Systematic and Random Errors

Spec mapping: OCR H556 Module 2.2 — Making measurements and analysing data, specifically the distinction between systematic and random errors, the sources of each, methods of detection, and methods of mitigation. This lesson anchors PAG 1 (measurement of physical quantities including the use of vernier callipers and micrometers). (Refer to the official OCR H556 specification document for exact wording.)

Every measurement in physics is imperfect. No matter how careful you are, no matter how expensive your apparatus, the value you record is never exactly the "true" value of the quantity being measured. Physicists are professional realists about this: they decompose the discrepancy between "measured" and "true" into two structurally distinct categories — systematic errors and random errors — and treat each in a different way. Random errors are the unavoidable scatter inherent in repeated measurement and respond to averaging; systematic errors are biases built into the apparatus or method that no amount of averaging will remove. The distinction is the foundation of every uncertainty calculation, every graphical-treatment-of-data question, and every PAG-related practical write-up in the H556 course.

This lesson lays the conceptual groundwork. By the end you should be able to: define systematic and random error precisely as OCR demand; classify a given source of measurement discrepancy as one or the other; explain how to detect each (the signature of random error is scatter, the signature of systematic error is a non-zero intercept or a consistent offset); and explain the specific mechanism by which each can be reduced. The last point is the discriminator: a top-band candidate doesn't just say "repeat and average", they say "repeat the measurement and calculate the mean, because random fluctuations are normally distributed about the true value and the standard error in the mean scales as $1/\sqrt{N}$ ".

Key definitions. A random error is the unpredictable, statistically unbiased fluctuation that causes repeated measurements of the same quantity to scatter about a central value. A systematic error is a consistent, predictable bias that shifts every measurement in the same direction by the same (or a closely related) amount.

Vocabulary OCR insists on — error, uncertainty, accuracy, precision

Before classifying errors we need a clean vocabulary. OCR mark schemes are strict on these definitions and lose top marks where students conflate them.

Error. The difference between a measured value and the true value of a quantity. An error is a specific numerical discrepancy that (in principle) exists but may not be known. "Error" in physics does not mean "mistake". A measurement subject to a systematic error is not a blunder; it is a measurement carrying an unavoidable bias that needs to be identified and corrected.
Uncertainty. The range within which the true value is believed to lie, based on the measurement procedure. It is a quantitative statement about how much confidence to place in a result. Uncertainty is normally quoted as " $x \pm \Delta x$ " — see the next lesson for the numerical treatment.
Accuracy. The degree to which a measurement is close to the true value. High accuracy means small systematic error.
Precision. The degree to which repeated measurements cluster together. High precision means small random error. Precision is independent of accuracy — a measurement can be precise (tight cluster) without being accurate (cluster in the wrong place).

A famous archery analogy makes this clear: precision is how tight the cluster of arrows is; accuracy is whether that cluster is centred on the bullseye. The two are independent.

Random errors

Random errors are fluctuations that cause repeated measurements of the same quantity to scatter about a central value. They are inherent to the measurement process and arise from:

Observer limitations. Human reaction time when starting/stopping a stopwatch (typical $\sim 0.2\,\text{s}$ ); the difficulty of judging when a needle is exactly aligned with a scale mark; reading a meniscus.
Environmental fluctuations. Draughts and vibrations affecting balance readings; temperature changes affecting resistance, gas volumes, and oscillation periods; mains-frequency electrical noise.
Random instrument noise. Thermal noise (Johnson noise) in resistors; shot noise in photo-detectors; quantisation noise in digital instruments where the least-significant bit oscillates.
Quantum noise in extreme cases. Counting statistics in radioactive decay (Poisson noise); shot noise in low-light photon counting.

Because random errors scatter both above and below the true value, they can be reduced — though not eliminated — by averaging. If you take $N$ repeated measurements, the standard error in the mean scales as $1/\sqrt{N}$ of the standard deviation of the individual measurements:

$\sigma_{\bar{x}} = \frac{\sigma}{\sqrt{N}}$

This is the central limit theorem at work. Doubling the number of measurements does not halve the random uncertainty — it reduces it only by a factor of $\sqrt{2} \approx 1.41$ . To halve the uncertainty you need four times as many measurements; to reduce it by a factor of ten, a hundred times.

Worked illustration — timing a pendulum

Imagine timing 10 oscillations of a pendulum with a hand-operated stopwatch. You record:

$14.5,\,14.8,\,14.3,\,14.7,\,14.6,\,14.4,\,14.8,\,14.5,\,14.6,\,14.5\,\text{s}$

The mean is $\bar{t} = 14.57\,\text{s}$ . No single reading equals the "true" period; but the mean is a much better estimate than any individual reading. The scatter ( $\sigma \approx 0.17\,\text{s}$ ) traces back to human reaction-time fluctuation — typically a fraction of a second per stop-start cycle. This is the classic signature of random error: scatter that averages away with repetition.

