified uncertainty.
Accuracy and precision: Explain how closeness to a reference and repeatability describe different features of data.
Uncertainty: Estimate uncertainty from instrument resolution, repeated readings, specifications, and calibration information.
Propagation: Calculate how input uncertainties affect a derived result.
Error bars: Plot and interpret uncertainty visually without treating error-bar overlap as a complete significance test.
Experimental design: Improve a method by reducing important sources of uncertainty and controlling systematic effects.

What a Measurement Result Means

A measurand is the quantity you intend to measure, such as the length of a wire, the period of a pendulum, or the resistance of a component. A measurement result assigns a value to that measurand using a stated method.

In modern metrology, measurement uncertainty describes the dispersion of quantity values that could reasonably be attributed to the measurand on the basis of the available information. In school laboratory work, you will often express a result in the form:

measured value ± uncertainty, with a unit

For example, a wire diameter might be reported as 0.52 ± 0.01 mm. This statement communicates more than writing only 0.52 mm because it shows the scale of the remaining doubt in the result.

Do not confuse error and uncertainty. Measurement error is the difference between a measured value and a reference value when that reference is available. Uncertainty describes how well the measurand is known after measurement. You can have a small uncertainty around a result that is nevertheless shifted by an uncorrected systematic effect.

Accuracy, precision, resolution, and repeatability

Accuracy is a qualitative description of how close a result is to an accepted reference or true quantity value. Precision describes how closely repeated measured values agree with one another under stated conditions. Resolution is the smallest change in input that produces a detectable change in an instrument indication. Repeatability concerns agreement when measurements are repeated under the same or nearly the same conditions.

A precise set of readings can still be inaccurate if a systematic effect shifts all of them. Repeating a biased procedure many times can reduce random scatter in the mean, but it does not automatically remove that bias.

Instruments, Scales, and Reading Uncertainty

Instrument choice should match the size of the quantity and the uncertainty you need. A metre rule may be suitable for a desk length, while a vernier caliper or micrometer is better for small dimensions.

A vernier scale lets you resolve fractions of the main-scale division. Before measuring, check that the instrument reads zero when it should. A non-zero indication at zero input is a zero error and is evidence of a systematic effect that may require correction.

Datei:VernierscaleHow a vernier scale works.gif

For an analog scale, a common school estimate is to take half of the smallest scale division as the reading uncertainty when the scale can be interpolated clearly. For a digital display, a common first estimate is one unit in the last displayed digit. These are practical starting rules, not universal laws. If a manufacturer specification, calibration certificate, or laboratory instruction provides a more appropriate uncertainty, use that information instead.

A real micrometer is designed for precise measurement of small dimensions. Recognising the instrument, its contact surfaces, and its scale is part of practical measurement competence.

The diagram below connects the physical instrument with a scale reading and an explicitly stated uncertainty.

Fehler beim Erstellen des Vorschaubildes:

The micrometer illustrations show why reporting digits and uncertainty together matters: a finely divided scale supports a smaller reading uncertainty than a coarse scale, but only if the instrument is properly used and calibrated.

Avoiding parallax and alignment errors

Parallax occurs when a scale is viewed from an angle rather than along the intended line of sight. Read pointers and liquid levels at eye level and perpendicular to the scale. Also align rulers with the dimension being measured and avoid measuring from a damaged or rounded zero edge.

Datei:Parallax pointer error.PNG

The measuring image below can be used as a virtual laboratory: decide which digits are justified, estimate the reading uncertainty, and consider how uncertainty would affect a calculated volume.

Datei:Measuring objects (a cylinder and a parallelepiped) with a ruler.png

Random and Systematic Effects

Random effects produce unpredictable variation from reading to reading. Examples include reaction-time variation, electrical noise, small fluctuations in temperature, and difficulty judging an exact endpoint. Repetition can reveal random scatter and usually improves the estimate of the mean.

Systematic effects shift results in a consistent or predictable way. Examples include a zero offset, an incorrectly calibrated sensor, heat loss ignored in a model, or a ruler that has stretched. Repetition alone does not reveal every systematic effect, because repeated values can cluster tightly around the wrong value.

