Every measurement you make in a chemistry lab is wrong. Not badly wrong, hopefully but wrong. A burette reading of 24.35 mL is not exactly 24.35 mL, and no amount of care will make it so. This isn’t a failure of technique. It’s a property of measurement itself.
Uncertainty in measurement is how chemists deal with that reality honestly. Instead of pretending a number is exact, we attach a range to it that says: here is our result, and here is how confident we are in it. A result reported without uncertainty is, strictly speaking, incomplete it tells you nothing about whether it can be trusted.
This guide covers what measurement uncertainty means in chemistry, where it comes from, how to calculate it step by step, and how to report it the way accredited labs do.
What Is Uncertainty in Measurement in Chemistry?
Uncertainty in measurement is a quantitative estimate of the range of values that could reasonably be attributed to the quantity being measured.
The internationally accepted definition comes from the GUM (Guide to the Expression of Uncertainty in Measurement, published by the JCGM/BIPM), which describes it as a parameter associated with the result of a measurement that characterises the dispersion of values that could reasonably be attributed to the measurand.
In plain English: when you report a concentration as 0.1024 mol/L ± 0.0006 mol/L, you’re saying the true value almost certainly lies somewhere in that interval. The ± part isn’t an admission of sloppiness. It’s a statement of scientific rigour.
A few terms worth fixing in your head early:
- Measurand – the quantity you actually intend to measure (e.g. the concentration of chloride in a water sample).
- Standard uncertainty, u(x) – uncertainty of a single input expressed as one standard deviation.
- Combined standard uncertainty, u<sub>c</sub>(y) – all the individual uncertainties merged into one figure for the final result.
- Expanded uncertainty, U – the combined uncertainty multiplied by a coverage factor, giving a stated confidence level.
Uncertainty vs Error: Why They Are Not the Same Thing
Students mix these up constantly, and it costs marks in exams and credibility in reports.
Error is the difference between your measured value and the true value. Since you almost never know the true value, error is a single fixed number you can’t actually calculate. It’s a theoretical concept.
Uncertainty is a range you can calculate, based on everything you know about your method, your instruments, and your data. It’s practical.
| Aspect | Error | Uncertainty |
|---|---|---|
| Nature | A single value, signed (+ or −) | An interval, always positive |
| Knowable? | Usually not, true value unknown | Yes, always estimable |
| Can it be corrected? | Systematic error can be corrected | Never removed, only reduced |
| Direction | Has a direction | Has no direction |
There’s also the classic pair worth revisiting:
- Random error causes scatter repeat the measurement and values fall on either side. Averaging more replicates reduces its effect.
- Systematic error (bias) pushes every result the same way an uncalibrated balance reading 0.002 g high on everything. More replicates won’t help; only calibration will.
Uncertainty accounts for both, which is exactly why it’s more useful than either concept alone.
Accuracy, Precision and Trueness
These three get thrown around loosely, so here’s the distinction that actually matters in analytical work.
Precision describes how closely repeated measurements agree with each other. High precision, tight cluster. It says nothing about correctness.
Trueness describes how close the average of many measurements comes to the accepted reference value.
Accuracy is the combination of both close together and close to the truth.
The dartboard analogy still works best: darts clustered tightly in the bottom-left corner are precise but not true. Darts scattered all around the bullseye average out true but are imprecise. Only a tight cluster on the bullseye is accurate.
Uncertainty estimation forces you to account for both the scatter (precision) and the possible bias (trueness).
Types of Uncertainty: Type A and Type B Evaluation
The GUM classifies uncertainty by how you evaluate it, not by where it comes from. This trips people up, so read carefully.
Type A Evaluation
Uncertainty evaluated by statistical analysis of repeated observations. You made the measurement several times and calculated the spread.
For n repeated measurements, the standard uncertainty is the standard deviation of the mean:
u(x) = s / √n
where s is the sample standard deviation.
