Skip to content

Latest commit

 

History

History
48 lines (38 loc) · 1.51 KB

cba_assig_notes.org

File metadata and controls

48 lines (38 loc) · 1.51 KB

Question 1

Binomial likelihood, the posterior with a conjugate beta prior is: \begin{equation} p(θ | y)=\frac{1}{B(\overline{α}, \overline{β})} θ\overline{α-1}(1-θ)\overline{β-1} \end{equation} with \begin{equation} \begin{aligned} \overline{α} &=α0+y
\overline{β} &=β0+n-y \end{aligned} \end{equation}

The beta prior can be specified as: \begin{equation} ≡ \text { binomial experiment with }\left(α0-1\right) \text { successes in }\left(α00-2\right) \end{equation}

Question 2

Check PPD for binomial likelihood on p. 151. We should take into account sampling variability of $\hat{θ}$

Question 3

Contour probability: posterior evidence of $H0$ with HPD interval. Defined as: \begin{equation} P\left[p(θ | \boldsymbol{y})>p\left(θ0 | \boldsymbol{y}\right)\right] ≡\left(1-pB\right) \end{equation}

$pB$ is computed from the smallest HPD interval containing $θ0$.

$\operatorname{Beta}\left(α0, β0\right)$ prior is equivalent to a binomial experiment with $α0 - 1$ successes in ($α0 + β0 - 2$) experiments.

The non-informative beta prior has $α0=1, β0=1$ and is equal the uniform prior on $[0, 1]$.

Question 4

Popular priors for BGLIM are normal proper priors with large variance. Gelman et al. however suggest Cauchy density with center 0 and scale parameter 2.5 for standardized continuous covariates.