The Odds of Dodecahedral Craps

It is remarkable that a science which began with the consideration of games of chance should have become the most important object of human knowledge. — Pierre-Simon Laplace, 1812

craps table

At first glance, craps might seem like a game of chance in its purest form: two dice are thrown, their numbers are added, and fortunes are made or lost according to the result. Yet, beneath the apparent simplicity lies a surprisingly interesting mathematical structure. In fact, craps offers an excellent opportunity to explore several fundamental ideas in probability, including the distribution of sums, conditional probability, sequences of independent events, and expected value.

The game is particularly interesting because the significance of a particular dice total depends on the stage of play. A number that brings good fortune on one roll may become the very number a player wishes to avoid on the next.

Let’s examine the basic rules.

A Game of Two Acts

In a standard game of craps, one player, called the shooter, throws two ordinary six-sided dice. Although several players may place different wagers on the same throw, we will concentrate on the Pass Line bet, one of the most fundamental wagers in the game.

The game begins with what is known as the come-out roll. If the shooter obtains a sum of 77 or 1111, the player wins immediately. These outcomes are called naturals. Conversely, a sum of 22, 33, or 1212 results in an immediate loss; these are known as craps.

But what happens if neither of these situations occurs?

If the initial throw produces 44, 55, 66, 88, 99, or 1010, that number becomes the player’s point. The shooter must then continue rolling the dice until one of two things happens:

  • The point is rolled again, in which case the player wins.
  • A 77 appears before the point is repeated, in which case the player loses.

Any other result simply means that the game continues12.

Notice the curious role assigned to the number 77. On the come-out roll, it is a winning outcome; once a point has been established, however, it becomes the losing outcome. The dice haven’t changed, nor have their probabilities, but the rules have transformed what their results mean.

This simple observation is the key to understanding the mathematics of craps.

Why Seven Is Special

Before examining the game itself, let us consider the probability of obtaining each possible sum when two ordinary dice are thrown.

Since each die has six faces, there are

6×6=366\times 6=36

equally likely outcomes. The possible sums range from 22 to 1212, but these eleven numbers are certainly not equally likely. For example, a sum of 22 can only be obtained by rolling a 11 on each die. A sum of 1212 is equally restrictive. If, on the other hand, we draw a sample space (see Fig. 1), we can see that a sum of 77 can be obtained in six different ways:

sums table
Figure 1. This chart shows the sums of two dice simultaneously. For example, the 2 in the upper-left corner of the grid represents the outcome of getting two 1s after rolling the dice. Note that 7 in blue divides the distribution into two symmetric halves.

Consequently, 77 is the most likely sum, occurring with probability

ℙ(7)=636=16.\mathbb P(7)=\frac{6}{36}=\frac16.

The following table summarizes the elegant pattern displayed by these sums:

SumNumber of combinationsProbability
2 or 1211/361/36
3 or 1122/362/36
4 or 1033/363/36
5 or 944/364/36
6 or 855/365/36
766/366/36

The distribution is symmetric around 77 (See Fig. 2): the probabilities of obtaining 44 and 1010, for instance, are identical, as are those of obtaining 55 and 99. Moreover, the closer a sum is to 77, the more combinations can produce it.

This pattern is not merely an interesting curiosity. It determines virtually everything that follows in the game. The unusually high probability of rolling a 7 explains why it is advantageous on the come-out roll but dangerous after a point has been established.

We can now investigate precisely how likely a player is to make his point before rolling a 7.

probability distribution of the sum of two dice
Figure 2. Probability distribution of the sum obtained by rolling two fair six-sided dice. The distribution is symmetric about 7, which is the most likely outcome, with a probability of 6/36=1/6.

The Point: A Race Against Seven

Suppose the shooter establishes a point of 66. There are five combinations that produce a 66, but six combinations that produce a 7. At first, one might be tempted to conclude that the probability of winning is simply 536\frac{5}{36}. That would be incorrect, however, because the shooter is not limited to a single additional throw. He may roll the dice several times, and any result other than 66 or 77 leaves the game unresolved.

The relevant question is therefore not whether the next roll produces a 66, but whether a 66 appears before a 77. We can simplify the problem by disregarding all outcomes that do not affect the point. Eventually, one of two decisive events must occur: either the point is repeated, or a 77 appears. So this is equivalent to asking for the probability of rolling a 66 given that we’ve rolled either a 66 or a 77.

