3
$\begingroup$

While I am looking for the sun-rise problem, in part of prior selection they said that ...

in a section of Beta distribution https://en.wikipedia.org/wiki/Beta_distribution#Rule_of_succession

Jaynes[53] questions (for the uniform prior Beta(1,1)) the use of these formulas for the cases s = 0 or s = n because the integrals do not converge (Beta(1,1) is an improper prior for s = 0 or s = n).

in proof of rule of succession which is using Beta(1,1) as its prior https://en.wikipedia.org/wiki/Rule_of_succession

This improper prior is 1/(p(1 − p)) for 0 ≤ p ≤ 1 and 0 otherwise.[5]

Where the p is an unknown parameter for the probability of success.
Though I don't know where 1/(p(1 − p)) comes from.


Beta(1,1) is Uniform(0,1) which is proper prior. Apparently, it is integrated into one, thus a proper prior.

Question: why prior Beta(1,1) in proof for the rule of succession is improper?

$\endgroup$
1
  • 6
    $\begingroup$ +1. It looks like Wikipedia is confused. The "integrals" to which it refers must be the three in the preceding paragraph, but only the one with the Beta$(0,0)$ prior fails to converge for $s=0$ or $s=n.$ It's possible Jaynes was using a different parameterization, one in which the equivalent of Beta$(1,1)$ is improper, so it would be nice to find the relevant passage in Jaynes (which Wikipedia does not cite specifically). $\endgroup$
    – whuber
    Mar 8, 2020 at 13:32

1 Answer 1

5
$\begingroup$

The Wikipedia passage

Jaynes [53] questions (for the uniform prior Beta(1,1)) the use of these formulas for the cases $s = 0$ or $s = n$ because the integrals do not converge (Beta(1,1) is an improper prior for $s = 0$ or $s = n$). In practice, the conditions $0<s<n$ necessary for a mode to exist between both ends for the Bayes prior are usually met, and therefore the Bayes prior (as long as $0 < s < n$) results in a posterior mode located between both ends of the domain.

is clearly a typo as the rest of the section

Haldane's prior probability (Beta(0,0))

The Beta(0,0) distribution was proposed by J.B.S. Haldane,[64] who suggested that the prior probability representing complete uncertainty should be proportional to $p^{−1}(1−p)^{−1}$ (...)

calls the Bayes-Laplace prior a Beta(1,1) distribution, Haldane's prior a Beta(0,0) [improper] distribution, and Jeffrey's middle ground a Beta(1/2,1/2) distribution.

Ànother sentence that seems to be wrong is

According to Jaynes [53] the main problem with the rule of succession is that it is not valid when $s=0$ or $s=n$ (see rule of succession, for an analysis of its validity).

since Jeffreys' prior, which is supported by Jaynes, has no issue when $s=0$ or $s=n$.

$\endgroup$
0

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.