1
$\begingroup$

Say I have a coin, how many flips would I need to do and record to determine that results of said flips are accurate, that there is indeed a 50% chance to land on heads or tails? 100, 1000, 10000?

How do I extend this to a dice roll with n number of sides. how many rolls would I need to determine that there is indeed a 10% chance to land on the number 7 on a 10 sided die? 100, 1000, 10000?

Is there a formula for determining how many trials I must conduct to decided the probability of an outcome with a 95% or 99% confidence level?

$\endgroup$
0

1 Answer 1

3
$\begingroup$

No finite number of trials is sufficient to determine that the probability of getting a head (say) is exactly $\frac12$.

Consider the difference between P(head) = $\frac12$ and P(head) = $\frac12+\varepsilon$. For every finite $n$, there's a value for $\varepsilon>0$ which is so close to $0$ that you will find the two probabilities very difficult to distinguish with $n$ trials.

So unless you specify a value for $\varepsilon$ that you regard $\frac12+\varepsilon$ as "close enough to 50%", there's no value of $n$ that will do.

A 95% interval for P(head) will have a "margin of error" (have interval half-width) of $\frac{1}{\sqrt{n}}$.

The ideas work similarly for dice, or any other similar situation (but the numbers differ).

Is there a formula for determining how many trials I must conduct to decided the probability of an outcome with a 95% or 99% confidence level?

You need to give more information to pin down a sample size.
If you specify both an interval-width (or specify a margin of error, the half-width) and coverage probability for a confidence interval (such as "I want a 95% CI for P(head) to be no wider than 0.01" for example), then you can get a sample size from that.

$\endgroup$
2
  • $\begingroup$ This question is answered in the wikipedia article, "Checking whether a coin is fair". See the section, "Estimator of true probability" for specifics of the general method described by Glen_b. As a supplement to the above, see: StackExchange answer on calculating maximum error $\endgroup$
    – ZachL
    Aug 22, 2015 at 20:53
  • $\begingroup$ "Everyone knows" that with a standard die, each of six numbers is as likely, don't we? When my school class of 20-some tried throwing hundreds of dice each, almost every child reported 2 and 5 coming up greatly more than any other number. Had we all been using the same dice, that would most obviously indicate a physical bias. (Sidewise, since we clearly did not use the same dice, does anyone think those results should not have led our teacher to undertake more research, perhaps leading to a relevant doctorate?? $\endgroup$ Feb 26, 2021 at 22:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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