Parameters

Continuous parameters

Use param for continuous state:

param beta(p);
param u(m, 2);
param logsigma(1);

A dimension can be a positive integer literal or the name of a scalar supplied in the R data list. Model indexing is one-based, so the declarations above are addressed as beta(j), u(j,l), and logsigma(1).

Bounded discrete parameters

Use dparam for bounded integer state:

dparam gamma(p, 0, 1);

The final two values are inclusive lower and upper bounds. HOBBS updates a discrete coordinate by evaluating the local block target over every value in its finite support and drawing from the resulting conditional distribution.

Save only posterior means

Large latent arrays can be sampled normally without writing every retained draw:

param u(m, 2) save=mean;

The full parameter remains part of the Markov state. Only its post-burn-in mean is written to the companion mean output. Use this when posterior means are sufficient; quantiles and chain diagnostics require saved draws.

Choose the continuous sampler

param beta(p) sampler=rwmh;
param u(m, 2) sampler=slice;
param sigma(1) sampler=slice save=mean;

rwmh is the default. sampler= and save=mean are independent modifiers and can be written in either order. A dparam does not accept a continuous sampler modifier.