Measurement Uncertainty and Significant Figures
Measure the width of this page with a ruler and you might get 12.4 cm. Measure it
again, a touch more carefully, and you get 12.5 cm. A friend reads 12.3
cm. Nobody made a mistake — a real measurement is never a single perfect number. It always
comes with a little fog around it: a small band of values the true length could really be.
That band is the measurement's uncertainty. It isn't a blunder and it isn't
carelessness — it's an honest statement of how sure you are. Every good scientist writes a result as
a best value plus-or-minus a range, like this:
\text{length} = 12.4 \pm 0.2 \ \text{cm}.
Read that out loud: "twelve-point-four, give or take nought-point-two centimetres." It says the true
length is almost certainly somewhere between 12.2\,\text{cm} and
12.6\,\text{cm}. The \pm 0.2 is not decoration —
it is part of the answer, and a number quoted without it is only half a measurement.
Uncertainty from a single reading
Suppose you only take one reading. How big is the uncertainty then? It comes from the
resolution of your instrument — the size of its smallest step. You've met resolution
already in
precision, accuracy and resolution;
here we turn it into a number.
A common rule of thumb: the uncertainty in a single reading is about half the
resolution. If the smallest step is one whole unit, the true value is at most half a step
away from the nearest mark.
\text{uncertainty} \approx \tfrac{1}{2} \times (\text{resolution}).
Worked example. A ruler is marked in millimetres, so its resolution is
1\,\text{mm}. You read a length as 84\,\text{mm}.
- Half the resolution is \tfrac{1}{2} \times 1\,\text{mm} = 0.5\,\text{mm}.
- So the result is 84 \pm 0.5\ \text{mm} — the true length lies between
83.5 and 84.5\,\text{mm}.
A finer instrument shrinks the uncertainty: a set of digital callipers reading to
0.01\,\text{mm} has a resolution of 0.01\,\text{mm},
so a single reading is uncertain by only about \pm 0.005\,\text{mm} —
a hundred times tighter than the ruler.
Uncertainty from repeats: mean and range
One reading is a lonely thing. Take the measurement several times and two good things
happen: you get a better best-estimate, and the readings themselves reveal how uncertain you are.
The best estimate is the mean (the average) of your repeats — adding
the readings and dividing by how many there are. Averaging lets the little wobbles above and below the
truth partly cancel out.
\text{mean} = \frac{\text{sum of readings}}{\text{number of readings}}.
The uncertainty from repeats comes from how spread out they are. Find the
range — the biggest reading minus the smallest — and take half of it:
\text{uncertainty} = \frac{\text{range}}{2} = \frac{\text{max} - \text{min}}{2}.
Worked example. A ball's fall time is measured five times, in seconds:
0.62, \quad 0.65, \quad 0.63, \quad 0.66, \quad 0.64.
- Mean: \dfrac{0.62 + 0.65 + 0.63 + 0.66 + 0.64}{5} = \dfrac{3.20}{5} = 0.64\,\text{s}.
- Range: largest 0.66 minus smallest
0.62 is 0.04\,\text{s}.
- Uncertainty: half the range, \dfrac{0.04}{2} = 0.02\,\text{s}.
- Result: 0.64 \pm 0.02\ \text{s}.
Imagine the fall times had been 0.62, 0.65, 0.63, \mathbf{0.98}, 0.64.
That 0.98 sticks out like a sore thumb — probably you mistimed the
stopwatch or fumbled the release. A reading that clearly doesn't belong is called an
anomaly, and the honest thing to do is discard it before you
average. Cross it out, then take the mean and range of the remaining good readings. One
slip shouldn't be allowed to poison the whole result — but be sure it's a genuine mistake, not just
a value you didn't expect.
See it move: spread becomes uncertainty
Below are six repeated readings of the same length, scattered along a number line. The
mean is the best estimate and the shaded band is the ± range the true
value could lie in. Drag the slider to spread the readings out: the more your repeats
disagree, the wider the band and the bigger the uncertainty — while the mean, your best guess, stays
put in the middle.
