Probability simulation classroom activity with dice

By Jude Wallis · Published

Students estimate the probability of at least one six in four rolls of a die by rolling four dice many times and pooling results, then compute it exactly as 1 - (5/6)^4 = 671/1296, about 0.518. Many students first guess 4/6, and the simulation is what shows that guess is too high.

AP Statistics: Unit 2 (topics 2.3 Estimating Probabilities Using Simulation, 2.7 Independent Events and Unions of Events, 2.10 The Binomial Distribution). Students estimate a probability by simulation (topic 2.3), compute it exactly by multiplying probabilities of independent events and taking a complement (topic 2.7), and see the count of sixes as a binomial random variable (topic 2.10).

What the activity shows

The question for the period is short: roll a fair die four times. What is the probability of getting at least one six?

A common first answer is 4/6≈0.6674/6 \approx 0.667, from adding 1/61/6 four times. The correct answer is about 0.518. The two are far enough apart that a class can tell them apart with dice in ten minutes, and close enough that a single pair of students often cannot. That contrast is the lesson: a simulation estimates a probability, the estimate improves as trials accumulate, and an exact calculation settles what the simulation could only suggest.

Run it when students are learning to estimate probabilities by simulation, before or alongside the complement rule.

Materials and timing

  • Four six-sided dice and a cup per pair. One die rolled four times also works.
  • A tally sheet with two columns: at least one six, and no six.
  • A table on the board with one row per pair: successes, trials, running total of successes, running total of trials, and running proportion.
  • A set of axes on the board for the running proportion, with total trials across the bottom and proportion from 0.3 to 0.8 up the side. Draw light horizontal lines at 0.5 and at 0.667.

Plan on 40 minutes: 5 for predictions, 10 to roll, 8 to pool and plot, 10 for the exact calculation, and 7 for discussion.

Steps

  1. Pose the question and have every student write a predicted probability. Collect a few on the board without comment.
  2. Define one trial: shake four dice, tip them out, and record whether at least one shows a six. The number of sixes does not matter, only whether there is one.
  3. Each pair runs 25 trials and computes its own estimate: successes divided by 25.
  4. Pairs report one at a time. After each report, add to the running totals and compute the running proportion.
  5. Plot the running proportion against total trials after each pair. Connect the points.
  6. Ask where the graph is settling, and whether it supports 0.667.
  7. Compute the exact probability together. The complement of at least one six is no six on any of the four rolls. The rolls are independent, so P(no six)=(5/6)4P(\text{no six}) = (5/6)^4 and P(at least one six)=1−(5/6)4P(\text{at least one six}) = 1 - (5/6)^4.
  8. Draw a horizontal line on the graph at the exact value and compare it with the class estimate.

What students should see

The exact value. (5/6)4=625/1296≈0.4823(5/6)^4 = 625/1296 \approx 0.4823, so the probability of at least one six is 671/1296≈0.5177671/1296 \approx 0.5177.

Single pairs disagree. An estimate from 25 trials has a standard deviation of 0.5177(0.4823)/25≈0.100\sqrt{0.5177(0.4823)/25} \approx 0.100. Pair estimates from about 0.32 to 0.72, which is 8 to 18 successes out of 25, are all ordinary. A pair gets 17 or more successes, an estimate of 0.68 or higher that appears to confirm the wrong answer, with probability 0.076. A pair gets 12 or fewer successes, an estimate under one half, with probability 0.429.

The pooled estimate settles. With 15 pairs the class has 375 trials, and the standard deviation of the estimate falls to 0.5177(0.4823)/375≈0.026\sqrt{0.5177(0.4823)/375} \approx 0.026. The class estimate will usually land within about 0.05 of 0.518, which puts it clearly below 0.667. The running-proportion graph swings widely for the first few pairs and flattens as trials accumulate. That flattening is the law of large numbers.

Where 4/6 comes from. The value 4×1/6=0.6674 \times 1/6 = 0.667 is real, but it answers a different question: it is the expected number of sixes in four rolls. The number of sixes is binomial with n=4n = 4 and p=1/6p = 1/6:

Sixes in four rollsProbability
0625/1296 = 0.4823
1500/1296 = 0.3858
2150/1296 = 0.1157
320/1296 = 0.0154
41/1296 = 0.0008

The mean of this distribution is (500+2⋅150+3⋅20+4⋅1)/1296=864/1296=0.667(500 + 2 \cdot 150 + 3 \cdot 20 + 4 \cdot 1)/1296 = 864/1296 = 0.667. A trial with two sixes counts twice toward that mean and only once toward the event at least one six, which is why the probability is lower than the mean count.

