- Measuring length, mass and time
- Physical quantities
- Ordering and rounding decimals · coming soon
- Recording and processing data · coming soon
- Rounding and significant figures
Two players, two different problems
Two players each throw six darts at the centre of a target. Look at where their darts land.
Which player is better? It is not an easy question. Player A's darts are all over the board, but they are spread evenly around the centre. Player B's darts are close together, but every one of them misses to the upper left. Player A has a problem with scatter. Player B has a problem with direction. They need completely different fixes. Player A needs a steadier hand. Player B needs to find out what is pulling every throw the same way.
Measurements behave exactly like darts. The centre of the target is the true value of a quantity, and every reading lands somewhere near it. The lecture on measuring length, mass and time showed how to read instruments carefully. But even careful readings miss the true value a little. This lecture is about those misses: the two kinds, how to recognise each, and how to reduce each.
Error is not the same as a mistake
In everyday English, an "error" is a mistake. In science, it means something different. The error in a reading is the difference between the reading and the true value:
Every measurement has some error, even one made perfectly. A ruler marked in millimetres cannot show a tenth of a millimetre. A hand cannot press a stopwatch at exactly the right instant. So error is not a sign that you did something wrong. It is part of measuring.
A mistake is different: misreading a scale, writing a number down wrongly, or counting 9 swings instead of 10. A mistake is not part of measuring, and it is not handled like an error. You find it, remove that reading, and measure again. Keep this difference in mind, because it matters below.
There is one more difficulty. In real experiments, you almost never know the true value. If you knew it, you would not need to measure. So you cannot simply look at your error. Instead, you learn to recognise the pattern that each kind of error leaves in your readings.
Random errors: readings that scatter
A random error makes readings scatter unpredictably above and below the true value. Each reading is a little different from the last, and you cannot say in advance whether the next one will be too high or too low.
You have met one cause already: your reaction time when you start and stop a stopwatch. Sometimes you press early, sometimes late. Other causes are judging a reading that falls between two marks, or a draught moving the pan of a balance. Small changes you cannot control, such as a slightly different angle each time you look, also cause random errors.
Random errors have one very useful property. Because they push readings both ways, they partly cancel out when you take several readings and find the mean (add the readings, then divide by how many there are). A high reading balances a low one. The more readings you average, the closer the mean usually gets to the true value. This is why scientists repeat measurements.
Systematic errors: readings that all lean one way
A systematic error pushes every reading in the same direction, usually by the same amount. All the readings are too high, or all are too low. This is player B's problem.
The most common cause is a zero error: an instrument shows a reading when it should show zero.
Every reading this balance gives will be 0.3 N too high. Other causes of systematic error are reading a scale from the same wrong angle every time, and measuring from the worn end of a ruler. A clock that runs slow causes one too.
Now the key point. Averaging does not help with a systematic error. If every reading is 0.3 N too high, the mean is also 0.3 N too high. You could take a thousand readings and the mean would still be 0.3 N too high. A systematic error has to be found and removed. Check the zero before you start. Compare the instrument with something whose value you know: this is called calibrating it. Or change the method.
That is it. That is the whole difference. Random errors scatter, and averaging helps. Systematic errors shift, and averaging does nothing. See it for yourself below.
A rod is exactly 50.0 cm long. A model student measures it many times. Choose how much random scatter there is, add a systematic error, and choose how many readings to average. Watch the mean (the triangle) and the true value (the dashed line).
Random scatter:
Systematic error: 0.0 cm
Readings averaged:
The table shows one set of a model student's readings of a rod exactly 50.0 cm long, with large random scatter (up to ±1.5 cm). The mean of more readings gets close to 50.0 cm when there is no systematic error. With a systematic error of +0.8 cm, the mean settles near 50.8 cm instead, however many readings are taken.
| Systematic error | Readings averaged | Mean (cm) | Error of the mean (cm) |
|---|---|---|---|
| none | 1 | 51.40 | +1.40 |
| none | 5 | 50.56 | +0.56 |
| none | 20 | 50.14 | +0.14 |
| +0.8 cm | 1 | 52.20 | +2.20 |
| +0.8 cm | 5 | 51.36 | +1.36 |
| +0.8 cm | 20 | 50.94 | +0.94 |
With only random scatter, one reading could be well over a centimetre out, but the mean of 20 readings landed within a few tenths of 50.0 cm. With a systematic error added, the dots moved together, and the mean moved with them. More readings made the mean steadier, but steadier around the wrong value. No number of readings can show you a systematic error by itself. You find it only by checking against something you know.
