Compute residuals for a glmmTMB object

Usage

# S3 method for class 'glmmTMB'
residuals(
  object,
  type = c("response", "pearson", "working", "deviance", "dunn-smyth"),
  re.form = NULL,
  ...
)

# S3 method for class 'glmmTMB'
deviance(object, ...)

Arguments

object: a “glmmTMB” object
type: (character) residual type
re.form: NULL to specify individual-level predictions; ~0 or NA to specify population-level predictions (i.e., setting all random effects to zero)
...: for method compatibility (unused arguments will throw an error)

Details

Residuals are computed based on predictions of type "response", i.e. equal to the conditional mean for non-zero-inflated models and to mu*(1-p) for zero-inflated models
Computing deviance residuals depends on the implementation of the dev.resids function from the object's family component; at present this returns NA for most "exotic" families (i.e. deviance residuals are currently only implemented for families built into base R plus nbinom1, nbinom2). Deviance residuals are based on the conditional distributions only, i.e. ignoring zero-inflation components.
Deviance is computed as the sum of squared deviance residuals, so is available only for the families listed in the bullet point above. See deviance.merMod for more details on the definition of the deviance for GLMMs.
for distributions in the exponential dispersion family (Gaussian, Poisson, binomial, Gamma), for models with a fixed dispersion parameter (Poisson, binomial) or constant dispformula component, reported Pearson residuals are only scaled by a factor proportional to the residual standard deviation (for compatibility with base R); divide these values by sigma(fitted_model) to get raw residuals scaled by the standard deviation. For all other distributions/models, Pearson residuals are scaled by the residual standard deviation. (The beta-binomial currently returns unscaled residuals.)