Six-sided and four-sided dice question contradiction

diceprobability

Suppose we have two fair dice, with six sides (numbered 1 to 6) and with four sides (numbered 1 to 4). Suppose we pick a die at random and throw it, the result is announced to be 2 and the die is discarded. Then we pick the remaining die and throw it. What is the expectation of the second throw?

On the one hand, us getting 2 on the first throw, does not provide us with information on whether the first die was six-sided or four-sided, so we could argue that by total expectation the expectation of the second throw is just the average of expectations:
$$
E(X)=(3.5)\frac{1}{2} + (2.5)\frac{1}{2}=3
$$

On the other hand, we know that we got 2 on the first throw, so the expectation of the second throw is:
$$
E(X)=(1)\frac{2}{9} + (2)\frac{1}{9} + (3)\frac{2}{9} + (4)\frac{2}{9} + (5)\frac{1}{9} + (6)\frac{1}{9}=3.22
$$

It seems to me that there is a contradiction?

Best Answer

On the one hand, us getting 2 on the first throw, does not provide us with information on whether the first dice was six-sided or four-sided

This sentence is false.

Imagine that instead of a 6-sided die and a 4-sided die, you had a 1000000-sided die, numbered 1-1000000, and a 4-sided die, numbered 1-4. You select one of the two dice uniformly at random, and roll it, and the result is a 2. Do you really claim that you have no information about which die was selected? Intuitively, it is very probable that the 4-sided die was selected; if the 1000000-sided die had been selected, the result could have been anything between 1 and 1000000, and it would look like quite an unprobable coincidence that the result was a 2.

A great tool to formalise this intuition and turn it into precise numbers, and make it work with a 6-sided die instead of a 1000000-sided die, is Bayes' theorem.

Let $D_6$ be the event "we picked the 6-sided die first" and $D_4$ the event "we picked the 4-sided die first". Let $R_2$ be the event "we rolled a 2 on the first roll". Then, in the absence of information, the probability of picking the 6-sided die first is $P(D_6) = \frac{1}{2}$; but the probability of having picked the 6-sided die, knowing that we rolled a 2, is noted $P(D_6 \,|\, R_2)$ and can be calculated using Bayes' theorem:

\begin{align*} P(D_6 \,|\, R_2) & = \frac{P(R_2 \,|\, D_6)P(D_6)}{P(R_2)} \\ P(D_6 \,|\, R_2) & = \frac{P(R_2 \,|\, D_6)P(D_6)}{P(R_2 \,|\, D_6)P(D_6) + P(R_2 \,|\, D_4)P(D_4)} \\ \end{align*}

Filling in these values:

  • $P(D_6) = P(D_4) = \frac{1}{2}$;
  • $P(R_2 \,|\, D_6) = \frac{1}{6}$;
  • $P(R_2 \,|\, D_4) = \frac{1}{4}$.

We get: \begin{align*} P(D_6 \,|\, R_2) & = \frac{\frac{1}{6}\times\frac{1}{2}}{\frac{1}{6}\times\frac{1}{2} + \frac{1}{4}\times\frac{1}{2}} \\ P(D_6 \,|\, R_2) & = \frac{2}{5} \\ \end{align*} Thus, after observing that we rolled a 2, the probability that we picked the 6-sided die is $\frac{2}{5}$, and the probability that we picked the 4-sided die is $\frac{3}{5}$.

(If we had used a 1000000-sided die and a 4-sided die, using the same reasoning we would have found that after observing a 2, the probability of having picked the 1000000-sided die is only $\frac{1}{250001}$, or $0.0004\%$.)

Now you can calculate the expectation of the second roll after having observed a 2 on the first roll. Calling $X$ the random variable equal to the result of the second roll:

\begin{align*} \mathbb{E}(X \,|\, R_2) & = \mathbb{E}(X \,|\, D_6)P(D_6 \,|\, R_2) + \mathbb{E}(X \,|\, D_4)P(D_4 \,|\, R_2) \\ \mathbb{E}(X \,|\, R_2) & = 2.5 \times \frac{2}{5} + 3.5 \times \frac{3}{5} \\ \mathbb{E}(X \,|\, R_2) & = 3.1 \\ \end{align*}

Thus the expectation of the second roll, having observed a 2 on the first roll, is 3.1. This is slightly higher than the expectation of the second roll we would have calculated if we hadn't observed the first roll: $$\mathbb{E}(X) = \mathbb{E}(X \,|\, D_6)P(D_6) + \mathbb{E}(X \,|\, D_4)P(D_4) = 2.5 \times \frac{1}{2} + 3.5 \times \frac{1}{2} = 3$$