Running LOOCV on hierarchical model outputs

Performs Leave-One-Out Cross-Validation (LOOCV) on hierarchical model outputs to assess the influence of individual simulated animals on population-level estimates. Supports analyses with or without groups.

In each iteration, the function removes one individual, refits the hierarchical model to the remaining dataset, and recalculates the target population-level estimates. This process is repeated until every individual has been excluded once.

This approach provides insight into how sensitive overall conclusions are to specific individuals. This helps identify influential individuals and assess robustness.

Usage

run_meta_loocv(
  rv,
  set_target = c("hr", "ctsd"),
  subpop = FALSE,
  trace = FALSE,
  ...
)

Arguments

rv

A reactiveValues object or list containing simulation inputs, fitted models, and (optionally) group assignments.

set_target

Character vector specifying the target metrics. Options are "hr" for home range area and/or "ctsd" for movement speed. Defaults to c("hr", "ctsd").

subpop

Logical; if TRUE, analyzes population-level inferences by groups (e.g., males vs. females). Requires valid group assigments in rv.

trace

Logical; if TRUE, prints progress and diagnostic messages. Default is FALSE.

...

Additional arguments for advanced control:

.only_max_m

Logical. If TRUE, runs the meta-analysis only at the maximum population sample size, skipping all intermediate sample sizes.

.progress

Integer. Displays a progress bar.

.m

Integer. Specifies exact sample size to use. Overrides automatic sequence generation. Accepts a single value.

.lists

List (optional); supplies precomputed input objects, typically created via .build_meta_objects().

Value

A data frame containing summarized simulation outputs.

Author

Inês Silva i.simoes-silva@hzdr.de

Examples

if(interactive()) {
   run_meta_loocv(rv, set_target = "hr")
}