Discussion questions

  1. Your pair's estimate and the class estimate both came from the same dice. Which do you trust more, and why?
  2. One pair got 17 successes in 25. Did they do something wrong? No. That result has probability about 0.076, so in a class of 15 pairs it would be unsurprising to see it once.
  3. Adding 1/6 four times gives 4/6. Adding it seven times gives 7/6. What does that tell you about the method? A probability cannot exceed 1, so adding is not valid here. The four events overlap, because more than one roll can be a six.
  4. Why was it easier to compute the chance of no six than the chance of at least one six directly? No six is a single case: every roll must avoid a six, so the four probabilities multiply. At least one six covers one, two, three or four sixes, which are separate cases to add.
  5. Did the simulation prove the answer is 0.518? No. It gave an estimate near 0.52 and ruled out 0.667. The calculation gives the exact value.
  6. If the class ran 3,000 trials instead of 375, what would change about the graph and what would stay the same?

Extension

The second bet. Roll two dice 24 times. What is the chance of at least one double six? The same faulty reasoning gives 24×1/36=0.66724 \times 1/36 = 0.667 again. The exact answer is 1−(35/36)24≈0.49141 - (35/36)^{24} \approx 0.4914, just under one half. With 25 rolls it is 1−(35/36)25≈0.50551 - (35/36)^{25} \approx 0.5055, just over. This pair of bets is a famous seventeenth-century gambling problem.

Could dice settle the second bet? The gap between 0.4914 and 0.5 is 0.0086. For two standard deviations of the estimate to fit inside that gap, a simulation needs (1/0.0086)2(1/0.0086)^2, more than 13,000 trials of 24 rolls each. No class can roll that many. Here calculation succeeds where a hand simulation cannot, which is a good place to discuss what simulations are for.

Check with a calculator. In the binomial probability calculator, set n to 4 and p to 0.166667 and ask for at least 1 success. For how to design a simulation with random digits when no dice are available, see how to design and run a simulation.

The exact probability of at least one six in four rolls

A fair die is rolled four times. Find the probability that at least one roll is a six.

  1. Write the complement: the opposite of at least one six is no six on any roll.

  2. One roll avoids a six with probability 5/65/6.

  3. The four rolls are independent, so multiply: P(no six)=(5/6)4=625/1296≈0.4823P(\text{no six}) = (5/6)^4 = 625/1296 \approx 0.4823.

  4. Subtract from 1: P(at least one six)=1−625/1296=671/1296≈0.5177P(\text{at least one six}) = 1 - 625/1296 = 671/1296 \approx 0.5177.

The probability is 671/1296, about 0.518. It is a little better than an even chance, and well below the tempting answer of 4/6.

How close should the class estimate be?

Example data, not from a real class: 15 pairs run 25 trials each and record 201 successes in 375 trials. Find the class estimate and judge whether it is consistent with the exact probability, 0.5177, and with the guess of 0.667.

  1. Class estimate: 201/375=0.536201/375 = 0.536.

  2. Standard deviation of an estimate from 375 trials when the true probability is 0.5177: 0.5177(0.4823)/375≈0.0258\sqrt{0.5177(0.4823)/375} \approx 0.0258.

  3. Distance from the exact value: 0.536−0.5177=0.01830.536 - 0.5177 = 0.0183, which is about 0.7 standard deviations. That is an ordinary result.

  4. Distance from the guess: 0.667−0.536=0.1310.667 - 0.536 = 0.131. If the true probability were 0.667, the standard deviation would be 0.667(0.333)/375≈0.0243\sqrt{0.667(0.333)/375} \approx 0.0243, so the estimate would be about 5.4 standard deviations too low.

An estimate of 0.536 from 375 trials sits comfortably near 0.518 and is far too low to be consistent with 0.667. The simulation rules out the guess, and the calculation supplies the exact value.

Frequently asked questions

How many trials does the class need?

About 375, which is 15 pairs running 25 trials each, gives an estimate with a standard deviation near 0.026. That is enough to separate 0.518 from 0.667 clearly. A small class can have each pair run 40 or 50 trials to reach a similar total.

We have no dice. What can stand in?

A calculator command such as randInt(1,6,4) returns four simulated rolls. A deck of cards also works: use the ace through six of any one suit, draw one card, put it back, shuffle, and repeat four times, counting the six as the success.

Why do some pairs get estimates above 0.6?

With only 25 trials the estimate has a standard deviation of about 0.10, so 0.6 is less than one standard deviation above 0.518. Estimates of 0.68 or higher still happen about 8% of the time. This is the reason for pooling.

Does rolling four dice at once differ from rolling one die four times?

No. Four fair dice rolled together are four independent rolls, the same as one die rolled four times. Rolling together is faster.

Should students know the complement rule first?

It works either way. Taught first, the activity confirms the rule. Taught after, the simulation creates the need for it, since students can see 4/6 is wrong before they can say why.