Extract spatial range and variance from a fitted spatial model

Summarises the posterior of spatial hyperparameters. For sampler-tier fits this reads the raw hyperparameter draws; for a nested-Laplace spatial fit it summarises the outer hyperparameter grid. Works with ICAR, BYM2, GP (NNGP), CAR, SPDE, and SVC spatial types.

Usage

spatial_range(object, probs = c(0.025, 0.975))

Arguments

object: A tulpa_fit object fitted with a spatial component.
probs: Quantile probabilities for the summary (default 0.025, 0.975).

Value

A data.frame with rows for each spatial hyperparameter and columns mean, sd, and one quantile column per entry of probs (named from the probability, e.g. q2.5, q97.5 for the defaults).

Details

A range is reported in the units of the coordinates as supplied, also when the field was fitted on standardized coordinates (scale_coords = TRUE, the default of spatial_gp() and spatial_svc()). A mode = "laplace" fit conditions on the field's hyperparameters rather than inferring them, so it carries no posterior for this function to summarise.

Examples

# \donttest{
set.seed(1)
S <- 12L
adj <- matrix(0, S, S)
for (i in 1:(S - 1)) adj[i, i + 1] <- adj[i + 1, i] <- 1
d <- data.frame(region = factor(rep(1:S, each = 5)), x = rnorm(S * 5))
field <- as.numeric(scale(cumsum(rnorm(S, 0, 0.5))))
d$y <- rpois(nrow(d), exp(0.3 + 0.4 * d$x + field[d$region]))
fit <- tulpa(y ~ x + spatial(region), data = d, family = "poisson",
             spatial = spatial_car(adj, level = "group", group_var = "region"))
spatial_range(fit)
# }