This is an application of conditional probability. Remember that P(A|B)P\left( A|B \right)is the probability that event AA occured given that event B occurred, and there is a formula for it:

P(A|B)=P(A∩B)P(B)P\left( A|B \right)=\frac{P\left( A\cap B \right)}{P\left(B\right)}

So, in our example, let A be the event that the sum of the dice is 66, and let BB be the event that the sum of the dice is either 66 or 77. We wish to determine P(A|B)P\left( A|B \right). Then, A∩BA\cap B is simply AA. Thus, P(A∩B)=536P\left( A\cap B \right)=\frac{5}{36} and P(B)=536+636=1136P\left( B \right)=\frac{5}{36}+\frac{6}{36}=\frac{11}{36} since five combinations are producing 66 and six producing 77. Consequently, the probability that the point occurs first is3

P(A|B)=5361136=511.P\left( A|B \right)=\frac{\frac{5}{36}}{\frac{11}{36}}=\frac{5}{11}.

The same reasoning applies to every possible point:

PointWays to obtain pointWays to obtain 7Probability of making point first
4 or 10363/9=1/33/9=1/3
5 or 9464/10=2/54/10=2/5
6 or 8565/115/11

An interesting consequence emerges: a point of 66 or 88 gives the player a better chance of winning than a point of 44 or 1010. This is because the former numbers have more possible combinations, bringing them closer in frequency to the dreaded 7.7.

The argument can be made more rigorous by examining all possible sequences of non-decisive rolls, along with a general counting argument to determine the different ways a sum can be achieved. This leads us naturally to a geometric series and counting technique that will be useful for exploring other variants (e.g., using dodecahedrons as dice instead of cubes), which we will apply in the following calculations.

Let ss be the sum of the numbers on the faces of two dice. Define the function D:{2,3,...,12}D:\{ 2,3,…,12 \}→\to {1,2,3,4,5,6}\{ 1,2,3,4,5,6 \} by the rule: For any given sum ss,

D(s)={s−1if 2≤s≤713−sif 7≤s≤12D(s) = \begin{cases} s-1 & \text{if } \: 2\le s\le 7 \\ 13-s & \text{if } \: 7\le s\le 12 \\ \end{cases}
=The number of ways to obtains=\text{The number of ways to obtain}\: s

Why, then, you may ask, does this function give the correct number of ways in which a sum ss can be obtained from some dice throw? To answer this question, remember that we wish to count the number of integer pairs(n,m)∈{1,2,3,4,5,6}2 \left( n,m \right)\in \{ 1,2,3,4,5,6 \}^{2} with

n+m=s,1≤n≤6,1≤m≤6,2≤s≤12n+m=s, \:\:\:1\le n\le 6,\:\:\: 1\le m\le 6,\:\:\: 2 \le s \le 12

If s≤7s\le 7, then both nn and mm may be at most s−1s-1, which is 66 at most (the upper bound comes from m=s−n≥1⇒s−1≥nm=s-n\ge1 \Rightarrow s-1\ge n ). In this case, there is no restriction, and the number of solutions is simply s−1s-1 (pick n=1,2,3,...,s−1n=1,2,3,…,s-1). For example, let s=5s=5, which can be attained in four different ways: (1,4),(2,3),(3,2),(4,1)(1,4), (2,3),(3,2),(4,1).

We will call this our naive count N(s)N(s). For s>7s\gt 7, we must subtract the number of invalid solutions in which either m≥7m\ge 7 or n≥7n\ge 7, but not both (if m≥7m\ge 7 and n≥7n\ge 7, then m+n=s≥14m+n=s\ge 14, which is impossible!) from our naive count. If n≥7n \ge 7, then nn may be equal to 7,8,...,s−17, 8,…, s-1 because 7≤n≤s−17\le n\le s-1.

In other words, the upper bound constraint still applies in addition to the new lower bound. As a result, the number of possible solutions in this case is (s−1)−7+1=s−7(s-1)-7+1=s-7, inclusive. Similarly, if m≥7m \ge 7, the number of possible solutions is simply s−7s-7. Then, we subtract these invalid solutions from N(s)N(s) to get the expression for all s≥7 s \ge 7,

N(s)−2(s−7)=(s−1)−2s+14=13−sN(s)-2(s-7)=(s-1)-2s+14=13-s

And that is why the function DD indeed gives the correct count for the number of ways in which a particular sum ss is made. Why, you may also ask, go to the trouble of making this counting argument when we can make a table summing over all the possible pairs just as we did at the beginning of this article? Well, imagine what would happen if, instead of 6 faces, our dice had 1212, 2020, or more faces; then the brute-force counting would become particularly laborious and impractical (to say the least!).

