A coin has an unknown head probability $p$. Flip $n$ times, and observe $X=k$ heads. Assuming an uniform prior for $p$, then the posterior distribution of $p$ is $B(\alpha = k + 1, \beta = n - k + 1)$. Consider $Y$ = number of additional flips required until the first head appears. Find the following distributions:

  • $P(Y=j|p=\theta)$, for j = 1,2,3,...
  • $P(Y=j|X=k)$, for j = 1,2,3,...

I think $P(Y=j|p = \theta) = \frac{P(Y=j,p=\theta)}{P(p=\theta)}$ and similarly for part 2. But are the RV's independent? And how do I find their joint pdf?


$\mathsf P(Y=j\mid p=\theta)$ is the conditional probability of $j$ aditional flips until another head shows for a given bias for the coin (after $X=k$ in $n$ flips).   What is the (conditional) distribution used for this?   (Hint: you do not need to resort to Bayes' for this; just identify the model.)

$\mathsf P(p=\theta)$ is the prior distribution of the bias; you are told to evaluate it using two cases: (1) $p$ is Uniform$(0;1)$, (2) $p$ is Beta$(\alpha:=k+1,\beta:=n-k+1)$

Random variables $p, Y$ are not independent, and their joint probability is the product of the prior of $p$ and the conditional of $Y$ given $p$.   $\mathsf P(Y=j\mid p=\theta)~\mathsf P(p=\theta)$

Then, for each type of prior, you are tasked to find the posterior: $\mathsf P(p=\theta\mid Y=j)$, and this is where you apply Bayes' rule

$$\mathsf P(p=\theta\mid Y=j)~=~\dfrac{\mathsf P(Y=j\mid p=\theta)~\mathsf P(p=\theta)}{\int\limits_0^1\mathsf P(Y=j\mid p=t)~\mathsf P(p=t)\operatorname d t}$$

  • $\begingroup$ Thanks! But I'm not sure what $P(Y=j|p=\theta)$ is. Also, this is the actual question. I'm sorry for the confusion. $\endgroup$ – user3727610 Apr 25 '16 at 1:17
  • $\begingroup$ So for the first one, it's just $P(Y=j|p=\theta) =(1-\theta)^{j-1}\theta$? $\endgroup$ – user3727610 Apr 25 '16 at 3:11

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.