Accurate and precise are not the same
Two words describe the two kinds of error.
Readings are accurate when their mean is close to the true value. A systematic error spoils accuracy.
Readings are precise when they agree closely with each other. Random error spoils precision.
Look back at the darts. Player A is accurate on average but not precise. Player B is precise but not accurate. A good measurement needs both.
The lecture on measuring used "precision" for an instrument: the smallest difference it can show. The two meanings are linked. An instrument that reads only to the nearest centimetre cannot give readings that agree to a millimetre. But notice what neither meaning promises: being right. A stopped clock is perfectly precise, because it shows exactly the same time whenever you look. It is accurate only twice a day. A clock that runs five minutes fast is precise too, and never accurate.
A balance shows −0.2 g with nothing on it. A stone then reads 45.6 g. What is the stone's true mass?
Show the answer
45.8 g. With nothing on it, the balance reads 0.2 g too low, so every reading is 0.2 g too low. Corrected reading = reading − zero error = 45.6 − (−0.2) = 45.8 g. The tempting answer, 45.4 g, subtracts 0.2 without thinking about the direction. Always ask: does this instrument read too high or too low?
Now sort some real causes of error below. For each one, ask: does it scatter the readings both ways, shift them all one way, or is it simply a wrong reading?
Put each card in the right box. Read the reason that appears.
Nine causes of error, sorted with the reason for each.
- Random: Starting and stopping a stopwatch by hand. Your reaction is sometimes quicker and sometimes slower, so the times scatter on both sides.
- Systematic: A spring balance that reads 0.3 N with nothing on it. A zero error adds the same 0.3 N to every reading.
- Random: Reading a thermometer with your eye at a slightly different height each time. The eye is sometimes high and sometimes low, so readings scatter both ways.
- Systematic: Always reading a measuring cylinder from above the liquid level. Parallax from the same side every time shifts every reading the same way.
- Systematic: Measuring from the worn end of a ruler that has lost 2 mm. Every length comes out 2 mm too long.
- Mistake: Writing 15.6 s in the table as 16.5 s. This is a wrong record, not scatter. Repeat the reading and correct it.
- Random: A draught gently moving the pan of a balance. The draught pushes sometimes one way, sometimes the other.
- Mistake: Counting 9 swings of a pendulum instead of 10, once. A miscount is a mistake. That reading should be removed and repeated, not averaged.
- Systematic: A stopwatch that runs slightly slow. Every time it shows is too short, and by the same fraction.
How many digits should an answer have?
Errors have one more consequence: they limit how many digits you are allowed to write. Suppose you measure a room as 4.3 m long and 3.12 m wide. A calculator gives the area as 4.3 × 3.12 = 13.416 m². But the length was only measured to the nearest 0.1 m. Writing 13.416 m² claims to know the area to a thousandth of a square metre. That is information you never had.
Scientists use significant figures (s.f.) to show how precise a number is. The significant figures are the digits that carry real information. To count them:
- All non-zero digits count: 3.46 has 3 s.f.
- Zeros between other digits count: 3.02 has 3 s.f.
- Zeros at the start do not count; they only place the decimal point: 0.025 has 2 s.f.
- Zeros at the end after a decimal point count, because someone chose to write them: 12.0 has 3 s.f.
- Zeros at the end of a whole number are unclear: 1200 could have 2, 3 or 4 s.f. Standard form removes the doubt: 1.2 × 10³ has 2 s.f., and 1.20 × 10³ has 3.
