How Fair Are Your Dice? A Simple Home Test

What you are testing

A fair six-sided die gives each face a probability of 1/6. A biased die does not: its faces might be uneven because of how it was made, because of wear, or because of deliberate loading. The question for a home test is whether your rolls are consistent with 1/6 for each face, or whether the pattern differs by more than chance would usually explain.

The catch is that a fair die rarely produces perfectly even counts. Roll it 60 times and you expect about 10 of each face, but seeing 7 of one face and 13 of another is entirely normal. So the job is not to look for any difference, but to judge whether the difference is bigger than luck alone typically gives.

Step 1: Choose how many rolls

More rolls give a more sensitive test. As a rough guide, you want an expected count of at least 5 per face, so at least 30 rolls for a six-sided die. That is a minimum. Small biases are easy to miss with a small sample, and a few hundred rolls give a much more reliable picture. If you are checking dice for a game that matters to you, plan on several hundred, and take the numbers as evidence rather than proof. The reason is the same one behind the law of large numbers: results get closer to the underlying chances as the number of trials grows.

Roll in a consistent way, on a hard, flat surface, and with enough movement that the die tumbles rather than sliding. Use the same method each time. Rolling in batches and tallying after each batch helps avoid mistakes.

Step 2: Tally the results

Make a table with one row per face and record how often each face comes up. Count carefully, and check that the tallies add up to the number of rolls.

Step 3: Compare with what you expected

For a fair die, the expected count for each face is the total rolls divided by six. With 60 rolls, that is 10 per face.

Suppose your tallies are:

The observed counts add to 60, as they should. These numbers are made up for illustration.

Step 4: One number for the overall mismatch

The usual tool is the chi-square statistic. For each face, square the difference, divide by the expected count, and add the results.

• Face 1: 4 / 10 = 0.4
• Face 2: 4 / 10 = 0.4
• Face 3: 1 / 10 = 0.1
• Face 4: 1 / 10 = 0.1
• Face 5: 16 / 10 = 1.6
• Face 6: 16 / 10 = 1.6

The total is 4.2. Larger totals mean a bigger overall mismatch.

Step 5: Decide whether it is unusual

To judge 4.2, compare it with a reference value. For a six-sided die there are five degrees of freedom (six faces, minus one because the counts must add to the total). A commonly used cut-off at the 5 percent level is about 11.07. If your statistic is above that, the pattern is unusual enough that a fair die would produce it only about 1 time in 20.

Here 4.2 is well below 11.07, so these counts give no reason to doubt the die. That does not prove the die is fair; it only says this sample did not reveal a problem.

If the result had been above the cut-off, the sensible response would be to roll more and test again rather than to condemn the die on the spot. A 5 percent threshold means that about 1 in 20 perfectly fair dice will fail a test like this by chance. Repeating the test, or running a longer one, is the way to separate a real bias from a fluke. Ideas such as confidence intervals express the same caution in another way.

Other quick checks

A few simpler checks are useful for spotting obvious problems:

• Look and feel. Casting flaws such as rounded edges, bubbles, uneven corners or off-centre pips can matter.- Opposite faces. On standard dice, opposite faces add to 7. If a die does not follow this convention, it is not necessarily unfair, but it may be non-standard.
• A repeated tendency. If one face keeps showing up in the first few dozen rolls, that is worth testing further, but not worth concluding anything about.

What the test cannot tell you

A home test has limits:

• It cannot detect a very small bias without a large number of rolls.
• It cannot tell you why a die is uneven, only that the pattern looks unlikely.
• It cannot tell you about rolling technique. A thrower who sets the die down gently can produce a pattern even with a fair die.
• Many dice slightly deviate from perfection, and for most games the effect is far too small to matter.

For a die that will be used where fairness truly matters, such as a competition, a better route is to use dice from a trusted source and roll them in a tray or tower, which tends to randomise the throw. A simple Monte Carlo simulation can also help you get a feel for how much a fair die wanders from even counts, by generating many 60-roll sessions and seeing how often the statistic exceeds 11.07.

Putting it in perspective

Most dice are fair enough for any casual game. The point of a test is to learn how much natural variation to expect, and to avoid the opposite mistake of treating ordinary luck as proof that a die is cursed. Past results do not change what a die will do next, whether it looks fair or not.