Systematic and Random Error

No measurement is ever perfect. Time a trolley down a ramp with a stopwatch, weigh a beaker on a balance, read a thermometer — do it again and you almost never get exactly the same number twice. The gap between the value you write down and the true value is called error. Not a mistake, not a blunder — just the unavoidable wobble in every real measurement.

The clever move in working scientifically is that error comes in two different kinds, and — this is the whole point — you deal with them in two completely different ways. Muddle them up and you can spend a whole afternoon "improving" an experiment without fixing what was actually wrong. Get them straight and you know exactly what to do.

Random error — an unpredictable scatter, above the true value one time, below it the next.
Systematic error — a steady bias that pushes every reading the same way, every time.

Random error: unpredictable scatter

Press a stopwatch to start and stop a race. Your thumb is a little late one go, a little early the next — you can't help it, and you can't predict it. So your times land scattered around the real time: sometimes a bit too big, sometimes a bit too small, with no pattern. That jitter is random error.

It shows up everywhere something small and uncontrollable nudges the reading: your reaction time on a stopwatch, the last flickering digit on a meter that won't sit still, tiny draughts in the room, the exact spot your eye lands on a scale. Because the pushes go both ways and roughly cancel, random error mostly affects a measurement's precision — how tightly the repeats bunch together — not where they are centred.

And that hands you the cure. Because random errors scatter symmetrically above and below the true value, if you repeat the measurement several times and take the mean, the too-highs and the too-lows cancel and the average settles close to the truth. More repeats, tighter average, less random error. Repeating is exactly how a scientist beats randomness.

Systematic error: a bias the same way every time

Now imagine your stopwatch is fine but your balance is faulty: with the pan empty it already reads 2 g. Every single mass you weigh comes out 2 g too heavy — not scattered, not sometimes-up-sometimes-down, but wrong by the same amount in the same direction every time. That is a systematic error: a consistent bias baked into the whole set of readings.

Classic sources you will meet at GCSE:

A zero error — an instrument that doesn't read zero when it should (the balance above, or an ammeter whose needle rests off the zero mark). Every reading is shifted by that offset. A zero error is the purest, simplest kind of systematic error.
An uncalibrated or badly marked scale — measuring from the 1 cm mark of a ruler instead of the very end, so every length comes out 1 cm too long.
Parallax read the same wrong way — always looking at a scale from the same slanted angle, so every reading is nudged the same amount in the same direction.

Here is the sting in the tail. Repeating a systematic error does not help at all. Weigh the mass ten times on the faulty balance and you get ten readings that are all 2 g too big — averaging them just gives you the same bias, measured very precisely. To kill a systematic error you have to find and remove the cause: check for a zero error and subtract it, re-zero or calibrate the instrument against a known standard, or line your eye up square to the scale. Fix the source, not the sums.

See it: spread versus offset

Each dot below is a single reading on a number line; the tall coloured line is the true value, and the second line is the mean of all the readings. Play with the two sliders and watch what each one does — they behave nothing alike.

Turn up the random spread: the dots fan out wider and wider, but the mean stays parked right on the true value. That's random error — messy, but it averages away.
Turn up the systematic offset: the whole cluster slides sideways and the mean marches off the true value with it. No amount of spreading, and no amount of averaging, brings it back — only removing the offset does.

That is the whole lesson in one picture: spread is random error and repeats cure it; an offset of the mean is systematic error and only fixing the cause cures it.

How each shows up on a graph

When you plot your results and draw a best-fit line, the two errors leave two completely different fingerprints — so a graph is one of the best ways to spot which one you've got. Flip the switch below between the two cases.

Random error makes the points scatter above and below the best-fit line, but the line itself still lands in the right place — same gradient, and it passes through the origin where it should. The truth is still there; it's just fuzzy.
Systematic error shifts every point the same way, so the whole line moves — most tellingly it no longer passes through the origin. A line that should go through (0,0) but cuts the axis at, say, y = 1.6 is the classic sign of a systematic error (often a zero error).

Notice how neat the systematic points look — barely any scatter — yet the line is plainly wrong. A result can be beautifully precise (tightly bunched) and still be badly biased. Tidy is not the same as true.

Anomalies: the odd one out

Sometimes one reading isn't just a little off — it's wildly off, sitting far away from all the others. A single reading that clearly doesn't fit the pattern is called an anomaly (or an outlier). Maybe you misread the scale, knocked the apparatus, or wrote 47 when you meant 74.

The rule is: investigate, don't ignore. Spot an anomaly and first ask why — a genuine mistake, or a real effect you didn't expect? If you're confident it was a one-off blunder, you may leave it out when you calculate your mean (so it doesn't drag the average around). But you never quietly delete a point just because you dislike it, and you never leave an obvious anomaly sitting in your average pretending it's fine. On a graph, an anomaly is the lone point nowhere near the best-fit line — the one you circle and check.

Repeating fixes random error, not systematic error. Taking more readings and averaging squeezes out random scatter — but on a biased instrument it just pins down the same bias more precisely. A systematic error survives any amount of averaging.
A zero error is systematic, never random. Because it shifts every reading by the same fixed amount in the same direction, you cannot average it away — you must find it (check the reading with nothing on the instrument) and correct for it.
Precise is not the same as accurate. Tightly bunched readings are precise, but if they're all bunched around the wrong value they're still inaccurate. Low scatter tells you nothing about a hidden bias.
Don't bury an anomaly. An outlier should be investigated and, if it's a genuine mistake, excluded from the mean — not silently kept, and not silently deleted.

In 2011 the OPERA experiment fired neutrinos 730 km through the rock from CERN to a detector in Italy and measured them arriving about 60 nanoseconds early — apparently faster than light. It made headlines around the world. But the answer wasn't new physics; it was a textbook systematic error. A fibre-optic cable carrying a timing signal was slightly loose, adding the same small delay to every single measurement. Because the bias was identical every time, no amount of repeating the run could reveal it — the neutrinos looked consistently, precisely too fast. Once the connector was tightened and the clocks recalibrated, the neutrinos slowed right back down to light speed. A whole "discovery" undone by one steady offset — the very reason scientists hunt so hard for systematic errors.