`influ()` is the model-neutral entry point for influence diagnostics. It dispatches on the fitted model class and returns the same compact [influ_diag] structure for every backend.
Details
Standardised focus indices require one term depending only on the focus variable. If a focus term also involves another variable, or several terms contain the focus, the diagnostic retains their influence but warns and omits the standardised index. A prediction-grid reference changes the centring distribution; it does not automatically marginalise interactions. Define such indices in a separate model-specific prediction workflow. Step plots require an unambiguous focus effect. Generalised grouped residual departures do not require a coefficient baseline.
Fixed offsets/exposure are supported only for single-component log-link ratios and identity-link contrasts; other offset-dependent reference calculations fail explicitly. Nominal summaries describe the observed response, not response divided by exposure. brms lognormal models require an identity location link and constant `sigma`. Mean-parameterised lognormal backends require a log link; `glmmTMB` may have varying dispersion, but its dispersion effects are not separately decomposed.
Examples
data(lobsters_per_pot)
model <- glm(
lobsters ~ year + month + poly(depth, 2) + poly(soak, 2),
family = poisson(link = "log"),
data = lobsters_per_pot
)
diagnostic <- influ(model, focus = "year")
diagnostic
#> <influ_diag>
#> Backend: glm
#> Response: single poisson (log)
#> Focus: year
#> Terms: 4
#> Focus levels:18
#> Uncertainty: analytic covariance
#> Retained: summary
head(influ_effects(diagnostic))
#> focus level term component scale estimate std_error lower
#> 1 year 2000 year conditional link 0.26372139 0.05631806 0.15334002
#> 2 year 2001 year conditional link 0.32518319 0.05439295 0.21857498
#> 3 year 2002 year conditional link 0.35979168 0.05009542 0.26160645
#> 4 year 2003 year conditional link 0.06774274 0.04881851 -0.02793978
#> 5 year 2004 year conditional link 0.23960938 0.04452008 0.15235163
#> 6 year 2005 year conditional link 0.02245967 0.05465950 -0.08467098
#> upper method
#> 1 0.3741028 analytic covariance
#> 2 0.4317914 analytic covariance
#> 3 0.4579769 analytic covariance
#> 4 0.1634253 analytic covariance
#> 5 0.3268671 analytic covariance
#> 6 0.1295903 analytic covariance