I saw this exercise and its solution, but I'm not sure about it. (I'm gonna quote the solution and after I'll say why I have doubts)

According http://mathfaculty.fullerton.edu/sbehseta/Math538_Elena_Quiz%203.pdf, I have the following problem.

Problem 10.1: Number of simulation draws. Suppose that scalar variable $\theta$ is approximately normally distributed in a posterior distribution that is summarized by n independent simulation draws. How large does n have to be so that the 2.5% and 97.5% quantiles of $\theta$ are specified to an accuracy of 0.1sd($\theta|y$). Figure this out mathematically.

And the following solution (also given by the same author of the paper)

(Without mention why) The text says that the posterior probabilities are estimated to a standard deviation of $\sqrt{p(1-p)/S}$, where S is the number of draws. Suppose we have priors $\mu=\tau=\sigma=1$ and ${y}_{1},...,{y}_{n}|\theta\sim N(\theta,\sigma=1)$ and $\theta\sim N(\mu=1,\tau=1)$. Then the posterior is $\theta|y\sim N(\frac{1+n\overline{y}}{1+n},\frac{1}{2})$.

Suppose we want to know the amount of samples needed to find the probability of $\theta\ge 2$. We know, based on our assumptions above that $\mathbb{P}(\theta\ge 2|$y)=0.91. Then, $S=\frac{p(1-p)}{0.01^{2}}=\frac{.91(1-.91)}{0.01^{2}}\approx 853.3$ and we need 854 draws.

Now, these are my doubts:

  1. Why can we estimate the sd with $\sqrt{p(1-p)/S}$?

  2. Why $\mathbb{P}(\theta\ge 2|$y)$=0.91$? To me, this value don't match, don't care the parameters of the distribution (At least at this example). Also, I think p should be 0.975 because we are looking for 97.5% quantile.

  3. Clearly the calculus of the number of draws is incorrect, but whatever, why are they using 0.01 (instead of 0.1)? According to the proposed posterior, $sd(\theta|y)=\sqrt{1/2}$

Summarizing, I thought the solution of the problem assuming that que estimation of s.d. proposed by the paper is correct (Even when I'm not sure). And i'd only make changes that I wrote at 2. and 3.

  • 1
    Since this is clearly an exercise, please add the self-study tag and try to expand on what you attempted and where exactly you get stick. Reference for the exercise would also be needed. – Xi'an Mar 18 '17 at 12:36
  • 1
    Thanks @Xi'an , I edited my question. I attempted almost the same thing but changing (mainly) the probability given. (I can't figure out why they are using 0.91). I understand the priors election, but the development of the given solution is not clear for me – student Mar 18 '17 at 16:51

Your Answer

 
discard

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.