Example: Five replicate titrations give 24.32, 24.38, 24.35, 24.30, 24.36 mL. The mean is 24.342 mL and s = 0.0311 mL. So u = 0.0311 / √5 = 0.0139 mL.
Notice the useful consequence: quadrupling the number of replicates halves this component. That’s why repeat measurements matter.
Type B Evaluation
Uncertainty evaluated by any means other than repeated measurement calibration certificates, manufacturer tolerances, handbook data, or professional judgement.
You typically convert a stated tolerance into a standard uncertainty using an assumed distribution:
| Distribution | When to use | Divide the half-width by |
|---|---|---|
| Rectangular (uniform) | Manufacturer states limits ±a with no other info | √3 ≈ 1.732 |
| Triangular | Values near the centre more likely (e.g. volumetric glassware) | √6 ≈ 2.449 |
| Normal (from certificate) | Certificate gives U with k = 2 | 2 |
Example: A Class A 250 mL volumetric flask is marked ±0.12 mL. Assuming a triangular distribution, u = 0.12 / √6 = 0.049 mL.
Key point: Type A and Type B are not “random” and “systematic”. A systematic effect can be evaluated statistically, and a random effect can be estimated from a certificate. The labels refer purely to method of evaluation.
Common Sources of Uncertainty in Chemical Analysis
Before calculating anything, you have to find every contributor. The standard approach is a cause-and-effect (fishbone) diagram with four or five main branches. Typical sources include:
Instrumental
- Balance readability, linearity, and calibration drift
- Volumetric glassware tolerance (pipettes, burettes, flasks)
- Spectrophotometer wavelength accuracy and photometric noise
- pH meter calibration and electrode response
Reagent and sample
- Purity of primary standards (e.g. 99.95% ± 0.05%)
- Molar mass uncertainty from IUPAC atomic weight ranges
- Sample inhomogeneity and subsampling
- Stability and degradation over time
Environmental
- Temperature effects on solution volume a 5 °C deviation from calibration temperature shifts an aqueous volume by roughly 0.01% per °C
- Humidity affecting hygroscopic solids
- Air buoyancy in high-accuracy weighing
Operator and method
- Endpoint judgement in visual titrations
- Parallax in reading a meniscus
- Rounding and calculation approximations
- Recovery and matrix effects in a method
Not all of these will matter. Part of the skill is recognising which contributors are negligible if one component is less than a third of the largest one, its contribution after squaring is usually trivial.
Absolute Uncertainty vs Relative Uncertainty
Absolute uncertainty carries the same units as the measurement: 25.00 ± 0.03 mL.
Relative uncertainty is the absolute uncertainty divided by the value, usually expressed as a percentage or in parts per thousand:
Relative u = u(x) / x
For that pipette: 0.03 / 25.00 = 0.0012, or 0.12%.
Why bother with both? Because the rules for combining uncertainties depend on the operation. Addition and subtraction use absolute uncertainties. Multiplication and division use relative ones. Getting this backwards is the single most common calculation error in student work.
Relative uncertainty is also how you compare contributors fairly. A ±0.0002 g balance uncertainty sounds tiny but on a 0.0050 g sample it’s 4%, which will dominate everything else in your budget.
How to Calculate Uncertainty in Measurement: Step by Step
The GUM process has four stages. Follow them in order.
Step 1 – Specify the measurand and write the equation
State clearly what you’re measuring and express it as a formula. For a standard solution prepared by weighing:
c = (m × P) / (M × V)
where c is concentration (mol/L), m is mass (g), P is purity, M is molar mass (g/mol), and V is volume (L).
Step 2 – Identify all uncertainty sources
Build your fishbone diagram. List every input in the equation, then every effect on each input.
Step 3 – Quantify each component as a standard uncertainty
Convert everything to u(x) at one standard deviation. Type A from replicates, Type B from tolerances divided by the appropriate factor. Everything must be on the same footing before you can combine it.
Step 4 – Combine the components
Use the law of propagation of uncertainty (covered in the next section), then multiply by a coverage factor to get the expanded uncertainty.
Finally, round and report.