For a calculation, use two rules. When you multiply or divide, give the answer to the same number of significant figures as the least precise measurement. The room's length has 2 s.f., so the area is 13 m² (2 s.f.). When you add or subtract, give the answer to the same number of decimal places as the least precise measurement. Keep all the calculator's digits until the last step, and round only once, at the end.
Worked examples
A spring balance reads +0.5 N with nothing on it. A student hangs a stone on it, and it reads 8.5 N. What is the stone's true weight?
The balance reads 0.5 N too high, so every reading is 0.5 N too high. Remove that from the reading:
= 8.5 N − 0.5 N = 8.0 N
Check. If the stone weighs 8.0 N, the faulty balance should show 8.0 + 0.5 = 8.5 N, which is what the student saw ✓. A better habit for next time: set the balance to zero before hanging anything on it.
A student times 10 swings of a pendulum five times: 15.6 s, 15.4 s, 15.8 s, 17.9 s and 15.5 s. Find the best value for the time of one swing.
Four readings are close together, between 15.4 s and 15.8 s. The fourth, 17.9 s, is about 2 s longer, more than a whole swing. Reaction time cannot explain that. The student almost certainly miscounted the swings. A reading that does not fit the pattern like this is called an anomalous reading. It is a mistake, so leave it out and, if possible, measure again.
Mean of the other four: (15.6 + 15.4 + 15.8 + 15.5) ÷ 4 = 62.3 ÷ 4 = 15.575 s. The readings were taken to 0.1 s, so write 15.6 s. One swing takes 15.6 ÷ 10 = 1.56 s.
Check. If 17.9 s had been included, the mean would be 80.2 ÷ 5 = 16.04 s, longer than four of the five readings. A mean that sits outside most of the data is a sign that a mistake is hiding in it ✓.
(a) A block has mass 25.0 g and volume 9.6 cm³. Find its density. (b) Two lengths, 12.5 cm and 3.27 cm, are laid end to end. Find the total length.
(a) Density = mass ÷ volume = 25.0 ÷ 9.6 = 2.6041… g/cm³. This is a division, so count significant figures: 25.0 has 3 s.f., 9.6 has 2 s.f. The answer gets 2 s.f.: 2.6 g/cm³.
(b) Total = 12.5 + 3.27 = 15.77 cm. This is an addition, so count decimal places: 12.5 has 1, 3.27 has 2. The answer gets 1 decimal place: 15.8 cm.
Check. In (b), the first length is only known to the nearest 0.1 cm, so the hundredths in the total cannot mean anything ✓.
Repeating and averaging reduces random error only. If the question describes a zero error, a worn ruler or a slow clock, more readings will not help. The answer is to check the zero, calibrate against a known value, or change the method.
A zero error is positive if the instrument reads above zero with nothing on it, and negative if it reads below. Always use true value = reading − zero error. With a negative zero error, subtracting a negative means adding.
An anomalous reading caused by a miscount or a misread scale is not random error, so it should not be averaged in. Leave it out, say why, and repeat the measurement if you can.
Work each one on paper first. Open a hint only when you are stuck.
-
Is each a random error, a systematic error or a mistake? (a) A metre rule that has shrunk slightly in a hot room. (b) Judging the top of a bouncing ball by eye. (c) Reading the 30 cm mark as 20 cm. (d) A thermometer that reads 1 °C in melting ice.
Hint 1
Ask: would the next reading be wrong in the same direction, in either direction, or is it just a wrong reading?
Hint 2
Melting ice is at 0 °C.
Full answer
(a) Systematic: every length is read too long, by the same fraction. (b) Random: your judgement is sometimes high and sometimes low. (c) A mistake: re-read it. (d) Systematic: a zero error of +1 °C, so every reading is 1 °C too high. Check: each systematic case has a direction you can name ✓.
-
A top-pan balance shows −0.4 g with nothing on it. (a) A beaker reads 52.1 g. What is its true mass? (b) What should the student have done before using the balance?
Hint 1
Does this balance read too high or too low?