A good experimental strategy therefore combines repetition with method checks: calibrate instruments, compare against standards, reverse or swap equipment where appropriate, control environmental variables, and ask whether the mathematical model matches the physical system.

A useful diagnostic question

When you identify a source of uncertainty, ask: Would this effect mainly create scatter, mainly shift all results, or do both? The answer helps you decide whether repetition, correction, improved calibration, redesign, or better environmental control is the most effective response.

Repeated Measurements and Statistics

Suppose you measure the same quantity several times: $x_{1}, x_{2}, \dots, x_{n}$ . The arithmetic mean is

$\bar{x} = \frac{1}{n} \sum_{i = 1}^{n} x_{i}$

and is often used as the best estimate when the readings are independent and there is no reason to prefer one reading over another.

For a small school dataset, the half-range

$\frac{x_{\max} - x_{\min}}{2}$

is sometimes used as a simple estimate of spread. It is easy to calculate, but it is sensitive to extreme readings and does not use all data efficiently.

For a more statistical description, use the sample standard deviation:

$s = \sqrt{\frac{\sum (x_{i} - \bar{x})^{2}}{n - 1}}$

The standard deviation describes the spread of individual readings. If you are estimating the uncertainty of the mean from independent repeated observations, the standard error is approximately

$s_{\bar{x}} = \frac{s}{\sqrt{n}}$

This decrease with $\sqrt{n}$ applies to random sampling variation; it does not make calibration uncertainty, model error, or other systematic contributions disappear.

Datei:Standard deviation.svg

A normal distribution is often a useful model for repeated random variation, but it is not guaranteed for every experiment. The shape of the data and the measurement process should justify the model.

Type A and Type B evaluations

At an advanced level, uncertainty components are often grouped by how they are evaluated. A Type A evaluation uses statistical analysis of repeated observations. A Type B evaluation uses other information, such as calibration certificates, manufacturer specifications, previous data, resolution limits, or scientific judgment.

Type A does not simply mean random, and Type B does not simply mean systematic. A complete uncertainty budget may contain several components from both evaluation methods.

Absolute, Relative, and Percentage Uncertainty

An absolute uncertainty has the same unit as the measured quantity. If $L = 2.40 \pm 0.03 m$ , then the absolute uncertainty is 0.03 m.

The relative uncertainty is

$\frac{u_{L}}{L}$

and the percentage uncertainty is

$\frac{u_{L}}{L} \times 100 %$ .

For the length above, the relative uncertainty is 0.03/2.40 = 0.0125, or 1.25 percent.

Relative uncertainty is useful for comparing measurements on different scales. An uncertainty of 1 mm is small for a 2 m length but large for a 5 mm thickness.

Combining and Propagating Uncertainties

When a result is calculated from measured inputs, their uncertainties affect the final result. The method you use depends on what the uncertainties represent.

Conservative school rules

Many school courses use maximum or worst-case uncertainty rules:

Addition and Subtraction: Add absolute uncertainties.
Multiplication and Division: Add relative or percentage uncertainties.
Exponentiation: For a power $y = x^{n}$ , multiply the relative uncertainty of $x$ by $| n |$ .

These rules are intentionally conservative because they estimate a plausible maximum effect if all input deviations act in the same direction.

Root-sum-square method for independent standard uncertainties

When uncertainties are standard uncertainties and the input quantities are independent, the usual propagation law combines contributions in quadrature. For $y = f (x_{1}, x_{2}, \dots, x_{N})$ ,

$u_{y} = \sqrt{\sum_{i = 1}^{N} {(\frac{\partial f}{\partial x_{i}} u_{i})}^{2}}$ .

For a sum or difference of independent quantities, this gives

$u_{y} = \sqrt{u_{1}^{2} + u_{2}^{2} + \dots}$ .

For a product or quotient, relative standard uncertainties combine approximately as a root-sum-square when uncertainties are small and independent.

If inputs are correlated, covariance terms are needed. At Grades 11–13, the key idea is to state your assumptions and use the propagation rule required by your course or laboratory.

