You take $n = 50$ dice rolls and calculate the average value of these dice rolls. What is the probability (approximately -i.e using CLT) that this average is greater than or equal to $4$?

My attempt:

We know that each die has expectation and $3.5$ and variance $35/12$. So the sum has variance $3.5n = 105$ and variance $\frac{35}{12}n = 145.8\overline{3}$. So, I was thinking to evaluate the cumulative probability of a normal distribution with parameters $\mu = 105, \sigma^2 = 145.8\overline{3}$, however I'm not sure if this is correct.

  • $\begingroup$ You need to compute the expectation and variance of the average rather than the sum of 50 outcomes. $\endgroup$ – Math Lover Nov 13 '17 at 23:20
  • $\begingroup$ Ok, that's what I thought. what would I be computing the expectation and variance with respect to? I.e it is clear to me during the initial calculation for a single die that i take the weighted sum of each outcome 1 through 6, however it does not seem obvious to me what the calculation would look like in this case. $\endgroup$ – rubikscube09 Nov 13 '17 at 23:22
  • $\begingroup$ Let $Y = \sum X_i$. Then $m = Y/50$. You have found the expectation and variance of $Y$. Can you compute those of $m$? $\endgroup$ – Math Lover Nov 13 '17 at 23:27
  • $\begingroup$ So clearly $E[m] = 3.5$ and for the variance, we would have $Var [m] = \frac{1}{n^2} Var[Y]$ which would be $(35/12)/50$. $\endgroup$ – rubikscube09 Nov 13 '17 at 23:35
  • $\begingroup$ Yes, use them to solve your problem. $\endgroup$ – Math Lover Nov 13 '17 at 23:37

For completeness, your normal approximation might suggest a probability of about $$1-\Phi\left(\dfrac{4-3.5}{\sqrt{(35/12)/50}}\right) \approx 0.019217$$ where $\Phi$ is the cumulative distribution function of a standard normal distribution

This is not bad as the actual probability is about $0.021086$, but you can do slightly better with a continuity correction. A mean of at least $4$ corresponds to a sum of at least $200$ or a sum strictly more than $199$ on this discrete distribution, so checking $4$ or $\frac{199}{50}=3.98$ are both arguable on the continuous normal distribution, and a compromise would be to check $3.99$ so about the closer $1-\Phi\left(\tfrac{4-3.5}{\sqrt{(35/12)/50}}\right) \approx 0.021240$

As a warning, a normal approximation like this can be relatively poor in the tails of a distribution. If for example you had been looking for the probability of a mean of at least $5$, the actual probability would have been about $5.8 \times 10^{-11}$ but a straight normal approximation might have been about $2.6\times 10^{-10}$ and with continuity correction about $3.4\times 10^{-10}$, both several times too high


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.