Reducing random errors — the specific mechanisms

OCR mark schemes want specific mechanisms, not vague statements like "be more careful". Internalise these:

Repeat measurements and take the mean. Mark-scheme phrase: "repeat and take the mean".
Use automated timing. Light gates eliminate human reaction time entirely — mark-scheme phrase: "use a light gate instead of a stopwatch".
Use instruments with finer resolution. A vernier calliper ( $0.1\,\text{mm}$ resolution) or micrometer ( $0.01\,\text{mm}$ ) beats a metre ruler ( $1\,\text{mm}$ ) for measuring small lengths.
Minimise environmental influences. Close doors and windows to reduce air currents; thermally insulate water-calorimetry experiments; use shielded cables to reduce electrical interference.
Take many samples and plot. A least-squares straight-line fit through many data points averages out individual random deviations.
Use multiple cycles. Timing 20 oscillations and dividing by 20 reduces the fractional uncertainty by a factor of 20 because the reaction-time error is fixed per measurement, not per cycle.

Exam tip: When asked how to reduce random error, the magic phrases are "repeat and take the mean", "use an instrument of higher resolution", "use automated timing such as light gates", and "time multiple oscillations to reduce the percentage uncertainty". These appear verbatim in many mark schemes.

Systematic errors

Systematic errors are biases that shift every measurement in the same direction by the same (or a closely related) amount. They arise from problems with the apparatus or the method, not from the random limits of reading. Typical sources:

Calibration errors. A ruler whose "zero" mark is 1 mm inside the physical end of the rule, due to wear; a balance whose tare is off by 0.05 g; a thermometer that reads 1 °C high throughout its range because of incorrect calibration at the factory.
Zero errors. An ammeter that reads +0.02 A when no current is flowing; a top-pan balance that displays 0.03 g when empty.
Parallax errors. Consistently reading a scale from a non-perpendicular angle, so that every reading is biased.
Environmental biases. A thermometer calibrated for sea-level atmospheric pressure being used at altitude; an experiment that ignores the buoyancy of air in a precision-mass measurement.
Method-inherent biases. End-correction in a resonance tube (the antinode sits above the open end); friction in a pulley that always opposes motion; the temperature increase that arises when measuring a resistance with a finite current (the resistor self-heats).
Poor technique. Always starting the stopwatch slightly late because the user lacks practice with the trigger; always overshooting a meniscus when titrating.

The defining feature of a systematic error is that it cannot be reduced by averaging. Repeating a measurement ten times with a faulty ammeter simply gives ten identically-biased readings. The mean is no closer to the truth than a single reading.

Inline decision tree — is it random or systematic?

flowchart TD
    A[Spotted a discrepancy in a measurement] --> B{Does repeating the measurement<br/>give a different answer each time?}
    B -- Yes --> C[Likely RANDOM error]
    B -- No, same offset every time --> D[Likely SYSTEMATIC error]
    C --> E[Reduce by: repeat and take mean,<br/>use finer instrument, light gates]
    D --> F{Does a graph through the origin<br/>show a non-zero intercept?}
    F -- Yes --> G[Confirms SYSTEMATIC zero error]
    F -- No --> H[Systematic error may be<br/>multiplicative or method-inherent]
    G --> I[Reduce by: calibrate against standard,<br/>subtract zero reading, change method]
    H --> I

Worked illustration — a ruler worn at the zero

Suppose you measure the length of a pencil with a plastic ruler. The "0" mark on the ruler is 1 mm inside the physical edge, due to wear. Every measurement you take will be 1 mm too long. Ten measurements averaged will still be 1 mm too long. No amount of repetition helps. The fix is to either subtract 1 mm from every reading (correct for the known offset) or to measure between two readings on the ruler (the "difference method").

Worked illustration — zero error on an ammeter

An analogue ammeter reads $+0.02\,\text{A}$ when disconnected. You then measure a current and read $0.48\,\text{A}$ . The true value is $0.48 - 0.02 = 0.46\,\text{A}$ . The systematic error is $+0.02\,\text{A}$ , and subtracting it yields the correct current.

Reducing systematic errors — the specific mechanisms

Check for zero errors before starting. Note the reading when the instrument is disconnected/unloaded, and subtract it from every subsequent measurement (or, if the instrument allows, adjust to zero).
Calibrate against a known standard. Check a thermometer against pure melting ice at $0\,^\circ\text{C}$ and boiling water at $100\,^\circ\text{C}$ (at sea level). Check a balance with traceable calibration weights.
Change the method. If a systematic error is inherent to a technique (e.g. end-correction in a resonance tube), switch to a method where the error cancels.
Use the "difference method". Measure $L_2 - L_1$ rather than $L_1$ alone, so that a constant additive offset cancels. Classic example: in a resonance-tube measurement of the speed of sound, take both first and second resonance lengths, since $\lambda/2 = L_2 - L_1$ is independent of the end-correction.
Use instruments of known accuracy. Traceable weights, calibrated multimeters, NPL-certified thermometers.
Identify and subtract a known bias. If you know the systematic error is $+0.02\,\text{A}$ , subtract it from every reading.

