Returns the posterior draws of a rstanarm::stan_glm() or
rstanarm::stan_glmer() fit together with the log prior and
log-likelihood draws that the priorsense package needs for its
power-scaling diagnostics (Kallioinen et al., 2023), such as
priorsense::powerscale_sensitivity() and priorsense::powerscale_plot_dens().
priorsense does not read rstanarm fits directly; bmbeR rebuilds the log
prior from rstanarm::prior_summary() in the same way as
prior_sensitivity().
Arguments
- fit
A
stanregobject fromrstanarm::stan_glm(),rstanarm::stan_glmer()orfit_model_with_prior(), fitted withoutQR = TRUE.
Value
A posterior::draws_array (iterations x chains x variables).
Details
The draws contain:
every model parameter, with
(Intercept)evaluated at the predictor means (the parameter rstanarm places its prior on);lprior[1], ...,lprior[p]: the log density of each population-level prior, in the row order ofprior_table(), so that priorsense'sprior_selectionargument can pick individual priors;log_lik[1], ...,log_lik[n]: the pointwise log-likelihood.
For multilevel models the group-level covariance prior has no lprior
entry, so priorsense power-scales only the population-level priors.
References
Kallioinen, N., Paananen, T., Bürkner, P.-C., & Vehtari, A. (2023). Detecting and diagnosing prior and likelihood sensitivity with power-scaling. Statistics and Computing, 34, 57. doi:10.1007/s11222-023-10366-5
Examples
# \donttest{
data(kidiq, package = "rstanarm")
fit <- rstanarm::stan_glm(kid_score ~ mom_iq, data = kidiq,
chains = 2, iter = 1000, refresh = 0)
draws <- as_priorsense_draws(fit)
posterior::variables(draws)[1:6]
#> [1] "(Intercept)" "mom_iq" "sigma" "lprior[1]" "lprior[2]"
#> [6] "lprior[3]"
if (requireNamespace("priorsense", quietly = TRUE)) {
priorsense::powerscale_sensitivity(draws)
}
#> Sensitivity based on cjs_dist
#> Prior selection: all priors
#> Likelihood selection: all data
#>
#> variable prior likelihood diagnosis
#> (Intercept) 0.001 0.103 -
#> mom_iq 0.001 0.094 -
#> sigma 0.005 0.092 -
# }