Test Uncertainty: Why Two Instruments Disagree
Send your test requirement — get a tailored solution and a quote within one working day.
Two instruments measure the same quantity on the same asset and produce different values. One of them is not necessarily faulty, and repeating both usually confirms the disagreement rather than resolving it. Electrical test measurement uncertainty is the framework that explains the difference, and applying it is what allows a test team to decide whether a discrepancy matters.
The framework is not an academic exercise. It determines whether a result near an acceptance limit supports a verdict, whether two tests performed years apart can be compared, and whether a disputed measurement is worth investigating or worth accepting.
Where measurement uncertainty comes from
The sources fall into three groups. The first is the instrument itself: its accuracy specification, its resolution and its stability over the conditions in which it is used.
The second is the measurement chain: the calibration of the whole arrangement rather than the instrument, the characteristics of the leads, the coupling and the connections, and the way the quantity of interest is defined.
The third is the environment and the process: the temperature, the humidity where it matters, the stability of the applied quantity, the time at which the reading is taken, and the degree to which the act of measuring changes the quantity being measured.
In field measurements the third group usually dominates. A winding resistance measurement taken with a 0.2 percent instrument but an uncertain winding temperature has an uncertainty dominated by the temperature. The instrument’s specification describes a contribution that is real and small.
Instrument accuracy versus method uncertainty
Instrument accuracy describes how closely the instrument’s indication matches the quantity at its own terminals, under defined reference conditions. It says nothing about whether the quantity at the terminals is the quantity the engineer is interested in.
Method uncertainty covers that gap. It includes whether the connection arrangement places the measurement at the right point in the circuit, whether the test conditions match the reference conditions of the standard, whether the applied quantity is stable and known, and whether the measurement process itself perturbs the value.
The distinction explains a recurring situation: an organisation buys a more accurate instrument to resolve a discrepancy, and the discrepancy remains. The improvement addressed the smallest contributor while the dominant one, usually the conditions or the connection arrangement, was unchanged.
| Source | Typical contribution | How it is reduced |
|---|---|---|
| Instrument accuracy | Smallest in most field measurements | Better instrument, or one with a suitable range and resolution |
| Measurement chain calibration | Significant where the arrangement includes dividers, shunts or coupling | Calibrating the arrangement rather than the instrument alone |
| Temperature | Often the largest contributor for resistance and loss measurements | Measuring at comparable temperatures and recording the value used |
| Applied quantity stability | Significant where the source is a site supply or generator | Monitoring the applied quantity during the test and recording it |
| Connection or coupling arrangement | Significant in low-resistance and discharge measurements | Standardising the arrangement and recording it |
| Timing and settling | Significant where the quantity is still changing | Using a defined stabilisation criterion rather than a fixed delay |
Temperature, humidity and supply effects
Temperature affects almost every electrical quantity measured on power equipment. Resistance rises with temperature, insulation resistance falls, dielectric loss rises and battery capacity falls. The sensitivity differs, but the pattern is consistent: the measurement describes the asset at the temperature at which it was taken.
Where a correction is applied, the correction itself carries uncertainty. Correction factors are derived from relationships that hold under particular assumptions, and applying one outside those assumptions introduces an error that can exceed the effect it was intended to remove. The more reliable practice is to measure under comparable conditions and to record the temperature beside the value.
The supply introduces its own contributions. A test performed from a site generator with variable output produces a different result from one performed from a stable supply, even at the same nominal setting. Where the applied quantity is not monitored during the test, its variation appears as uncertainty in the result rather than as a measured quantity.
Reading resolution and rounding
Resolution is the smallest change the instrument can indicate; it is not the same as accuracy. An instrument that displays a value to four decimal places is not claiming accuracy to that resolution, and recording a value to that resolution implies a precision the measurement does not have.
Rounding should be applied consistently and in a way that reflects the uncertainty. Reporting a resistance to six significant figures when the measurement uncertainty is one percent is a common and avoidable error, and it invites a comparison between two values that differ only in the digits beyond the meaningful ones.
The consistent practice is to report the value to a resolution consistent with the uncertainty, and to state the uncertainty where the verdict depends on it. That combination gives the reader what they need without implying more precision than exists.
When two instruments legitimately disagree
Two instruments legitimately disagree whenever their combined uncertainties overlap. If each measurement carries an uncertainty of one percent, a difference of up to approximately two percent between them is consistent with both being correct.
The question to ask is therefore not which reading is right but whether the difference is larger than the combined uncertainty. Where it is, one of the measurements has a problem or the conditions differed. Where it is not, the disagreement is expected behaviour and neither instrument requires attention.
