DHARMa

DHARMa (Diagnostics for HierArchical Regression Models) checks whether a fitted regression model is consistent with the data it was fitted to. For generalized linear and mixed models, raw residuals are hard to interpret. DHARMa instead simulates new data from the fitted model and uses them to compute scaled (quantile) residuals. These are standardized to values between 0 and 1 and can be read as intuitively as residuals from a linear regression.

Building on these residuals, the package provides plots and tests for typical misspecification problems: over- and underdispersion, zero-inflation, and residual spatial, temporal and phylogenetic autocorrelation. Supported are models from lme4, glmmTMB, GLMMadaptive, spaMM, mgcv (GAMs), phylolm and simple brms models, as well as lm and glm (including negative binomial models from MASS). Simulations created elsewhere, for example posterior predictive simulations from JAGS, Stan or BUGS, can be processed too.

Since 2024 the package has been extended in a project funded by the DFG; see Projects.

Authors: Florian Hartig (maintainer). Contributors: Lukas Lohse, Melina de Souza Leite, Cosmina Werneke.