Worked example: density

A metal block has mass $m = 125.0 \pm 0.2 g$ and volume $V = 46.2 \pm 0.5 c m^{3}$ . Its density is $ρ = m / V$ , giving about 2.706 g cm^-3.

Using the conservative percentage rule, the relative uncertainty is approximately 0.2/125.0 + 0.5/46.2, or about 1.24 percent. The absolute uncertainty in density is therefore about 0.034 g cm^-3, so a reasonable report is 2.71 ± 0.03 g cm^-3.

Using independent standard uncertainties with root-sum-square propagation would produce a slightly smaller combined uncertainty. The important point is that the calculation method must match the meaning of the input uncertainties.

Significant Figures and Reporting

Uncertainty determines which digits in a reported result are meaningful. A practical reporting convention is:

Round the uncertainty to one significant figure, or sometimes two when the first digit is small or when greater detail is justified.
Round the measured value to the same decimal place as the uncertainty.
Keep extra digits during calculations and round only the final result.

For example, a calculator result of 9.813742 with an estimated uncertainty of 0.126 might be reported as 9.81 ± 0.13. Reporting 9.813742 ± 0.13 suggests unsupported precision.

Graphs, Error Bars, and Model Testing

Graphs are powerful because they show trends, scatter, outliers, and the relationship between measurements and a model. Plot measured quantities with units on both axes and include uncertainty when it is relevant.

Error bars show the uncertainty or variability associated with a plotted point. They can be vertical, horizontal, or both. Always state what an error bar represents, such as an instrumental uncertainty, one standard deviation, a standard error, or a confidence interval.

Datei:Graph with Data Points and Error Bars (NIH BioArt 722).svg

Do not use a simple rule such as "overlapping error bars mean no difference" as a universal statistical test. The meaning of overlap depends on what the bars represent, the experimental design, and the analysis.

Gradient uncertainty

For a linear relationship, many school practicals estimate gradient uncertainty by drawing a best-fit line and then the steepest and shallowest reasonable lines consistent with the error bars. If their gradients are $m_{\max}$ and $m_{\min}$ , a graphical estimate is

$u_{m} \approx \frac{m_{\max} - m_{\min}}{2}$ .

Computer regression can provide more formal estimates, but the result depends on the statistical model and assumptions. Never copy a software uncertainty without understanding what it represents.

Designing Better Experiments

Uncertainty analysis is most useful before and during an experiment, not only after it. Use it to decide where improvement matters.

A strong plan follows this logic:

Measurand: Define exactly what quantity you are trying to determine.
Measurement model: Write the equation connecting measured inputs to the desired result.
Instrument: Choose a range and resolution appropriate to the task.
Calibration: Check zero, reference values, and known offsets.
Repeatability: Collect enough repeats to reveal meaningful scatter.
Control variables: Keep important environmental and procedural conditions stable.
Uncertainty budget: Estimate the important contributions and identify which one dominates.
Evaluation: Decide whether the uncertainty is small enough to answer the scientific question.

Example: measuring gravitational acceleration with a pendulum

For a simple pendulum at small amplitude, $T = 2 π \sqrt{L / g}$ , so $g = 4 π^{2} L / T^{2}$ . A strong method measures the time for many oscillations rather than one, repeats the timing, keeps the swing angle small, measures length to the centre of the bob, and checks whether the pivot or air resistance affects the model.

If timing uncertainty dominates, increasing the total timed interval can reduce its relative effect. If uncertainty in pendulum length dominates, a better length measurement may help more. If the swing angle is too large, repeating the same flawed procedure more often will not fix the model mismatch.

Advanced Extension: Uncertainty Budgets and Expanded Uncertainty

An uncertainty budget is a structured list of uncertainty sources, their estimated sizes, how they were obtained, and how strongly each affects the final result. It turns a vague statement such as "there were many errors" into a quantitative analysis.

In professional metrology, a combined standard uncertainty is often formed by propagating the standard uncertainties of the inputs. An expanded uncertainty may then be written as $U = k u_{c}$ , where $k$ is a coverage factor. Under suitable assumptions, a factor near 2 is commonly associated with a coverage probability of about 95 percent. The coverage probability must be stated or justified rather than assumed automatically.

