Relative frequency: probability from an experiment
In 8.1 you found probabilities by counting equally likely outcomes — that is theoretical probability. But what is the probability that a drawing pin lands point-up? That a biased die shows a six? That a bus arrives late? There is no symmetry to count, so theory is silent. The only way in is to do the experiment and see how often the event actually happens:
Each repeat of the experiment is called a trial. Drop a drawing pin 200 times and see it land point-up 124 times, and your estimate is 124/200 = 0.62. The word estimate matters: run the same 200 drops again and you might get 118 or 131 — relative frequency wobbles from experiment to experiment, while a theoretical probability never moves.
Example 1A four-sided spinner numbered 1 to 4 is spun 200 times. The results are: 1 lands 44 times, 2 lands 61 times, 3 lands 58 times, 4 lands 37 times. (a) Find the relative frequency of landing on 3. (b) Do the results suggest the spinner is fair?
More trials, better estimate
Flip a coin 10 times and 7 heads is nothing unusual — relative frequency 0.7, miles from the true 0.5. Flip it 1000 times and the relative frequency will almost certainly sit very close to 0.5. Small experiments wobble; big experiments settle. This is the informal law of large numbers: as the number of trials grows, the relative frequency settles down towards the true probability.
Here is one real-style running record of coin flips. Watch the estimate wobble early on, then calm down:
| Flips so far | 10 | 20 | 50 | 100 | 200 | 500 | 1000 |
|---|---|---|---|---|---|---|---|
| Heads so far | 7 | 12 | 28 | 53 | 104 | 253 | 497 |
| Relative frequency | 0.7 | 0.6 | 0.56 | 0.53 | 0.52 | 0.506 | 0.497 |
The estimate settles towards the true probability
Example 2Three students flip the same coin. Ali: 13 heads in 20 flips. Bea: 54 heads in 100 flips. Carlos: 262 heads in 500 flips. (a) Whose result gives the best estimate of P(head)? (b) Write down the best possible estimate using all the data.
Experiment analyser
Type in your results so far, compare the estimate with theory, and predict a future count. Try 7 heads in 10 flips, then 497 in 1000 — watch the verdict change.
Expected results: probability × number of trials
Relative frequency turns an experiment into a probability. Expected frequency runs the same idea the other way: it turns a probability into a predicted count. If one trial succeeds with probability p, then in n trials you expect about p × n successes.
The probability you multiply by can come from theory (1/6 for a six on a fair die) or from an experiment (0.62 point-up from your drawing-pin data). Paper 4 loves chaining the two: estimate a probability from one data set, then use it to predict a count in a bigger population — see Example 4.
Example 3The probability that Nadia wins a game of chess against her computer is 0.15. She plays 300 games in a year. How many games should she expect to win?
Fair or biased? Compare with the theoretical value
To judge whether a die, coin or spinner is biased, put the experimental and theoretical numbers side by side. Two equivalent comparisons work, and mark schemes accept either:
Compare probabilities
Relative frequency vs theoretical probability. A fair die gives each score with 1/6 ≈ 0.167; a relative frequency of 0.235 for a six, from many trials, is well above it.
Compare counts
Observed frequency vs expected frequency. In 600 fair rolls you expect 1/6 × 600 = 100 sixes; observing 141 is far more than 100.
| Score on die (600 rolls) | 1 | 2 | 3 | 4 | 5 | 6 |
|---|---|---|---|---|---|---|
| Observed frequency | 95 | 105 | 98 | 102 | 59 | 141 |
| Expected if fair | 100 | 100 | 100 | 100 | 100 | 100 |
| Relative frequency | 0.158 | 0.175 | 0.163 | 0.17 | 0.098 | 0.235 |
Scores 1–4 sit comfortably near 100 — that much wobble is normal. But 141 sixes and only 59 fives, in a run of 600 trials, is a large, consistent distortion: this die looks biased towards 6 and against 5. Two ingredients make the conclusion safe: the gap is big, and the number of trials is large. Either one alone is not enough.
More worked examples
Example 4A drawing pin is dropped 240 times and lands point-up 168 times. (a) Estimate the probability that this drawing pin lands point-up. (b) The pin is dropped 1000 more times. Estimate the number of times it lands point-up.
Example 5Coin A shows heads 542 times in 1000 flips. Coin B shows heads 380 times in 500 flips. For each coin, decide whether the results suggest it is biased. Justify your answers.
Your turn
Practice set
- Out of 60 seeds planted, 45 germinate. Estimate the probability that a seed of this type germinates.
- Three students spin the same spinner and record how often it lands on red: 12 out of 40 spins, 33 out of 100 spins, 130 out of 400 spins. Which result gives the best estimate of P(red), and what is that estimate?
- The probability that a train is late is 0.35. Out of 200 journeys, how many trains would you expect to be late?
- A biased spinner lands on purple 320 times in 500 spins. (a) Estimate P(purple). (b) Estimate how many times it lands on purple in 2000 spins.
- A die is rolled 300 times and shows a six 92 times. (a) How many sixes would a fair die be expected to show? (b) Do the results suggest the die is biased? Give a reason.
- 45/60 = 3/4 = 0.75
- The 400-spin experiment (most trials): 130/400 = 0.325
- 0.35 × 200 = 70 trains
- (a) 320/500 = 0.64 (b) 0.64 × 2000 = about 1280 times
- (a) 1/6 × 300 = 50 sixes (b) 92 is far more than 50, and 300 trials is a reasonable sample, so yes — evidence of bias towards six (relative frequency 92/300 ≈ 0.307 vs 1/6 ≈ 0.167)
Key points to remember
- Relative frequency = frequency ÷ total trials — an estimate of probability from an experiment, always between 0 and 1.
- Use relative frequency when theory cannot help: biased objects, drawing pins, real-world data. Use counting only for fair, equally likely outcomes.
- More trials give a better estimate — the relative frequency settles towards the true probability. Best single estimate = most trials; best of all = pool the data.
- Never average relative frequencies from different-sized experiments — add successes, add trials, divide once.
- Expected frequency = probability × number of trials, and it may come out as a decimal — it is an average.
- “Expect” means on average, not guaranteed — a one-mark favourite.
- To judge bias, compare observed with expected (or relative frequency with theory): a verdict needs a big gap and many trials, and numbers quoted in your reason.
Where next?
🎥 Take the interactive module
Probability — the full teaching module for this topic, with real past-paper questions and mark schemes.
✏️ Practise exam questions
Probability — 25 exam-style questions worth 94 marks, with full answers and the mark scheme.
Progress is saved on this device only and appears on the notes home page.