Let $Y \sim \mathrm{Bin}(n, q)$ and $X \sim \mathrm{Bin}(Y, p)$, is there an expression for the distribution of $X/Y$?
$\begingroup$
$\endgroup$
4
-
3$\begingroup$ Binomial distributions take the value $0$ with positive probability, so you may be looking at $\frac00$ when $Y=0$, which has probability $(1-q)^n$. You may want to define this case to be $0$ or $1$ or $p$ or something else but you do need to handle it $\endgroup$– HenryCommented Apr 7, 2022 at 8:26
-
$\begingroup$ stats.stackexchange.com/questions/153951 ran into the same difficulty and therefore no answer could be supplied. stats.stackexchange.com/questions/458789 solves a related statistical problem indirectly -- perhaps that would be relevant to your application? $\endgroup$– whuber ♦Commented Apr 7, 2022 at 13:53
-
$\begingroup$ @Henry good point $\endgroup$– geometrikalCommented Apr 7, 2022 at 20:31
-
$\begingroup$ @whuber thanks for that. The application is related to machine learning. Say I have an image classifier that I've trained. I randomly select an images from a test set of which I know the classes (Y is number of images for a class that occurs with frequency q from n selections). I then classify them with the model and record true / false (X where p is the probability the model classifies correctly) and calculate the accuracy (X/Y). I hoping to get an idea for how many (n) images I need to select to have the accuracy occur within a certain confidence interval. $\endgroup$– geometrikalCommented Apr 7, 2022 at 20:38
Add a comment
|