Visualising the difference — the dartboard analogy

The classical mental model for random-vs-systematic error is the dartboard.

flowchart LR
    A["Small random, small systematic<br/>arrows clustered at bullseye<br/>ACCURATE and PRECISE"]
    B["Small random, large systematic<br/>arrows clustered but off-centre<br/>PRECISE but NOT ACCURATE"]
    C["Large random, small systematic<br/>arrows scattered around bullseye<br/>ACCURATE on average, NOT PRECISE"]
    D["Large random, large systematic<br/>arrows scattered and off-centre<br/>NEITHER ACCURATE NOR PRECISE"]

Tight cluster at the bullseye — small random, small systematic. Good measurement.
Tight cluster off-centre — small random, large systematic. Precise but inaccurate. The measurements agree with each other but not with reality.
Spread around the bullseye — large random, small systematic. Accurate on average but imprecise on individual readings.
Spread away from the bullseye — both errors large. Useless.

The key insight: precision and accuracy are independent. A measurement can be very precise (tight cluster) but inaccurate (centred away from the bullseye), or accurate on average (centred at the bullseye) but imprecise (large spread). The two failure modes require different remedies — averaging fixes precision but not accuracy; calibration fixes accuracy but not precision.

Worked Example 1 — Classifying errors in a resistance measurement

A student measures the resistance of a resistor by repeatedly applying a voltage $V$ and measuring the current $I$ , using the same ammeter each time. The readings are:

$V = 6.00\,\text{V}$ (set precisely);
$I = 0.302,\,0.298,\,0.305,\,0.301,\,0.299\,\text{A}$ .

The manufacturer states the resistor is $20.0\,\Omega$ . The ammeter has a known $+0.010\,\text{A}$ zero error.

(a) Calculate the mean current. (b) Correct for the zero error. (c) Calculate the resistance before and after correction. (d) Classify each error type and explain how it is mitigated.

Solution.

(a) $\bar{I} = (0.302 + 0.298 + 0.305 + 0.301 + 0.299)/5 = 1.505/5 = 0.3010\,\text{A}$ .

(b) Corrected mean: $\bar{I}_{\text{corr}} = 0.3010 - 0.010 = 0.2910\,\text{A}$ .

(d) The scatter in the five current readings (range $0.298$ to $0.305$ , total spread $\sim 2\%$ ) is random error. It comes from analogue-meter reading uncertainty and the ammeter's internal noise. Random error is reduced by repeating and taking the mean, which has been done — the mean $0.3010\,\text{A}$ is a better estimate than any single reading.

The $+0.010\,\text{A}$ bias is systematic error. It is a zero error of the ammeter and shifts every reading by the same amount. Systematic error of this kind is reduced by identifying and subtracting the known bias, which is what (b) does.

Note that the "more accurate" answer ( $20.62\,\Omega$ ) is further from the manufacturer's nominal $20.0\,\Omega$ than the uncorrected mean ( $19.93\,\Omega$ ). That is actually fine — the manufacturer's tolerance is typically $\pm 5\%$ (i.e. $\pm 1.0\,\Omega$ ) and the corrected measurement is the more scientifically honest one. A naïve student might "calibrate" by adjusting the data to match the manufacturer's value; that would be experimental misconduct.

Worked Example 2 — Spotting a systematic error from a graph

A student measures the extension $x$ of a spring under various loads $F$ using a metre ruler. The data should obey Hooke's law $F = kx$ (a straight line through the origin), but the best-fit line on the student's $F$ -versus- $x$ plot intersects the extension axis at $+1.5\,\text{mm}$ .

(a) What type of error does the intercept suggest? (b) How should it be corrected?

Solution.

(a) A non-zero intercept at zero load indicates a systematic error of $+1.5\,\text{mm}$ in the extension measurement — every $x$ reading is offset by 1.5 mm. The likely cause is a zero-reading error on the ruler (the pointer was 1.5 mm above the ruler's zero mark when the spring was unloaded), or a wear/misalignment in the ruler itself.

(b) Either subtract $1.5\,\text{mm}$ from every extension reading, or repeat the experiment with the pointer carefully aligned to the ruler's zero mark when the spring is in equilibrium under no added load.

This is a classic OCR practical-paper question. The systematic error is identified from the y-intercept of the graph; the random error is identified from the scatter of the points about the best-fit line. The two analyses are independent and both should be performed.

Systematic and Random Errors

Systematic and Random Errors

Vocabulary OCR insists on — error, uncertainty, accuracy, precision

Random errors

Worked illustration — timing a pendulum

Reducing random errors — the specific mechanisms

Systematic errors

Inline decision tree — is it random or systematic?

Worked illustration — a ruler worn at the zero

Worked illustration — zero error on an ammeter

Reducing systematic errors — the specific mechanisms

Visualising the difference — the dartboard analogy

Worked Example 1 — Classifying errors in a resistance measurement

Worked Example 2 — Spotting a systematic error from a graph

Worked Example 3 — Method-inherent error in a resonance tube

More in Physics