Skip to contents

Report identifiability diagnostics for a fitted model

Source: R/diagnostics.R
check_tls.Rd

check_tls() re-runs the data-adequacy diagnostics on a fitted profile_tls object, including the two post-fit checks that fit_tls() cannot run before the model exists: whether any fitted CTmax is extrapolated beyond the assayed temperatures (item 7), and whether phi has reached the binomial limit (item 8). Each concern is emitted as a cli::cli_warn(). Use it to audit a fit, or after suppressWarnings() around fit_tls().

Usage

check_tls(fit)

Arguments

fit

A profile_tls fit from fit_tls(), or a freq_tls workflow from fit_4pl().

Value

Invisibly, a character vector of the diagnostic codes that fired.

Details

The profile-geometry diagnostics (items 9-12) are emitted by confint() and profile() when those are called, not here, because they require the profile likelihood.

Recovery guide

The warning code returned by check_tls() identifies the next action:

  • temps: assay at least three distinct temperatures spanning the survival transition.

  • durations / durations_per_temp: assay at least three distinct durations per temperature, with times on both sides of the transition.

  • no_mortality: extend to hotter or longer exposures until mortality occurs.

  • all_mortality: add cooler or shorter exposures that retain survivors.

  • threshold: extend the design until observed survival straddles 0.5.

  • up_not_approached / low_not_approached: add milder / harsher conditions approaching survival 1 / 0, or do not interpret that asymptote.

  • ctmax_extrapolated: expand the assayed temperature range to bracket CTmax; otherwise report it explicitly as extrapolated.

  • phi_binomial_limit: consider the simpler binomial family.

After changing the design or family, refit and rerun check_tls(). For an existing data set that cannot be augmented, use the warning to limit the scientific claim; vignette("profile-likelihood") explains the strict fallback = FALSE diagnostic and the default bootstrap recovery attempt.

Examples

d <- simulate_tls(family = "binomial", CTmax = 36, z = 4, seed = 1)
fit <- fit_tls(d, y = survived, n = total, time = duration, temp = temp,
               family = "binomial", tref = 1)
codes <- check_tls(fit)
codes # character(0) means no data-adequacy diagnostic fired
#> character(0)