This extension shows why the same word "uncertainty" can refer to different numerical forms. Always identify whether you are reporting a half-range, standard deviation, standard error, standard uncertainty, confidence interval, or expanded uncertainty.

Reliable reference sources

For deeper study, compare school laboratory conventions with professional metrology guidance:

JCGM International Vocabulary of Metrology: measurement uncertainty: Standard terminology for measurement science.
NIST Technical Note 1297: Guidance on Type A and Type B evaluation, combined uncertainty, propagation, and reporting.
BIPM Guides in Metrology: International guides maintained by the Joint Committee for Guides in Metrology.

Interactive Tasks

Quiz: Test Your Knowledge

What information makes a measured result more complete than a number alone? (A value with a unit and justified uncertainty) (!A value with as many decimal places as possible) (!A value without any estimate of uncertainty) (!A value copied directly from the instrument)

Which statement best describes precision? (Repeated measurements agree closely with one another) (!A result is close to a reference value) (!An instrument always reads exactly zero) (!A measurement has no uncertainty)

What can repetition most directly reveal? (Random scatter in repeated readings) (!Every possible systematic effect) (!The exact true value of the measurand) (!A perfect calibration certificate)

What is a likely consequence of an uncorrected zero offset? (A systematic shift in measured values) (!A guaranteed reduction in uncertainty) (!A random change of sign on every reading) (!A larger SI unit)

When is the standard error of the mean most relevant? (When estimating random sampling uncertainty in the mean) (!When correcting a known scale offset) (!When replacing all calibration information) (!When changing the unit of a quantity)

What happens to the relative uncertainty when the same absolute uncertainty is attached to a larger measured value? (It becomes smaller) (!It always becomes larger) (!It always becomes one hundred percent) (!It becomes identical to the unit)

Which rule is used in the conservative school method for multiplying measured quantities? (Add the relative uncertainties) (!Subtract all absolute uncertainties) (!Ignore the smaller uncertainty) (!Multiply the number of significant figures)

What should an error bar label or caption make clear? (What quantity or interval the bar represents) (!That every data point is certainly correct) (!That overlapping bars prove equality) (!That the graph contains no systematic effects)

Why should final rounding usually occur after calculations are complete? (To avoid unnecessary rounding error in intermediate steps) (!To create more significant figures) (!To remove all experimental uncertainty) (!To make every result an integer)

What is the best response when one uncertainty source dominates an experiment? (Improve the method that controls that dominant contribution) (!Repeat unrelated measurements indefinitely) (!Add extra decimal places to the answer) (!Ignore the dominant contribution)

Memory Game

Measurand	Quantity intended to be measured
Resolution	Smallest detectable change indicated by an instrument
Precision	Closeness of repeated readings to one another
Calibration	Comparison or adjustment using a reference standard
Parallax	Reading shift caused by viewing a scale from an unsuitable angle
Standard deviation	Statistical measure of the spread of repeated values
Uncertainty budget	Structured account of important uncertainty contributions

Drag and Drop

Match the correct terms.	Topic
Random scatter	Repeat measurements and analyse their spread
Zero offset	Check zero and apply a justified correction
Parallax	Read the scale perpendicular to the line of sight
Coarse resolution	Select a more suitable measuring instrument
Environmental drift	Monitor and control relevant laboratory conditions

...

Crossword Puzzle

Uncertainty	What quantifies the remaining doubt or dispersion associated with a measured result
Resolution	What instrument property describes the smallest detectable change
Calibration	What process compares a measuring system with a reference
Precision	What describes close agreement among repeated measured values
Measurand	What is the quantity intended to be measured
Parallax	What viewing effect can shift a scale reading when your eye is at an angle

LearningApps

Cloze Text

A complete measurement result includes a value, a unit, and an estimate of

. The quantity that you intend to measure is called the

. Closely grouped repeated readings show high

. A consistent shift caused by a zero offset is a

effect. Variation from reading to reading can be described as

scatter. The smallest detectable change of an instrument is related to its

. The arithmetic average of repeated readings is the

. The sample

describes the spread of repeated values. Absolute uncertainty divided by the measured value gives the

. Error bars provide a visual representation of

. A structured list of uncertainty contributions is an

. Repeating a biased method does not automatically improve its

.

