Check model residuals

Detect overall deviations of residuals from the expected distribution using a simulation-based approach. Provides standardized residuals that are more interpretable for GLMs than classical residual plots.

Usage

check_residuals(object, n_simulations = 30)

Arguments

object

A fitted model object (e.g. of class "glm").

n_simulations

Number of simulations to generate residuals. Default = 30.

Value

An object of class "check_residuals", which is a list with:

df

Data frame with theoretical quantiles (x) and observed residuals (y).

p.val

P-value from Kolmogorov-Smirnov test against uniform(0,1).

Invisibly returns the object.

Details

Misspecifications in GLMs cannot reliably be diagnosed with standard residual plots, because the expected distribution of the data changes with fitted values. check_residuals() uses a simulation-based approach (similar to a parametric bootstrap or Bayesian p-value) to generate standardized residuals between 0 and 1, which can be interpreted intuitively like residuals from linear models.

This function wraps DHARMa::simulateResiduals(), adapted for convenience.

Note: If all simulations for a data point have the same value (e.g. all zeros), an error may occur (Error in approxfun: need at least two non-NA values). Increasing n_simulations can help in such cases.

References

Dunn, K. P., & Smyth, G. K. (1996). Randomized quantile residuals. JCGS, 5, 1–10.

Gelman, A., & Hill, J. (2006). Data analysis using regression and multilevel/hierarchical models. Cambridge University Press.

Hartig, F. (2020). DHARMa: Residual Diagnostics for Hierarchical (Multi-Level / Mixed) Regression Models. R package version 0.3.0. https://CRAN.R-project.org/package=DHARMa

Author

Martin Haringa

Examples

if (FALSE) { # \dontrun{
m1 <- glm(nclaims ~ area, offset = log(exposure),
          family = poisson(), data = MTPL2)
cr <- check_residuals(m1, n_simulations = 50)
autoplot(cr)
} # }