0
$\begingroup$

I have a set of binomial data $\{r_i\}$ (successes) for a set of $\{N_i\}$ trials for which I know the distribution given success parameter $p$: $P(\{r_i\}\ |\ \{N_i\},\ p)$.

Applying Bayes Theorem, the posterior for p is:

\begin{align} P(p\ |\{r_i\},\ \{N_i\})=\cfrac{P(\{r_i\}\ |\ \{N_i\},\ p)\ P_0(p)}{\int P(\{r_i\}\ |\ \{N_i\},\ p)\ P_0(p)\ dp} \end{align}

where $P_{0}(p)$ is my prior.

How would this change if my success probability $p$ is a function $f(A,\ E_0)$ of two parameters $A$ and $E_0$, for which I consider Gaussian priors $P_0(A)$ and $P_0(E_0)$?

How does one apply Bayes Theorem to get the $P(A,\ E_0\ |\ \{r_i\},\ \{N_i\})$ posterior in this case? Thanks in advance!

$\endgroup$
0
$\begingroup$

Your formula with Bayes Theorem is still correct, but the prior for $p$, $P_0(p)$, would be calculated as $$ P_0(p) = \sum_{\{(A, E_0) \ : \ f(A, E_0) = p\}} \ P_0(A) \ P_0(E_0), $$ that is, you would marginalise over $A$ and $E_0$ here. I'm assuming that $f$ is a deterministic function and that if $A, E_0$ are valid values, then $f(A, E_0)$ is a valid value of $p$.

$P(A, E_0 | \{r_i\}, \{N_i\})$ is given by $$ \frac{P(\{r_i\}, \{N_i\} \ | \ A, \ E_0) \ P_0(A) \ P_0(E_0)}{\int P(\{r_i\}, \{N_i\} \ | \ A, \ E_0) \ P_0(A) \ P_0(E_0) \ dA \ dE_0}, $$ and since both $p = f(A, \ E_0)$ and $\{r_i\}, \{N_i\}$ only depend on $A, E_0$ through $p$, then this can be written as $$ P(A, E_0 | \{r_i\}, \{N_i\}) = \frac{P(\{r_i\}, \{N_i\} \ | \ p = f(A, E_0)) \ P_0(A) \ P_0(E_0)}{\int P(\{r_i\}, \{N_i\} \ | \ p = f(A, E_0)) \ P_0(A) \ P_0(E_0) \ dA \ dE_0}. $$

$\endgroup$
  • $\begingroup$ Thanks! But does your first equation where you calculate the prior for p hold if A and E0 are not fully independent? $\endgroup$ – Radusaurus Apr 9 '18 at 16:03
  • $\begingroup$ You're welcome. The answer to your second question is that the first equation would not hold if $A$ and $E_0$ were not independent. You would need to replace $P_0(A) P_0(E_0)$ in the numerator and denominator with the joint distribution of $A, E_0$, labelled $P_0(A, E_0)$. $\endgroup$ – Alex Apr 12 '18 at 6:35

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.