Now, to determine the overall probability of winning this game, we must add up all the probabilities that contribute to it, i.e., all the ways a given player may win. As stated in the problem, if the player gets the sum 77 or 1111 on his first roll, he wins. Then,

ℙ(A Natural)=D(7)+D(11)36=6+236=836=29\mathbb{P}\left(\text{A Natural} \right)=\frac{D(7)+D(11)}{36}=\frac{6+2}{36}=\frac{8}{36}=\frac{2}{9}

Another way to win is by setting a point first (any other sum has occurred on his first roll except 22, 33, or 1212). The probability of setting a point on his first roll is

D(s)36,wheres∈{4,5,6,8,9,10}\frac{D(s)}{36}, \text{where} \:s\in \{ 4,5,6,8,9,10 \}

If a player has set a point, they continue to roll until either a subsequent roll matches their point and wins or they roll a 7 and lose. Letp=P(roll s)p=\text{P}(\text{roll s})(his point), q=P(roll 7)q=\text{P}(\text{roll 7}) (lose), and r=1−p−qr=1-p-q (other sums). After a point has been set, each roll is either decisive, with probabilities of winning or losing, pp and qq respectively, or non-decisive, with probability rr (In which case he must throw the dice again).

For a player to win at all, the first decisive roll must be his point ss, which can happen on the very next roll, having a probability of pp, after a non-decisive roll, having a probability of r⋅pr\cdot p, after two non-decisive rolls, having a probability of r2⋅pr^{2}\cdot p, and so forth. Hence,

ℙ(Eventually Win)=p+rp+r2p+r3p+...=p∑k=0∞rk=p1−r\mathbb{P}\left( \text{Eventually Win} \right)=p+rp+r^{2}p+r^{3}p+…=p\sum_{k=0}^{\infty }r^{k}=\frac{p}{1-r}

Then, since r=1−p−qr=1-p-q, by substitution,

p1−(1−p−q)=pp+q=P(roll s)P(roll s)+P(roll 7)=D(s)36D(s)36+D(7)36=D(s)D(s)+6\frac{p}{1-(1-p-q)}=\frac{p}{p+q}=\frac{\text{P}(\text{roll s})}{\text{P}(\text{roll s})+\text{P}(\text{roll 7})}=\frac{\frac{D(s)}{36}}{\frac{D(s)}{36}+\frac{D(7)}{36}}=\frac{D(s)}{D(s)+6}

By the general product rule of probability (the conjunction of setting a point first and then winning once that point is matched on any subsequent roll), the total contribution from all possible points is 

∑s∈pointsD(s)36⋅D(s)D(s)+6=∑s∈pointsD(s)236(D(s)+6)\sum_{s\in \text{points}}^{}\frac{D(s)}{36}\cdot\frac{D(s)}{D(s)+6}=\sum_{s\in \text{points}}^{}\frac{D(s)^{2}}{36\left( D(s)+6 \right)}

Consequently, since the events corresponding to the different ways of winning are mutually exclusive, the addition rule of probability gives

ℙ(Overall Win)=29+∑s∈{4,5,6,8,9,10}D(s)236(D(s)+6)\mathbb{P}\left( \text{Overall Win} \right)=\frac{2}{9}+\sum_{s\in \{ 4,5,6,8,9,10 \}}^{}\frac{D(s)^{2}}{36\left( D(s)+6 \right)}

=29+136(323+6+424+6+525+6+525+6+424+6+323+6)=\frac{2}{9}+ \frac{1}{36}\left(\frac{3^{2}}{3+6} +\frac{4^{2}}{4+6}+\frac{5^{2}}{5+6}+\frac{5^{2}}{5+6}+\frac{4^{2}}{4+6}+\frac{3^{2}}{3+6} \right)
=29+136(99+1610+2511+2511+1610+99)=\frac{2}{9}+ \frac{1}{36}\left(\frac{9}{9} +\frac{16}{10}+\frac{25}{11}+\frac{25}{11}+\frac{16}{10}+\frac{9}{9} \right)
=244495≈49.29%=\frac{244}{495} \approx 49.29 \%

A Dodecahedral Game of Craps

So far, we have relied on two familiar cubic dice, each with six faces. Their geometry gives rise to 3636 equally likely ordered outcomes and the characteristic symmetric distribution of sums. But is the cube essential to the mathematics?

