Why use a beta distribution on the Bernoulli parameter for hierarchical logistic regression?

Question

I'm currently reading Kruschke's excellent "Doing Bayesian Data Analysis" book. However, the chapter on hierarchical logistic regression (chapter 20) is somewhat confusing.

Figure 20.2 discribes a hierarchical logistic regression where the Bernoulli parameter is defined as a the linear function on the coefficients transformed through a sigmoid function. This seems to be the way the hierarchical logistic regression is posed in most of the examples I've seen in other sources online as well. For instance - http://polisci2.ucsd.edu/cfariss/code/SIMlogit02.bug

However, when the predictors are nominal, he adds a layer in the hierarchy - the bernoulli parameter is now drawn from a beta distribution (figure 20.5) with parameters determined by mu and kappa, where mu is the sigmoid transformation of the linear function of coefficients, and kappa uses a gamma prior.

This seems reasonable and analogous to the coin-flipping example from chapter 9, but I don't see what having nominal predictors has to do with adding a beta distribution. Why wouldn't one do this in the case of metric preditors and why was the beta distribution added for the nominal predictors?

EDIT: Clarification on the models I'm referring to. First, a logistic regression model with metric predictors (no beta prior). This is similar to other examples of hierarchical logistic regression, such as the bugs example above:

$$ y_i \sim Bernoulli(\mu_i) \\ \mu_i = sig(\beta_0 + \sum_j \beta_j x_{ji} ) \\ \beta_0 \sim N(M_0, T_0) \\ \beta_j \sim N(M_\beta, T_\beta) \\ $$

Then the example with nominal predictors. Here's where I don't quite understand the role of the "lower" level of the hierarchy (incorporating the logistic outcome into a beta prior for a binomial) and why it should be different than the metric example.

$$ z_i \sim Bin(\theta_i, N) \\ \theta_i \sim Beta(a_j, b_j) \\ a_j = \mu_j \kappa \\ b_j = (1- \mu_j) \kappa \\ \kappa \sim \Gamma(S_\kappa, R_\kappa) \\ \mu_j = sig(\beta_0 + \sum_j \beta_j x_{ji} ) \\ \beta_0 \sim N(M_0, T_0) \\ \beta_j \sim N(0, \tau_\beta) \\ \tau_\beta = 1/\sigma_{\beta}^2 \\ \sigma_{\beta}^2 \sim folded t (T_t, DF) $$

Answer 1

The two models you compare have many extraneous features, and I think you can restate your question more clearly in the context of the following two simplified models:

Model 1:

\begin{align} y_i | \mu_i &\sim Bern( \mu_i ) \\ \mu_i &\sim \pi(\mu_i) \end{align}

Model 2:

\begin{align} y_i | \theta_i & \sim Bern( \theta_i ) \\ \theta_i | \mu_i,\kappa &\sim Beta\big( \mu_i\kappa, (1-\mu_i)\kappa \big) \\ \mu_i&\sim \pi(\mu_i) \end{align}

Your questions are: (1) what role is played by the beta distribution; and related, (2) how (if at all) is Model 2 different from Model 1?

On the surface these appear to be pretty different models, but in fact, the marginal distributions of $\mu_i$ in both models are identical. The posterior distribution of $\mu_i$ in Model 1 is \begin{gather} p(\mu_i|y_i) \propto \mu_i^{y_i}(1-\mu_i)^{1-y_i}\pi(\mu_i) \end{gather} whereas the marginal posterior distribution of $\mu_i$ in Model 2 is: \begin{align} p(\mu_i|y_i,\kappa) &\propto \int^1_0 \frac{\theta_i^{y_i + \mu_i\kappa - 1}(1-\theta_i)^{\kappa(1-\mu_i)-y_i}}{B\big(\kappa\mu_i,\kappa(1-\mu_i)\big)} d\theta \,\pi(\mu_i) \\ &\propto \frac{B\big(y_i+\mu_i\kappa,1-y_i+\kappa(1-\mu_i)\big)\pi(\mu_i) }{B\big(\kappa\mu_i,\kappa(1-\mu_i)\big)} \\ &\propto \mu_i^{y_i}(1-\mu_i)^{1-y_i} \pi(\mu_i) \end{align}

Thus any advantage gained from using Model 2 is computational. Overparameterizing hierarchical models, such as the addition of $\theta_i$ in Model 2, can sometimes improve the efficiency of the sampling procedure; for example, by introducing conditionally conjugate relationships between groups of parameters (see Jack Tanner's answer), or by breaking correlation among parameters of interest (google "Parmeter Expansion").

Answer 2

The reason for drawing the Bernouilli parameter from a beta distribution is that the beta is conjugate to the binomial. Using a conjugate prior distribution enables a closed-form solution to finding the posterior.

EDIT: clarifying. Either model will work. Even with MCMC, it's useful to have conjugate priors because that permits the use of specialized samplers for various types of distributions that are more efficient than generic samplers. For example, see the JAGS user manual sec. 4.1.1 and sec 4.2.