Propagation of Uncertainty: Combining Multiple Sources
Uncertainties combine in quadrature squares added, then square rooted. They do not simply add, because independent errors partially cancel.
For addition and subtraction
If y = a + b − c, combine absolute uncertainties:
u<sub>c</sub>(y) = √[ u(a)² + u(b)² + u(c)² ]
Example: Weighing by difference. Container + sample = 12.4527 g, container = 11.9483 g, each with u = 0.0001 g. Sample mass = 0.5044 g, u = √(0.0001² + 0.0001²) = 0.00014 g.
For multiplication and division
If y = (a × b) / c, combine relative uncertainties:
u<sub>c</sub>(y)/y = √[ (u(a)/a)² + (u(b)/b)² + (u(c)/c)² ]
For powers
If y = aⁿ, then u<sub>c</sub>(y)/y = |n| × u(a)/a. Squaring a value doubles its relative uncertainty.
The general rule
For any function, the combined standard uncertainty is:
u<sub>c</sub>(y) = √[ Σ (∂f/∂x<sub>i</sub>)² · u(x<sub>i</sub>)² ]
The partial derivative ∂f/∂x<sub>i</sub> is called the sensitivity coefficient it tells you how strongly the result responds to each input. The two simpler rules above are just special cases of this.
Expanded Uncertainty and the Coverage Factor
The combined standard uncertainty corresponds to roughly 68% confidence. That’s rarely enough for a report, a certificate, or a regulatory submission.
Multiply by a coverage factor, k:
U = k × u<sub>c</sub>(y)
| Coverage factor | Approximate confidence level |
|---|---|
| k = 1 | 68% |
| k = 2 | 95% |
| k = 3 | 99.7% |
k = 2 is the default convention in analytical chemistry and testing laboratories. It assumes an approximately normal distribution and sufficient degrees of freedom. When degrees of freedom are low, k is derived from the Student’s t distribution using the Welch–Satterthwaite formula instead.
Always state the coverage factor. “±0.0006 mol/L” is ambiguous; “±0.0006 mol/L (k = 2, approximately 95% confidence)” is complete.
Worked Example: Uncertainty in Preparing a Standard Solution
Let’s put it together. We prepare a sodium carbonate standard by weighing 0.5000 g of Na₂CO₃ into a 250 mL Class A volumetric flask.
Equation: c = (m × P) / (M × V)
Given values:
- m = 0.5000 g
- P = 0.999 (99.9% purity, stated as ±0.1%)
- M = 105.99 g/mol
- V = 0.2500 L
Nominal result: c = (0.5000 × 0.999) / (105.99 × 0.2500) = 0.018851 mol/L
Component 1 – Mass. Balance tolerance ±0.0002 g, rectangular distribution: u = 0.0002 / √3 = 0.000115 g per weighing. Weighing by difference means two readings: u(m) = √2 × 0.000115 = 0.000163 g Relative: 0.000163 / 0.5000 = 0.000327
Component 2 – Volume. Flask tolerance ±0.12 mL, triangular distribution: u = 0.12 / √6 = 0.049 mL Relative: 0.049 / 250 = 0.000196
Component 3 – Purity. Stated ±0.1%, rectangular distribution: u = 0.001 / √3 = 0.000577 Relative: 0.000577
Component 4 – Molar mass. Uncertainty around 0.0007 g/mol from atomic weight ranges. Relative: 0.0007 / 105.99 ≈ 0.0000066 negligible, drop it.
Combine in quadrature:
u<sub>c</sub>/c = √(0.000327² + 0.000196² + 0.000577²) = √(1.07×10⁻⁷ + 3.84×10⁻⁸ + 3.33×10⁻⁷) = √(4.78×10⁻⁷) = 0.000691 (i.e. 0.069%)
u<sub>c</sub> = 0.018851 × 0.000691 = 1.30×10⁻⁵ mol/L
Expand with k = 2:
U = 2 × 1.30×10⁻⁵ = 2.6×10⁻⁵ mol/L
Final report:
c(Na₂CO₃) = 0.018851 mol/L ± 0.000026 mol/L (expanded uncertainty, k = 2, approximately 95% confidence)
Look at which component dominated: purity, at 0.058% relative. Buying a higher-grade primary standard would improve this result far more than buying a better balance. That’s the real payoff of an uncertainty budget it tells you exactly where to spend effort and money.
