Family-specific default TDMM priors

Return the family-specific prior defaults used by tdmm(). A partial named list can be supplied through priors= to modify supported prior parameters.

Usage

tdmm.priors(family = c("gaussian", "bernoulli", "poisson"))

Arguments

family

Response family.

Details

All Gamma distributions use shape and rate parameterization.

For all response families, the smoothing precision has a Gamma(1, 0.001) prior.

For Gaussian outcomes, the random-intercept precision has a Gamma(1, 0.005) prior when random_intercept = TRUE; the residual precision has a Gamma(1, 0.005) prior in either setting.

For Bernoulli and Poisson outcomes, the random-intercept standard deviation has a Uniform(0, 5) prior when random_intercept = TRUE. When random_intercept = FALSE, the random-intercept prior settings returned by tdmm.priors() are not used by JAGS.

For Bernoulli outcomes, null_space = list(sd = 10) specifies the standard deviation used for the two unpenalized spline components.

Users may modify the numerical parameters of these priors, but the prior distribution associated with each response family is fixed.

Value

A named list containing the prior settings for the selected response family. The list includes smoothing and random_intercept, with additional family-specific components where applicable.

Examples

tdmm.priors("bernoulli")
#> $smoothing
#> $smoothing$shape
#> [1] 1
#> 
#> $smoothing$rate
#> [1] 0.001
#> 
#> 
#> $random_intercept
#> $random_intercept$distribution
#> [1] "uniform_sd"
#> 
#> $random_intercept$max_sd
#> [1] 5
#> 
#> 
#> $null_space
#> $null_space$sd
#> [1] 10
#> 
#> 

# Example partial override:
priors <- list(smoothing = list(rate = 0.002))