Prior predictive check

Draws parameter values from the priors, simulates data with the design of data (its sites, covariates, occasions, recorders, and verification), and compares a few summaries of the simulated data with the observed ones. If the observed values sit far in the tails of what the priors produce, the priors (or the model) are at odds with the data before any fitting.

Usage

fp_prior_predictive(
  data,
  occ = ~1,
  det = ~1,
  fp = ~1,
  ver = ~1,
  markov = NULL,
  random_recorder = !is.null(data$recorder),
  random_fp = FALSE,
  priors = fp_priors(),
  ndraws = 200,
  seed = NULL
)

# S3 method for class 'fp_prior_predictive'
plot(x, ...)

Arguments

data

An fp_data object.

occ, det, fp, ver, markov, random_recorder, random_fp, priors

As in fp_occu().

ndraws

Number of prior draws, one simulated dataset each.

seed

Optional integer seed.

x

An fp_prior_predictive object.

...

Unused.

Value

An object of class fp_prior_predictive: a data frame with one row per draw and one column per summary, with the observed values as the attribute observed. print() shows the prior predictive intervals and where the observed values fall; plot() draws them.

Details

Each prior draw takes the intercepts and slopes from their Normal priors, the gap from its truncated Normal prior, and the standard deviations of the random effects from their half-Normal priors, with new recorder and site effects drawn from them. Covariates are taken as they are in data, so the priors are judged on the scale the model will use. The summaries are the share of sites with at least one automated detection (naive occupancy), the share of automated occasions with a detection, the share of verified occasions with a detection, the mean occupancy probability, and the mean false-positive rate over sites.

See also

fp_priors(), simulate.fp_occu()

Examples

sim <- simulate_fp(nsite = 100, nocc = 10, seed = 1)
pp <- fp_prior_predictive(sim$data, occ = ~ x, fp = ~ 1,
                          random_recorder = FALSE, ndraws = 50, seed = 1)
pp
#> Prior predictive check (50 draws)
#> 
#>          summary     q05 median   q95 observed share_below_observed
#>  naive_occupancy 0.24750  0.845 1.000    0.700                 0.30
#>   automated_rate 0.03030  0.273 0.712    0.319                 0.58
#>    verified_rate 0.03594  0.208 0.532    0.208                 0.48
#>         mean_psi 0.22774  0.487 0.826       NA                   NA
#>          p10_bar 0.00465  0.107 0.686       NA                   NA
#> 
#> Observed values near 0 or 1 in `share_below_observed` sit in the tails of the prior predictive distribution.