Suppose a modified version of the dice game craps is played with two regular (i.e., perfectly symmetrical) dodecahedra. Each die has its sides numbered from 11 to 1212 so that after each throw of the dice, the sum of the numbers on the top two surfaces of the dice would range from 22 to 2424. If a player gets the sum 1313 or 2323 on his first throw (a natural), he wins. If he gets 2 2, 33, or 2424 on his first throw (craps), he loses.

If he gets any other sum (his point), he must throw the dice again. On this or any subsequent throw, the player loses if he gets the sum 13 and wins if he gets his point, but must throw both dice again if any other sum occurs. The player continues until he either wins or loses. To the nearest whole percent, what is the probability at the start of any game that a shooter wins?4

two dodecahedral dice
Figure 3. AI-Generated Image

Solution

Reveal solution

There are 12×12=14412\times 12=144 equally likely outcomes. Suppose ss is the sum of the two numbers on the top two surfaces of the dice. Define the functionD:{2,3,...,24} D:\{ 2,3,…,24 \}→\to{1,2,...,12}\{ 1,2,…,12 \} by the rule: For any given sum ss,5

D(s)={s−1if 2≤s≤1325−sif 13≤s≤24D(s) = \begin{cases} s-1 & \text{if } \: 2\le s\le 13 \\ 25-s & \text{if } \: 13\le s\le 24 \\ \end{cases}
=The number of ways to obtains=\text{The number of ways to obtain}\:s

Now, to determine the overall probability of winning this game, we must add up all the probabilities that contribute to it, i.e., all the ways a given player may win. As stated in the problem, if the player gets the sum 1313 or 2323 on his first roll, he wins. Then,

ℙ(A Natural)=D(13)+D(23)144=12+2144=14144=772\mathbb{P}\left( \text{A Natural} \right)=\frac{D(13)+D(23)}{144}=\frac{12+2}{144}=\frac{14}{144}=\frac{7}{72}

Another way to win is by setting a point first (any other sum has occurred on his first roll except 22, 33, or 2424. The probability of setting a point on his first roll is therefore

D(s)144,wheres∈{4,5,...,12,14,...,22}\frac{D(s)}{144}, \text{where} \:s\in \{ 4,5,…,12,14,…,22 \}

If a point has been set, the player continues to roll until either a subsequent roll matches his point and wins or gets the sum 1313and loses. Let p=P(roll s) p=\text{P}(\text{roll s})(his point) and q=P(roll 13)q=\text{P}(\text{roll 13}) (lose).By the result of the introductory problem, the probability in this case is

pp+q=P(roll s)P(roll s)+P(roll 13)=D(s)144D(s)144+D(13)144=D(s)D(s)+12\frac{p}{p+q}=\frac{\text{P}(\text{roll s})}{\text{P}(\text{roll s})+\text{P}(\text{roll 13})}=\frac{\frac{D(s)}{144}}{\frac{D(s)}{144}+\frac{D(13)}{144}}=\frac{D(s)}{D(s)+12}

Then, by the general product rule of probability (the conjunction of setting a point first and then winning once that point is matched on any subsequent roll), the total contribution from all possible points is 

∑s∈pointsD(s)144⋅D(s)D(s)+12=∑s∈pointsD(s)2144(D(s)+12)\sum_{s\in \text{points}}^{}\frac{D(s)}{144}\cdot\frac{D(s)}{D(s)+12}=\sum_{s\in \text{points}}^{}\frac{D(s)^{2}}{144\left( D(s)+12 \right)}

Consequently, by the sum rule of probability for mutually exclusive events, the overall probability of winning is the sum of all the aforementioned contributions, that is6,

