Law of large numbers; relative frequencies settling down and fluctuating
Machine translation
Learn
Julie tosses a coin many times. Every so many tosses she notes down what percentage of heads she has had so far. 70% means 70 out of every 100, that is, 7 out of every 10. This relative frequency changes a lot at the start: after 10 tosses she has 70% and after 20, 55%. We say that it fluctuates.
Julie's percentage of heads: it goes up and down at first and almost stops changing at the end With more tosses, the changes get smaller and smaller and the relative frequency stays close to one value. We say that it settles down. Here it stays close to 50%, which is what you expect with an ordinary coin. The law of large numbers says that the more times we repeat an experiment, the closer the relative frequency gets to the probability.
But settling down is not the same as becoming exact: after 1000 tosses Julie has 52%, not exactly 50%. Also, what gets close to 50% is the percentage, not the number of heads and tails. The difference between the two can grow, even if it is small compared with the total.
Tosses Heads Tails Difference % of heads 10 7 3 4 70% 100 54 46 8 54% 1000 520 480 40 52% The difference between heads and tails grows, and the percentage of heads gets closer to 50% The coin has no memory. After five heads in a row, the sixth toss is just as likely to be heads as tails. A run is not made up for: it gets diluted among all the tosses that come after it.
A result that surprises you with only a few tosses can be normal. With 10 tosses, 7 heads is not unusual. With 1000, 700 heads almost never happens with an ordinary coin. That is why, with more repeats, the relative frequency is a more reliable estimate of the probability.
Worked examples
Example 1
Gus does the same experiment with another coin. This is his graph of the percentage of heads. What happens at the start, and what value does it settle down to?
- After 10 tosses he has 30% and after 20, 40%: at the start it changes a lot.
- From 100 tosses onwards it hardly changes: 47%, 49%, 49% and 49%.
- It stays close to 50%.
It fluctuates at the start and settles down close to 50%.
Watch out! Here many people say that it reaches exactly 50%, but on the graph it stays at 49%.
Example 2
Tony tosses an ordinary coin and gets 5 heads in a row. He says: “Now it has to be tails, to make up for it.” What do you tell him?
- The coin has no memory: every toss is a new one.
- Heads and tails are still equally likely on the sixth toss.
- The 5 heads are not made up for: they get diluted among the tosses that come, until they count for very little.
On the sixth toss, heads and tails are equally likely.
Watch out! Here many people believe that, after a run, the other result is due.
Example 3
Gus gets 6 heads in 10 tosses. Julie gets 600 heads in 1000 tosses, with another ordinary coin. Who has more reason to doubt their coin?
- 6 out of 10 is only one head more than the 5 you expect: it happens often.
- In 1000 tosses you almost always get between 450 and 550 heads: 600 is way outside that.
- Both have 60%, but with 1000 tosses that 60% is much less believable as a coincidence.
Julie has more reason to doubt.
Watch out! Here many people say that both should doubt equally, because both have 60%.
A challenge to think about
Toss a coin 40 times. Every 10 tosses, note down the percentage of heads you have so far and draw the graph. Compare it with someone else's: how are they alike and how are they different? What do you think would happen if you tossed 1000 times?
Practise
Rounds of 10 questions. Choose a level:
Your grade in this topic:No grade yetDo a round at the Basic level to get a grade.