0
$\begingroup$

Suppose I have a large number of urns, each with a different ratio $r$ of white and black balls. Some urns may be full of white balls or full of black balls. What kinds of distributions or processes could I use to model the percentage of white balls across the urns? So I'm looking for a distribution with support $[0,1]$ that does not necessarily go to zero at the ends.

I initially thought that a beta distribution would make sense, but that assumes that the balls are coming from some independent process, and you typically get $p(0) = p(1) = 0$. In many applications, similar people tend to stick together, so you get higher probabilities for the ends.

Possible examples include:

  1. Percentage of a country's population that follows Christianity (there are many countries that are 100% or 0% Christian).

  2. Percentage of products made by companies that are of a particular type (there are many companies that sell exclusively clothes and many that sell exclusively non-clothes, and some that sell both).

  3. Percentage of a minority group in a neighborhood.

Improve this question
$\endgroup$

1 Answer 1

Reset to default
0
$\begingroup$

Maybe logit-normal is close enough. It still goes to 0 at the endpoints, but there can be "spikes" arbitrarily close to the endpoints. Arguably the existence of "100% Christian countries" is an artifact of the finite number of people in the country.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.