That reframing changes what a test team does. Instead of repeating both measurements and re-examining the instruments, the team examines the conditions and the arrangements, which is where the difference is most likely to originate. Method documentation from instrument suppliers such as Hioki and Fluke states the conditions under which a given accuracy figure applies, which is the information needed to make that comparison.
Arbitrating a disagreement in the field
The first step is to establish which factors differ between the two measurements. Temperature, applied quantity, connection arrangement, timing and duration are the candidates, and recording them in the first place is what makes the comparison possible.
The second step is to exchange instruments or probes where that is practical. If the results follow the instrument, the difference is in the instrument or its calibration. If they follow the position or the connection, the difference is in the arrangement.
The third step, where the difference persists, is to take a third measurement using a different method. A resistance measurement confirmed by a different technique, or a loss measurement confirmed by a different arrangement, resolves the question more convincingly than repeating the same method a third time.
Reporting uncertainty without confusing the client
Reporting uncertainty does not require a full budget for every measurement. What it requires is a statement of the dominant contributions and a resulting figure, together with the conditions under which the measurement was taken.
A practical form is to state the value, the conditions, the uncertainty, and the verdict with the margin. Where the margin is large relative to the uncertainty, the verdict is robust and the reader can see why. Where the margin is comparable to the uncertainty, the reader can see that the verdict depends on the measurement quality rather than on the asset.
That second situation is the one worth flagging explicitly, because it invites a decision: improve the measurement, accept the result with the uncertainty stated, or investigate the asset by another method.
Specifying acceptable uncertainty in a purchase
When specifying test equipment, the accuracy requirement should be derived from the acceptance limits the equipment will be used against, not chosen as an aspiration. An instrument whose uncertainty is a small fraction of the limit supports a verdict; one whose uncertainty is comparable to the limit does not, however good its other features.
The specification should also address the reference conditions, because an accuracy figure quoted under laboratory conditions may not hold in a substation in summer. Where the instrument is used outside those conditions, the manufacturer’s temperature coefficient is the relevant figure.
Finally, the specification should address the measurement chain rather than the instrument alone. Where the arrangement includes a divider, a shunt or a coupling, the uncertainty of the chain is what limits the result, and the chain is what needs calibration. The measurement system requirements that define how a high-voltage measuring system is qualified are set out in IEC 60060-2. The reference framework for the units and for traceability is published by BIPM, practical traceability guidance is described in the NIST handbooks, the accreditation framework that a calibration laboratory is measured against is described by ILAC and IAF, and legal metrology requirements are published by OIML. Equipment and calibration support arrangements are described under testing services and OEM/ODM solutions.
Two readings that differ by less than their combined uncertainty are not in conflict.
Send the two results with their conditions and instrument details to our engineering team and we will identify which contribution explains the difference. Precision dividers and the wider measurement range are listed under all electrical testing equipment.
FAQ
What contributes to measurement uncertainty in a field test?
The instrument’s own accuracy, the calibration of the measurement chain, the conditions under which the test is performed, the stability of the applied quantity, the connection or coupling arrangement, and the resolution with which the value is read. In most field measurements the last four contribute more than the first, which is why buying a more accurate instrument rarely resolves a disagreement.
How do instrument accuracy and method uncertainty differ?
Instrument accuracy describes how closely the instrument’s indication matches the true quantity at its terminals under reference conditions. Method uncertainty describes everything else: whether the quantity at the terminals is the quantity of interest, whether the conditions match the reference conditions, and whether the measurement process itself perturbs the value. A 0.2 percent instrument in a measurement whose method uncertainty is five percent does not produce a result accurate to 0.2 percent.
When can two instruments legitimately disagree?
Whenever their combined uncertainties overlap. Two instruments each accurate to one percent can differ by two percent on the same quantity and both be within specification. The question to ask is not which reading is right but whether the difference is larger than the combined uncertainty of the two measurements.
How should a disagreement be arbitrated in the field?
By identifying which factors differ between the two measurements, rather than by repeating both. Checking the temperatures, the applied quantities, the connection arrangements and the timing usually identifies the cause. Where the factors are the same and the difference persists, a third measurement using a different method resolves the question.
Why does uncertainty matter when the acceptance limit is far from the measured value?
It matters less, and that is the point of stating it. A result that sits comfortably inside a limit does not need a precise uncertainty statement to support a verdict. A result near the limit does, because the verdict may depend on the uncertainty rather than on the measurement. Stating the uncertainty is what allows a reader to tell which situation applies.