Significant figures: don't claim more than you know
A calculator will happily hand you 12.4375000, but your ruler never knew
those last digits — they're invented. Significant figures are the digits in a number
that actually carry information, and quoting the right number of them is how you tell the truth about
your precision.
Count significant figures with three simple rules:
- Every non-zero digit counts. 4.72 has 3 significant figures.
- Zeros between non-zero digits count. 50.3 has 3; 1002 has 4.
- Leading zeros don't count (they only place the decimal point), but trailing
zeros after the decimal do. 0.0340 has 3 significant figures — the
3, the 4 and that final 0;
the two front zeros are just spacers.
Worked example. Round 3847 to different precisions:
- to 3 significant figures: look at the 4th digit (7), round up →
3850;
- to 2 significant figures: look at the 3rd digit (4), round down →
3800;
- to 1 significant figure: look at the 2nd digit (8), round up →
4000.
The golden rule: a result should carry no more figures than the measurement justifies.
A stopwatch you thumb by hand is good to maybe a tenth of a second, so 7.4\,\text{s}
is honest but 7.418\,\text{s} is a fib — those extra digits pretend to a
precision the stopwatch simply doesn't have.
Percentage uncertainty: is ±1 big or small?
An uncertainty of \pm 1\,\text{cm} feels tiny when you're measuring a
cricket pitch and enormous when you're measuring a matchstick. To judge it fairly, compare the
uncertainty with the size of the value itself. That comparison is the percentage
uncertainty:
\text{percentage uncertainty} = \frac{\text{uncertainty}}{\text{value}} \times 100\%.
Worked example. The page measured 12.4 \pm 0.2\,\text{cm}:
\frac{0.2}{12.4} \times 100\% = 1.6\% \ \text{(to 2 s.f.)}.
Now compare two lengths, A = 50 \pm 1\,\text{cm} and
B = 5 \pm 1\,\text{cm}. They share the same
\pm 1\,\text{cm}, yet A is
\tfrac{1}{50}\times 100\% = 2\% uncertain while B is
\tfrac{1}{5}\times 100\% = 20\% uncertain. The same absolute
uncertainty is ten times more damaging on the smaller measurement — which is exactly why we
measure small things with finer instruments.
Three classic slips that make an uncertainty meaningless — dodge them all:
-
Don't quote false precision. A millimetre ruler resolves to about
0.1\,\text{cm}, so a reading is at best 12.4\,\text{cm}
— never 12.37500000\,\text{cm}. Copying a calculator's tail of digits doesn't
make you more accurate; it just makes you wrong more precisely.
-
Uncertainty is measured, not guessed. With repeat readings the uncertainty comes from
the range of what you actually got — (\text{max}-\text{min})/2 — not
from a hopeful "that feels about right" plucked from thin air.
-
Round the uncertainty to 1 significant figure, then match the value to it. If a
calculation spits out 12.437 \pm 0.213\,\text{cm}, round the uncertainty to
0.2 (1 s.f.), then give the value to the same decimal place:
12.4 \pm 0.2\,\text{cm}. Writing 12.437 \pm 0.2
mixes a precise value with a fuzzy uncertainty — nonsense.
Next time a poll reports "45% support, margin of error ± 3 points," you're looking at exactly the idea
on this page. The pollsters didn't ask everyone — they measured a sample of a few thousand
people, and a sample is like a set of repeat readings: it wobbles around the true figure. That
margin of error is their measurement uncertainty, so the real support is somewhere
between about 42\% and 48\%.
It also explains why "45% versus 44%" is a non-story: the two numbers overlap inside each other's
uncertainty, so the poll genuinely can't tell them apart. And to halve the margin of error a
pollster must ask about four times as many people — better precision costs dearly,
whether you're wielding a ruler or a clipboard.