ℙ(Win)=772+∑s∈{4,5,...,12,14,...,22}D(s)2144(D(s)+12)\mathbb{P}\left( \text{Win} \right)=\frac{7}{72}+\sum_{s\in \{ 4,5,…,12,14,…,22 \}}^{}\frac{D(s)^{2}}{144\left( D(s)+12 \right)}
=772+2144(323+12+424+12+525+12+626+12+727+12+828+12+929+12+10210+12+11211+12)=\frac{7}{72}+ \frac{2}{144}\left(\frac{3^{2}}{3+12} +\frac{4^{2}}{4+12}+\frac{5^{2}}{5+12}+\frac{6^{2}}{6+12}+\frac{7^{2}}{7+12}+ \frac{8^{2}}{8+12} +\frac{9^{2}}{9+12}+\frac{10^{2}}{10+12} +\frac{11^{2}}{11+12} \right)
=772+2144(915+1616+2517+3618+4919+6420+8121+10022+12123)=\frac{7}{72}+ \frac{2}{144}\left(\frac{9}{15} +\frac{16}{16}+\frac{25}{17}+\frac{36}{18}+\frac{49}{19}+ \frac{64}{20} +\frac{81}{21}+\frac{100}{22} +\frac{121}{23} \right)
=3004412968643960≈44%Rounded to the nearest whole percent=\frac{30044129}{68643960}\approx 44\% \:\: \text{Rounded to the nearest whole percent}

I don’t know about you, but I definitely wouldn’t play this version!

The House Edge: A Question of Expectations

Knowing the probability of winning a wager is important, but it does not tell us the entire story. After all, winning a bet that pays one dollar is quite different from winning one that pays ten dollars. To evaluate a wager properly, we must consider both the probability of each possible outcome and the corresponding gain or loss.

This brings us to the concept of expected value.

In probability theory, the expected value of a random variable is the weighted average of all its possible values, where each value is weighted according to its probability of occurrence. In other words, it tells us what the average result would be if we could repeat the same experiment a very large number of times. For a Pass Line bet, the player receives an even-money payout: he gains one unit if he wins and loses one unit if he loses. From our previous calculation,

ℙ(Win)=244495,ℙ(Lose)=251495.\mathbb P(\text{Win})=\frac{244}{495}, \qquad \mathbb P(\text{Lose})=\frac{251}{495}.

Let (X)(X) represent the player’s net gain from a single wager. Its expected value is therefore

𝔼[X]=(+1)244495+(−1)251495=244−251495=−7495≈−0.01414.\begin{aligned} \mathbb E[X] &=(+1)\frac{244}{495} +(-1)\frac{251}{495} &=\frac{244-251}{495} &=-\frac{7}{495} &\approx -0.01414. \end{aligned}

The negative sign is particularly important: it tells us that the wager is unfavorable to the player. On average, the player loses approximately 1.4141.414 units for every 100100 units wagered.

For example, if someone repeatedly places a one-dollar Pass Line bet, the expected loss per resolved wager is approximately 1.414 1.414 cents. Of course, this does not mean that the player will actually lose that amount on every occasion. Individual outcomes may vary considerably; the expected value describes the long-run average, not the result of any particular game.

This small but persistent disadvantage is known as the house edge. In this case, the casino has an advantage of approximately 1.414%1.414\% over the player’s initial wager7.

Thus, although the probability of winning a Pass Line bet is remarkably close to one-half, the odds are not quite in the player’s favor. The difference is small enough to be easily overlooked in a single game, but probability has a way of making such differences significant when the experiment is repeated often enough. What, then, would be the house edge for the Dodecahedral Game of Craps?8

References & Further Reading

Technical notes & references
  1. MGM Resorts International. (n.d.). Guide to craps. https://www.mgmresorts.com/en/gamesense/guide-to-craps.html ↩︎
  2. Obringer, L. (2004, July 8). How craps works. HowStuffWorks. https://entertainment.howstuffworks.com/craps.htm ↩︎
  3. This analysis follows closely that of Hoshino, R. (n.d.). Probability of winning at craps. Dalhousie University. https://www.mscs.dal.ca/~hoshino/book/ch20craps.pdf. A brilliant exposition of basic probability applied to this game ↩︎
  4. Adapted from Hoeflin, R. K. (1997, June). Problem 17 in The Hoeflin Power Test. Retrieved November 11, from http://miyaguchi.4sigma.org/hoeflin/power/power.html ↩︎
  5. To derive this counting function, the argument is completely analogous to the original problem of Craps at the beginning ↩︎
  6. Note that we multiply by 2 the whole sum of probabilities of matching one’s point after setting a point because the distribution of sums around the sum 13 is completely symmetrical ↩︎
  7. Shackleford, M. (2026, August 3). How the house edge for each bet is derived. Wizard of Odds. https://wizardofodds.com/games/craps/appendix/1/ ↩︎
  8. For a fuller mathematical treatment of craps, we recommend Mark Bollman’s Mathematics of the Big Four Casino Table Games: Blackjack, Baccarat, Craps, & Roulette (2021). ↩︎

Leave a Comment

Your email address will not be published. Required fields are marked *