screen_control() stores threshold settings used by
screen_gllvmTMB(). The defaults are deliberately conservative and
advisory: they flag data conditions to inspect before fitting, but they
do not remove traits, choose a latent rank, or validate a model.
Usage
screen_control(
module = "binomial",
rare_warn_n = 10,
rare_strong_n = 5,
prevalence_warn = c(0.05, 0.95),
prevalence_strong = c(0.02, 0.98),
phi_warn = 0.9,
phi_strong = 0.95,
discordant_warn_n = 10,
discordant_strong_n = 5,
hamming_rate_warn = 0.01,
separation = c("none", "fixed"),
separation_tolerance = 1e-04
)Arguments
- module
Screening module. The first implemented module is
"binomial".- rare_warn_n, rare_strong_n
Minority success/failure counts below these values trigger increasingly strong warnings.
- prevalence_warn, prevalence_strong
Length-two numeric vectors giving lower and upper prevalence values for advisory imbalance notes.
- phi_warn, phi_strong
Pairwise absolute-phi thresholds.
- discordant_warn_n, discordant_strong_n
Pairwise discordant-count thresholds for near-duplicate binary traits.
- hamming_rate_warn
Pairwise normalized Hamming-distance threshold.
- separation
Optional fixed-design separation screen. The default
"none"preserves the existing advisory screen and does not requiredetectseparation;"fixed"returns one certificate per maximal coefficient-connected fixed-effect block.- separation_tolerance
Solver tolerance passed to
detectseparation.