How would one interpret the strength of prior belief associated with a parameter with a prior beta(10,8) distribution, compared to a beta(0,0) prior, (with data from a binomial distribution)? Some intuition and calculations to support your answer would be really appreciated.

  • $\begingroup$ I took the liberty to edit your title to something more informative. Feel free to revert the change it edit it if you feel that it is inaccurate. $\endgroup$ – Tim Mar 12 '18 at 5:47

Consider the standard Bayesian conjugate model:

$$\theta \sim \text{Beta}(\alpha, \beta),$$ $$X | \theta \sim \text{Bin}(n, \theta).$$

The posterior distribution for this model is:

$$\theta | x \sim \text{Beta}(\alpha + x, \beta + n-x).$$

In this model, the value $n_0 \equiv \alpha + \beta$ in the prior is a measure of prior strength that is sometimes called the number of pseudo data points. Observation of the count value $x$ from $n$ trials adds to this to the posterior strength $n_0 + n$, so this is like you had $n_0$ pseudo data points and then added $n$ actual data points.


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.