1
$\begingroup$

How many times must a fair die be rolled in order for there to be less than 1% chance that the mean of the scores differ from 3.5 by more than 0.1?

The answer is $n≥1936$, but how do you get to this answer?

$\endgroup$
  • $\begingroup$ Did you try to determine the formula for the confidence interval? $\endgroup$ – user3371583 Apr 4 '15 at 13:17
  • $\begingroup$ How do I do that? $\endgroup$ – Andrew Apr 4 '15 at 13:27
  • 1
    $\begingroup$ If $X$ is the number of faces showing on one roll, then $E(X) = 3.5$ and you need to find $V(X)$, and from it $V(\bar X)$. Also, assume $\bar X$ is normal. $\endgroup$ – BruceET Apr 5 '15 at 9:37
1
$\begingroup$

Find mean and variance for one fair die. Let $X$ be the number of faces showing on one fair die. $$\mu = E(X) = (1 + 2 + \cdots + 6)/6 = [6(7)/2]/6 = 7/2 = 3.5,$$ where we have used the formula that the sum of the first $n$ integers is $n(n+1)/2.$ Also, $$E(X^2) = (1^2 + 2^2 + \cdots + 6^2)/6 = \frac{6(7)(13)/6}{6} = 91/6,$$ where we have used the formula that the sum of the first $n$ squares is $n(n+1)(2n+1)/6.$ Then $$ \sigma^2 = V(X) = E(X^2) - \mu^2 = \frac{91}{6} - \frac{49}{4} \frac{182 - 147}{12} = 35/12.$$ There are, of course, other methods of finding $\mu$ and $\sigma^2.$

Find the mean and variance for the average of $n$ dice. For the average $\bar X_n$ on $n$ rolls of a die, we have $$\mu_n = E[(X_1 + X_2 + \dots + X_n)/n] = n\mu/n = \mu.$$ and, because we assume the $n$ rolls of the die are independent, $$\sigma^2_n = V[(X_1 + X_2 + \dots + X_n)/n] = n\sigma^2/n^2 = \sigma^2/n.$$

Assume the average is normally distributed. If $n$ is even moderately large the Central Limit Theorem indicates that $\bar X_n$ has approximately the normal distribution with mean $\mu_n = 7/2$ and variance $\sigma_n^2 = \sigma^2/n = \frac{35}{12n}.$

Thus, upon standardizing (here, dividing by $\sigma_n$), we have $$P\{-0.1 < \bar X_n - 7/2 < 0.1\} \approx P\{-0.1\sqrt{12n/35} < Z < 0.1\sqrt{12n/35} \},$$ where $Z$ is standard normal. Because we want this probability to equal $.01 = 1\%,$ we note that $P\{-2.576 < Z < 2.576\} = 0.1.$

So in order to find the desired $n,$ we set $2.576 = 0.1\sqrt{12n/35},$ solve for $n$ and round up to the nearest integer. This gives the claimed answer $n = 1936$.

$\endgroup$
  • $\begingroup$ Some of this may already be in your textbook, and you may have to fill in some gaps. If anything is unclear, leave a comment. $\endgroup$ – BruceET Apr 5 '15 at 18:44
  • $\begingroup$ The law of large numbers indicates nothing about the distribution of $\bar X_n$ except that $\bar X_n\to\mu$ almost surely. $\endgroup$ – Did Apr 5 '15 at 19:04
  • $\begingroup$ @Did. Thanks. Wrote it before coffee. Changing it to CLT. $\endgroup$ – BruceET Apr 5 '15 at 20:20
  • $\begingroup$ @Did Doesn't it say that it approaches a degenerate distribution? $\endgroup$ – JohnK Apr 5 '15 at 20:24
  • $\begingroup$ @JohnK: If 'it' refers to the LLN of my pre-coffee blunder, then Yes. If 'it' refers to the CLT of the edited version, then No. Very roughly speaking, the LLN has $n$ in the denominator and the CLT has $\sqrt{n}$ in the denominator. Two fundamentally different limits result. $\endgroup$ – BruceET Apr 5 '15 at 22:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.