Composition-weighting specification

Holds priors on the Dirichlet-multinomial composition-weighting overdispersion. The DM weight (the "theta" that scales the effective sample size) is otherwise an unpenalized free parameter; a linkage lets you put a prior on it through the same grammar as every other parameter. The three parameters target the three DM likelihoods:

theta_comp

age / length composition weight (comp_weights), per fleet.

theta_caal

conditional-age-at-length weight (caal_weights), per fleet.

theta_diet

diet composition weight (diet_comp_weights), per predator.

This is a prior-only process: the DM weight is a scalar, not a year-varying quantity, so only intercept formulas (~ 1) with a priors entry are meaningful. The prior is placed on the natural-scale DM weight theta = exp(weight), so a gamma() prior reads naturally. A linkage on a fleet whose Comp_distribution (or CAAL_distribution) is not "DirichletMultinomial" errors, since the weight has no effect there.

Usage

build_composition(linkages = NULL)

Arguments

linkages

Optional named list of linkage_spec() objects keyed by theta_comp / theta_caal (per fleet by default, by = ~ fleet) or theta_diet (per predator by default, by = ~ species).

Value

A list of composition-weighting settings for fit_mod().

Examples

# \donttest{
# A weak gamma prior on the DM overdispersion of every fleet's age comps
build_composition(linkages = list(
  theta_comp = linkage_spec(~ 1, by = ~ fleet,
                            priors = list(`(Intercept)` = gamma(2, 0.5)))))
#> $linkages
#> $linkages$theta_comp
#> <Rceattle linkage spec>
#>   param:   theta_comp
#>   formula: ~1
#>   prior:    (Intercept) ~ gamma(2, 0.5)
#>   link:    log
#> 
#> 
# }