Open-Ended Tasks

Easy

Measurement poster: Create a one-page poster that explains value, unit, uncertainty, accuracy, precision, and resolution with one original example of each.
Instrument photo audit: Photograph or sketch four measuring instruments available at school or home and annotate their range, smallest division or display step, and a reasonable first estimate of reading uncertainty.
Parallax demonstration: Produce a short image sequence or video showing how an analog scale reading changes with viewing angle and explain how to avoid the effect.
Reporting practice: Measure one object with a ruler at least five times and write a short laboratory note that reports the mean, spread, unit, and appropriately rounded result.

Standard

Pendulum timing study: Measure the period of a pendulum by timing different numbers of oscillations and compare the percentage timing uncertainty of each strategy.
Calibration interview: Interview a laboratory technician, science teacher, engineer, or craft professional about how they check measuring instruments and summarize the role of calibration and traceability.
Error bar investigation: Collect paired experimental data, plot the values with justified error bars, fit a trend, and explain what the graph does and does not prove.
Measurement workplace visit: Visit a school laboratory, maker space, workshop, pharmacy laboratory, engineering department, or other suitable place and document three ways measurement quality is controlled.

Advanced

Uncertainty budget project: Design an experiment to determine density, resistance, gravitational acceleration, or another derived quantity and build a quantitative uncertainty budget before collecting final data.
Propagation comparison: Analyse the same derived result using conservative worst-case propagation and root-sum-square propagation, then explain why the answers differ and which assumptions each method makes.
Model limitation investigation: Choose an experiment in which a theoretical approximation can fail, vary the relevant condition, and decide when model limitations become larger than the measurement uncertainty.
Research presentation: Produce a five-minute scientific presentation or video that compares school uncertainty conventions with professional Type A, Type B, combined standard uncertainty, and expanded uncertainty terminology.

Learning Assessment

Method critique: Given a flawed experimental procedure, identify the dominant random and systematic effects, rank their importance, and propose changes that would most improve the final result.
Uncertainty calculation: From a table of repeated measurements and instrument specifications, calculate a mean, an appropriate spread measure, relative uncertainty, and a correctly rounded reported result, explaining every choice.
Propagation transfer: Derive the uncertainty of a new calculated quantity from several measured inputs and justify whether conservative addition or root-sum-square propagation is appropriate.
Graph interpretation: Evaluate a graph containing error bars and a fitted model, estimate gradient uncertainty where appropriate, and explain which conclusions are supported and which would require further statistical evidence.
Experimental design: Plan an investigation to distinguish two competing physical models when their predicted values differ by only a few percent, including a target uncertainty and a strategy for reaching it.
Reflection on evidence: Compare two groups that measured the same quantity with different instruments and decide which result is more scientifically convincing by considering uncertainty, calibration, repeatability, and model assumptions.

Evidence of Learning

Strong evidence of learning includes:

Knowledge: You can explain uncertainty, measurand, accuracy, precision, resolution, random effects, systematic effects, Type A evaluation, Type B evaluation, and uncertainty propagation.
Calculation skills: You can calculate means, spread, relative and percentage uncertainties, and propagated uncertainties with clearly stated assumptions.
Practical skills: You can select and read instruments correctly, reduce parallax, check zero, repeat measurements appropriately, and record units consistently.
Data skills: You can plot error bars, fit and critique a trend, identify outliers responsibly, and distinguish scatter from bias.
Scientific communication: You can report a value and uncertainty with justified rounding and explain what the uncertainty represents.
Products: You can produce a laboratory report, uncertainty budget, annotated graph, experimental video, or research presentation that documents your reasoning.
Transfer: You can apply uncertainty thinking to unfamiliar experiments, engineering measurements, chemistry, biology, environmental science, and quantitative decision-making.

OERs on the Topic

The English Wikipedia article below provides an open overview of measurement uncertainty and links to related concepts such as propagation, statistics, and metrology.