Uncertainty of Common Laboratory Apparatus
Approximate tolerances for Class A borosilicate glassware and typical balances at 20 °C. Always check your own calibration certificate these are indicative only.
| Apparatus | Nominal capacity | Typical tolerance | Standard uncertainty (triangular) |
|---|---|---|---|
| Volumetric pipette | 10 mL | ±0.02 mL | 0.008 mL |
| Volumetric pipette | 25 mL | ±0.03 mL | 0.012 mL |
| Burette | 50 mL | ±0.05 mL | 0.020 mL |
| Volumetric flask | 100 mL | ±0.10 mL | 0.041 mL |
| Volumetric flask | 250 mL | ±0.12 mL | 0.049 mL |
| Volumetric flask | 1000 mL | ±0.40 mL | 0.163 mL |
| Graduated cylinder | 100 mL | ±1.0 mL | 0.41 mL |
| Analytical balance | — | ±0.0002 g | 0.000115 g (rectangular) |
Two practical notes. First, a burette delivers by difference, so two readings contribute combine them in quadrature. Second, graduated cylinders are roughly an order of magnitude worse than volumetric glassware. If a cylinder appears anywhere in your quantitative procedure, it will almost certainly dominate your budget.
Reporting Uncertainty and Significant Figures
Some conventions that separate a professional report from a student one:
Round the uncertainty to one or two significant figures. Two is standard practice for intermediate work; one is acceptable for a final statement. Writing ±0.0002613 mol/L implies a precision in the uncertainty estimate that doesn’t exist.
Round the result to match the uncertainty’s decimal place. If U = 0.000026, report the value to six decimals: 0.018851. A result of 0.0188513 ± 0.000026 is internally inconsistent.
Round the uncertainty upward, not to nearest. Uncertainty estimates should err on the conservative side.
Always state the coverage factor and confidence level.
Keep full precision through intermediate calculations and round only at the very end. Rounding early introduces its own error.
Correct form:
Chloride concentration: 254.3 mg/L ± 8.2 mg/L (k = 2, ≈95% confidence)
Incorrect forms:
254.32167 mg/L ± 8.24913 mg/L ← false precision 254 mg/L ± 8.2 mg/L ← mismatched decimal places 254.3 ± 8.2 ← no units, no coverage factor
Why Measurement Uncertainty Matters in Real Laboratories
This isn’t just academic bookkeeping.
Regulatory compliance. ISO/IEC 17025, the international standard for testing and calibration laboratories, requires accredited labs to estimate measurement uncertainty for the tests they perform. In India, NABL accreditation is built on the same requirement.
Conformity decisions. Suppose a drinking water limit for lead is 10 µg/L and your result is 9.6 µg/L with U = 1.2 µg/L. The measured value passes, but the uncertainty interval crosses the limit. Whether you declare a pass, a fail, or “indeterminate” is a decision rule that must be defined in advance and it’s impossible to make without an uncertainty figure.
Method comparison. Two labs report 45.2% and 46.8% for the same sample. Disagreement? If both have U = 1.5%, the intervals overlap and the results are statistically consistent. Without uncertainty, you’d be arguing over nothing.
Cost efficiency. The uncertainty budget shows exactly which component dominates. Improving anything else is wasted money.
How to Reduce Uncertainty in Chemical Measurements
Once you know your largest contributor, these are the levers worth pulling:
- Increase replicates. Type A uncertainty falls as 1/√n. Going from 3 to 12 replicates halves it though the returns diminish quickly after that.
- Calibrate against certified reference materials. Removes bias and gives you a documented, traceable uncertainty.
