A thermometer reads slightly higher today than yesterday. Has the environment warmed, or has something changed in the measurement? Measurement uncertainty describes the doubt associated with a reported result. Scientists separate possible measurement effects from real change by defining what they measure, checking instruments, repeating observations and comparing evidence from different methods.<\/p>\n
The aim is not to pretend that every result is exact. It is to understand how much confidence the evidence supports and which conclusions remain uncertain. This matters in classroom experiments, environmental monitoring and research alike. A small numerical difference can be meaningful when measurements are well controlled, while a larger difference can mislead if the instruments, locations or procedures have changed unnoticed.<\/p>\n
A reported measurement is an estimate of a quantity under specified conditions. Uncertainty expresses the limitations of that estimate. It can reflect instrument performance, calibration, sampling, environmental conditions and the definition of the quantity itself.<\/p>\n
Uncertainty is not simply a confession that an experiment failed. A careful result with a stated uncertainty is often more useful than a confident number without one. It tells readers how the measurement can reasonably be compared with other results.<\/p>\n
NIST guidance distinguishes uncertainty from error. Error concerns the difference between a measured result and a reference or true value, while the true error may not be known. Uncertainty characterises the information available about the result rather than providing a known correction for every possible mistake.<\/p>\n
Scientists use the term measurand for the quantity intended to be measured. Defining it avoids hidden differences between observations. \u201cThe temperature of a pond\u201d is incomplete unless the depth, location, time and method are specified.<\/p>\n
A reading near a sunlit surface may genuinely differ from one in deeper shaded water. Treating those readings as repeated measurements of one identical quantity would confuse spatial variation with instrument behaviour.<\/p>\n
In an original classroom example, two groups measure the width of a leaf. One measures its widest point; another measures halfway along its length. Their results can disagree even with accurate rulers because they have defined different quantities. Agreement begins with a shared question.<\/p>\n
Random effects produce variation between repeated observations under similar conditions. Examples include small fluctuations in a signal or differences in identifying a scale position. Systematic effects can shift results in a consistent or predictable way, such as a balance that does not read zero when empty.<\/p>\n
Repeating measurements helps characterise scatter, but repetition alone does not necessarily reveal a common offset. Ten readings from the same incorrectly adjusted balance can agree closely and still be wrong in the same direction.<\/p>\n
The distinction is useful, although real measurement processes are often more complicated than one random and one systematic component. Conditions can drift over time, and an effect that looks random over one timescale may reveal a pattern over another.<\/p>\n
Precision concerns agreement among repeated results under specified conditions. Accuracy concerns closeness to the relevant true value. A set of tightly grouped readings can be precise while affected by bias.<\/p>\n
Imagine a reference object whose mass is known sufficiently well for a classroom exercise. Balance A gives scattered readings around the reference value. Balance B repeatedly gives a value two grams higher. B looks more consistent, but that consistency does not establish greater accuracy.<\/p>\n
Neither should be judged from a single reading. The useful response is to investigate the sources of variation and offset. Correcting an identified systematic effect can improve results, but the correction itself may have uncertainty.<\/p>\n
Calibration establishes a relationship between an instrument’s indications and reference values under specified conditions. It may reveal an offset, a scale problem or behaviour that changes across the instrument’s range. Calibration is not always the same operation as adjusting the instrument.<\/p>\n
A thermometer checked at one temperature has not automatically been verified across every temperature it might encounter. Likewise, a balance that behaves well with a small load may require further checks before being used near its maximum capacity.<\/p>\n
NIST’s discussion of calibration emphasises its relationship with uncertainty and measurement conditions. A calibration record is useful only when users understand its scope, the reference employed and whether the instrument has changed since the check.<\/p>\n
Resolution is the smallest change an instrument can distinguish or display under relevant conditions. A digital display with many decimal places may look impressive, but those digits do not guarantee a small uncertainty.<\/p>\n
Suppose a device displays 20.000 degrees while its calibration and operating conditions only support much coarser confidence. Reporting every digit without qualification suggests more knowledge than the measurement provides. The display format and the scientific conclusion are separate matters.<\/p>\n
Earth Site’s guide to decimals<\/a> provides the numerical background. In science, deciding how many digits to report also requires understanding the measurement process, not merely copying the screen.<\/p>\n Consider a fictional experiment comparing a sample’s length before and after treatment. The reported values are 100.0 millimetres and 100.4 millimetres. The numerical increase is 0.4 millimetres, but its meaning depends on uncertainty and on how the measurements were made.<\/p>\n If positioning the sample differently can change the reading by roughly a millimetre, the result does not provide strong evidence for a 0.4-millimetre expansion. If the procedure is controlled much more tightly and independently checked, the same difference may be convincing.<\/p>\n There is another complication: the two measurements may share a calibration offset. A constant offset can partly cancel when taking a difference, while a changing offset cannot. This is why combining uncertainties requires attention to relationships between measurements rather than blindly adding every number.<\/p>\n The right conclusion is not \u201csmall changes are meaningless\u201d. It is that the ability to detect change depends on the quality and structure of the evidence.<\/p>\n Repeated readings can help estimate variability and reduce some random contributions to an average when the assumptions are appropriate. But taking many readings from one unchanged setup does not test every source of uncertainty.<\/p>\n A second operator may position the sample differently. A different instrument may reveal a shared bias in the first. Measurements on another day can show drift or sensitivity to room conditions. Each check asks a different question.<\/p>\nA worked example: has a sample changed?<\/h2>\n
Repeats, averages and independent checks<\/h2>\n