Hint 2
True value = reading − zero error, and the zero error here is −0.4 g.
Full answer
(a) 52.1 − (−0.4) = 52.5 g. The balance reads 0.4 g too low, so the true mass is 0.4 g more than the reading. (b) Pressed the zero (tare) button with nothing on the pan, so that it read 0.0 g. Check: a 52.5 g beaker on a balance reading 0.4 g low shows 52.1 g ✓.
-
(a) How many significant figures are there in 0.0450, in 3.002 and in 7.0? (b) A car travels 125 m in 9.8 s. Find its average speed to a sensible number of significant figures.
Hint 1
Leading zeros never count. Zeros at the end after a decimal point do count.
Hint 2
Speed = distance ÷ time is a division, so use the fewest significant figures in the data.
Full answer
(a) 0.0450 has 3 s.f. (4, 5 and the final 0). 3.002 has 4. 7.0 has 2. (b) 125 ÷ 9.8 = 12.755… m/s. The time has 2 s.f., so the speed is 13 m/s. Check: 13 × 9.8 = 127.4 m, close to 125 m, as it should be for a 2 s.f. answer ✓.
-
A metal rod is known to be exactly 20.0 cm long. Two students each measure it five times. Arjun: 20.4, 20.5, 20.4, 20.6, 20.4 cm. Bea: 19.6, 20.5, 20.1, 19.8, 20.2 cm. (a) Find each student's mean. (b) Whose readings are more precise? Whose are more accurate? (c) Suggest a cause of each student's main error, and how to reduce it.
Hint 1
Precision is about how close the readings are to each other; look at the spread. Accuracy is about how close the mean is to 20.0 cm.
Hint 2
Arjun's spread is 20.6 − 20.4 = 0.2 cm. Bea's is 20.5 − 19.6 = 0.9 cm.
Full answer
(a) Arjun: 102.3 ÷ 5 = 20.46 cm, about 20.5 cm. Bea: 100.2 ÷ 5 = 20.04 cm, about 20.0 cm. (b) Arjun's readings are more precise (spread 0.2 cm against 0.9 cm). Bea's are more accurate (her mean is within 0.1 cm of 20.0 cm; Arjun's is about 0.5 cm too long). (c) Arjun has a systematic error of about +0.5 cm: perhaps he measured from the worn end of a ruler, or always read from the same angle. He should start from a clear mark and look straight down, or compare his ruler with a known length. Bea has a larger random error: perhaps she lined up the rod carelessly or looked from different angles. More careful readings and more of them would help her. Check: 20.46 − 20.0 ≈ +0.5 cm, and every one of Arjun's readings is above 20.0 cm, the fingerprint of a systematic error ✓.
In April 1990, NASA launched the Hubble Space Telescope. Two months later, NASA announced that its pictures were blurred. An investigation found that its main mirror, 2.4 m across, had been polished very carefully to the wrong shape: it was too flat near its edge. The error was only about two thousandths of a millimetre, but that was about ten times more than the design allowed.
The cause was a systematic error. The device used to test the mirror's shape while it was polished had a part set 1.3 mm out of position. Every test the polishers made agreed with that faulty device, so the mirror was made extremely precisely, to match a wrong reference. Repeating the test thousands of times could never have revealed it. Only a check against something independent could. Because the error was known exactly, it could be corrected: in December 1993, astronauts fitted correcting optics, much as glasses correct an eye.
Every measurement in physics, from a school ruler to a space telescope, faces the same two enemies. Repetition defeats one. Independent checking defeats the other.
- The error in a reading is the difference between the measured value and the true value, and every measurement has some.
- Random errors scatter readings both ways, so repeating readings and finding the mean reduces them.
- Systematic errors shift every reading the same way, so averaging cannot remove them, and they must be found by checking the zero or calibrating.
- Readings are accurate when their mean is close to the true value, and precise when they agree closely with each other.
- A mistake or anomalous reading should be left out and repeated, not averaged.
- An answer should not have more significant figures (for × and ÷) or decimal places (for + and −) than the least precise measurement allows.