Goodness-of-fit tests for a tobs fit — tobs_gof

The tulpa goodness-of-fit generics for tobs_fit: dispersion, zero-inflation, and outlier counts compared against the fitted model's simulated replicates. For the Kolmogorov-Smirnov uniformity test, hand pit_residuals() to tulpa::test_uniformity(), which takes a PIT vector directly.

Usage

# S3 method for class 'tobs_fit'
test_dispersion(
  object,
  n.samples = 250,
  nsim = NULL,
  seed = NULL,
  observed = NULL,
  alternative = c("greater", "two.sided", "less"),
  ...
)

# S3 method for class 'tobs_fit'
test_zero_inflation(
  object,
  n.samples = 250,
  nsim = NULL,
  seed = NULL,
  observed = NULL,
  ...
)

# S3 method for class 'tobs_fit'
test_outliers(
  object,
  n.samples = 250,
  nsim = NULL,
  seed = NULL,
  observed = NULL,
  ...
)

Arguments

object: A fitted tobs_fit.
n.samples: Number of posterior-predictive replicates to simulate.
nsim: The same budget under the name the tulpa default methods use; overrides n.samples when given.
seed: Optional random seed for the replicates; the caller's RNG stream is restored afterwards.
observed: Must be NULL: a tobs fit scores the response it was fitted to. Accepted so a call spelled for the tulpa default reaches the method.
alternative: Tail of the dispersion p-value: "greater" (default, over-dispersion), "two.sided" or "less".
...: Must be empty; an unrecognised argument is an error.

Value

A list with the observed statistic (observed), its posterior-predictive expectation (expected), their ratio (ratio) and a tail p-value (p.value). The names are the same on every family, so a caller can read the statistic without branching on the model type.