- Use the largest practical sample size. A ±0.0002 g balance uncertainty is 0.04% on a 0.5 g sample but 4% on a 0.005 g sample.
- Pick the right glassware. Volumetric over graduated, every time, in quantitative work.
- Control temperature. Let solutions equilibrate to the calibration temperature before making up to the mark.
- Use instrumental endpoint detection. A potentiometric titration removes the operator’s colour judgement entirely.
- Work in the middle of the calibration range. Instruments are least reliable at the extremes of their working range.
- Maintain equipment properly. A clean, correctly drained pipette performs to specification; a dirty one doesn’t.
Frequently Asked Questions
What is uncertainty in measurement in simple words?
It’s the range within which the true value of a measurement is likely to lie. Rather than claiming a result is exactly 25.00 mL, you state it as 25.00 ± 0.03 mL acknowledging that the true value sits somewhere in that window.
What is the difference between uncertainty and error in chemistry?
Error is the actual difference between your measured value and the true value a single number you usually can’t determine because the true value is unknown. Uncertainty is a calculable range that expresses how much your result might reasonably vary. Error is theoretical; uncertainty is practical.
How do you calculate uncertainty in measurement?
Define the measurand and its equation, identify every source of uncertainty, convert each to a standard uncertainty (Type A from statistics, Type B from tolerances and certificates), combine them in quadrature, then multiply by a coverage factor usually k = 2 for 95% confidence.
What is the difference between Type A and Type B uncertainty?
Type A is evaluated by statistical analysis of repeated measurements essentially a standard deviation. Type B is evaluated by any other means, such as calibration certificates, manufacturer specifications, or published data. The distinction is about the method of evaluation, not whether the effect is random or systematic.
Why do we use k = 2 for expanded uncertainty?
For a normal distribution, roughly 95% of values fall within two standard deviations of the mean. Multiplying the combined standard uncertainty by 2 therefore gives about 95% confidence, which has become the standard convention in analytical chemistry and accredited testing.
What is the uncertainty of a 25 mL pipette? A Class A 25 mL volumetric pipette typically has a tolerance of ±0.03 mL. Assuming a triangular distribution, the standard uncertainty is 0.03/√6 ≈ 0.012 mL. Always check the actual calibration certificate for your pipette.
Can measurement uncertainty ever be zero?
No. Every measurement carries uncertainty from the instrument, the method, and the environment. It can be reduced through better equipment and technique, but it can never be eliminated. A reported uncertainty of zero indicates an incomplete analysis, not a perfect measurement.
Do uncertainties add or combine in quadrature?
They combine in quadrature you square each component, sum the squares, and take the square root. Simple addition would overestimate the total, because independent errors tend to partly cancel rather than reinforce each other.
What is a sensitivity coefficient?
It’s the partial derivative of the measurement equation with respect to one input variable. It quantifies how much the final result changes when that input changes, and it weights each component’s contribution to the combined uncertainty.
Is uncertainty in measurement required for ISO/IEC 17025?
Yes. ISO/IEC 17025 requires accredited testing and calibration laboratories to identify uncertainty contributions and estimate measurement uncertainty for their methods. It’s a core requirement of accreditation, including under NABL in India.
Conclusion
Uncertainty in measurement isn’t a confession that your work is unreliable. It’s the opposite it’s the evidence that you understand your method well enough to know its limits.
The core ideas are straightforward once you’ve used them a few times. Distinguish uncertainty from error. Classify each component as Type A or Type B and convert everything to a standard uncertainty. Combine in quadrature, using absolute values for sums and relative values for products. Expand with k = 2 and report the coverage factor alongside the result.
What makes the exercise genuinely useful is the uncertainty budget itself. In the worked example above, purity contributed nearly three times more than the volumetric flask information that would be invisible without the calculation, and information that tells you precisely where the next improvement should come from.
Start with your most routine analysis. Build the budget once. You’ll almost certainly find the dominant contributor isn’t the one you expected, and after that the whole process becomes second nature.
