If I understand correctly, a distribution in the exponential family... $$\underline X\sim f_{\underline\theta}(\underline x) = exp\{\sum\limits_{i}\eta_i(\underline\theta)T_i(\underline x)-B(\underline\theta)\}~h(\underline x)$$ ...where $\underline\eta(\centerdot)$ is a (possibly vector valued) parameter, $\underline T(\centerdot)$ is a corresponding vector of sufficient statistics, $B(\theta)$ is the log partition, and $h(\underline x)$ is the base measure and $\underline T(\underline x)$ itself has the following distribution:

$$\underline T(\underline x) =\underline t\sim g_{\underline\theta}(\underline t) = exp\{\sum\limits_{i}\eta_i(\underline\theta)t_i-B(\underline\theta)\}~h^*(\underline t)$$

...where $h^*(\centerdot)$ is not necessarily the same function as the one from $f_{\underline\theta}(\underline x)$ above.

My questions are:

  • Are the other two functions ($B(\underline\theta)$ and $\underline\eta(\underline\theta)$) identical to the ones from $f_{\underline\theta}(\underline x)$?
  • What is the general strategy for finding $h^*(\centerdot)$?

1 Answer 1


The functions are identical (same carrier measure, same log-normalizer).


Your Answer

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

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