# Concluding that a coin is biased depending on number of flipping trials

I am given the task to check if a coin is biased to land on heads. The bias must exceed a certain threshold i.e. $$p > 0.5 + \epsilon$$ for some given $$\epsilon$$. I would like to know how the number of flips affects the certainty of the conclusion that the coin is biased according to this definition.

Concretely, if I am allowed to flip the coin $$n$$ times, what is the probability of a false positive (coin is not biased but I claim that it is) and a false negative (coin is biased but I fail to spot it) as a function of $$n$$ and $$\epsilon$$?

• Please take a look at the Binomial Test. – denklo Jan 28 at 11:26
• The question is wrong. It is no more answerable than "How long is a piece of string?". To make a sensible question, more information is needed. If we know the prior probability that the coin is biased, then we can calculate a posterior probability of it being so from the result of any given number of flips. – John Bentin Jan 28 at 12:18
• @JohnBentin, I took a look at the binomial test. Isn't it correct to have a null hypothesis that the coin is biased such that $p>0.5 +\epsilon$ and the alternative hypothesis that it is biased but with $p\leq 0.5 +\epsilon$? I wish to accept/reject the null hypothesis. Is this a well defined question? – user1936752 Jan 28 at 13:43
• The question didn't make clear that the set-up was to test the null hypothesis $p>0.5+\epsilon$ versus the opposite. For any $p\in[0\,\Bbb,\, 1]$, you can calculate the probability of the outcome being as extreme as, or more extreme than, what is observed given that the probability of a head is $p$. But this looks like a pretty meaningless exercise. Even this useless calculation becomes impossible if we weaken the assumption from fixing P(head) to merely P(head) > $0.5+\epsilon$ without specifying a distribution for P(head) in $[0.5+\epsilon\,\Bbb, \,1]$. – John Bentin Jan 28 at 19:23