Conditional posterior for \(\Sigma\)
\[\begin{align}Y_i \mid \boldsymbol{\theta}, \Sigma & \overset{ind}{\sim} N(\boldsymbol{\theta}, \Sigma)\\
\Sigma & \sim \textrm{IW}_p(\eta_0, \boldsymbol{S}_0^{-1}) \\
\boldsymbol{\theta} & \sim N(\mu_0, \Psi_0^{-1})
\end{align}\]
The conditional posterior (full conditional) \(\Sigma \mid \boldsymbol{\theta}, \boldsymbol{Y}\), is then \[\Sigma \mid \boldsymbol{\theta}, \boldsymbol{Y} \sim \textrm{IW}_p\left(\eta_0 + n, \left(\boldsymbol{S}_0+ \sum_{i=1}^n (\boldsymbol{Y}_i - \boldsymbol{\theta})(\boldsymbol{Y}_i - \boldsymbol{\theta})^T\right)^{-1} \right)\]
posterior sample size \(\eta_0 + n\)
posterior sum of squares \(\boldsymbol{S}_0+ \sum_{i=1}^n (\boldsymbol{Y}_i - \boldsymbol{\theta})(\boldsymbol{Y}_i - \boldsymbol{\theta})^T\)
Posterior Derivation
- The conditional posterior (full conditional) \(\Sigma \mid \boldsymbol{\theta}, \boldsymbol{Y}\), is \[\begin{align*}
\pi(\Sigma & \mid \boldsymbol{\theta}, \boldsymbol{Y})\propto p(\Sigma) \cdot p( \boldsymbol{Y} \mid \boldsymbol{\theta}, \Sigma)\\
& \propto \left|\Sigma\right|^{\frac{-(\eta_0 + p + 1)}{2}} \textrm{exp} \left\{-\frac{1}{2} \text{tr}(\boldsymbol{S}_0\Sigma^{-1}) \right\} \cdot \prod_{i = 1}^{n}\left|\Sigma\right|^{-\frac{1}{2}} \ \textrm{exp} \left\{-\frac{1}{2}\left[(\boldsymbol{Y}_i - \boldsymbol{\theta})^T \Sigma^{-1} (\boldsymbol{Y}_i - \boldsymbol{\theta})\right] \right\} \\
& \\
& \\
& \\
&
\end{align*}
\]
\[\Sigma \mid \boldsymbol{\theta}, \boldsymbol{Y} \sim \textrm{IW}_p\left(\eta_0 + n, \left(\boldsymbol{S}_0+ \sum_{i=1}^n (\boldsymbol{Y}_i - \boldsymbol{\theta})(\boldsymbol{Y}_i - \boldsymbol{\theta})^T\right)^{-1} \right)\]