0
$\begingroup$

On this Wikipedia page on 'conjugate priors' the following notation is used for the posterior predictive of a normal random variable with known mean $\mu$ but unknown variance $\sigma^2$:

$$t_{2\alpha'}(\tilde{x}|\mu,\sigma^2 = \beta'/\alpha')$$

Here's my interpretation of the notation:

$\tilde{x}$ is a new i.i.d. variable the distribution of which is described by the t-distribution.

$\mu$ is the known mean and is the first parameter of the distribution

$\sigma^2$ is second parameter of the distribution and is the Bayesian variance estimate given by the ratio of the posterior $\beta$ and $\alpha$

$\alpha'$ and $\beta'$ are the posterior hyper-parameters of $\sigma^2$

$t_{2\alpha'}$ is a t-distribution with $\nu = n - 1$ degrees of freedom that has been shifted by $\mu$ and multiplied by $\sigma$.

My guess is that the pdf is given by:

$$f(x) = \left(\frac{\Gamma(\frac{\nu+1}{2})} {\sqrt{\nu\pi}\,\Gamma(\frac{\nu}{2})} \left(1+\frac{x^2}{\nu} \right)^{\!-\frac{\nu+1}{2}} + \mu\right)\sigma$$

Can anyone more knowledgeable clear this up for me?

$\endgroup$
4
$\begingroup$

$\tilde{x}$ is a new i.i.d. variable the distribution of which is described by the t-distribution.

Formally, $\tilde{x}$ is the argument of the posterior predictive pdf and hence a dummy. The posterior predictive pdf is the density of the distribution of a new and independent variate with same distribution as the observation.

μ is the known mean and is the first parameter of the distribution

μ is the known mean of both the observation and the predictive variate. It is the location parameter of the $t$ distribution

$t_{2α′}$ is a t-distribution with ν=n−1 degrees of freedom that has been shifted by μ and multiplied by σ.

$t_{2α′}$ is a $t$ distribution with $2α′$ degrees of freedom, not $n-1$, location parameter $\mu$ and scale parameter $\sigma$.

$σ^2$ is second parameter of the distribution and is the Bayesian variance estimate given by the ratio of the posterior β and α

$σ$ is the scale parameter of the $t$ distribution, provided by the ratio of the hyperparameters of the posterior distribution.

the pdf is given by $$f(x) = \left(\frac{\Gamma(\frac{\nu+1}{2})} {\sqrt{\nu\pi}\,\Gamma(\frac{\nu}{2})} \left(1+\frac{x^2}{\nu} \right)^{\!-\frac{\nu+1}{2}} + \mu\right)\sigma$$

No, the density of a location-scale $t_\nu(\cdot;\mu,\sigma)$ distribution is $$\frac{\Gamma(\frac{\nu+1}{2})}{\sigma\sqrt{\nu\pi}\,\Gamma(\frac{\nu}{2})}\left(1+\frac{(x-\mu)^2}{\nu\sigma}\right)^{\!-\frac{\nu+1}{2}} $$

$\endgroup$

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.