Residuals and diagnostics for multinomial regression models

In this paper, we extend the concept of a randomized quantile residual to multinomial regression models. Customary diagnostics for these models are limited because they involve difficult‐to‐interpret residuals and often focus on the fit of one category versus the rest. Our residuals account for associations between categories by using the squared Mahalanobis distances of the observed log‐odds relative to their fitted sampling distributions. Aside from sampling variation, these residuals are exactly normal when the data come from the fitted model. This motivates our use of the residuals to detect model misspecification and overdispersion, in addition to an overall goodness‐of‐fit Kolmogorov–Smirnov test. We illustrate the use of the residuals and diagnostics in both simulation and real data studies.

• Publication year

  2023

• Venue

  Statistical analysis and data mining

• Publication date

  2023-09-29

• Fields of study

  Mathematics, Computer Science

• Identifiers
  DOI 10.1002/sam.11645
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• A comparison of residual diagnosis tools for diagnosing regression models for count data

  2020cited by this paper
• A mixed effects multinomial logistic-normal model for forecasting baseball performance

  2019cited by this paper
• Chapter 8: Generalized Linear Models: Diagnostics

  2018cited by this paper
• Residuals and Diagnostics for Ordinal Regression Models: A Surrogate Approach

  2018cited by this paper
• Residuals and Diagnostics for Binary and Ordinal Regression Models: An Introduction to the sure Package

  2018cited by this paper
• Reprint of: Mahalanobis, P.C. (1936) "On the Generalised Distance in Statistics."

  2018cited by this paper
• How to define deviance residuals in multinomial regression.

  2015cited by this paper
• R: A language and environment for statistical computing.

  2014influential reference
• General models for resource use or other compositional count data using the Dirichlet-multinomial distribution.

  2013cited by this paper
• Multivariate Generalizations of the Multiplicative Binomial Distribution: Introducing the MM Package

  2012cited by this paper
• Modeling Baseball Player Ability with a Nested Dirichlet Distribution

  2009cited by this paper
• Graphical diagnostics to check model misspecification for the proportional odds regression model

  2009cited by this paper
• A better lemon squeezer? Maximum-likelihood regression with beta-distributed dependent variables.

  2006cited by this paper
• Generalized Linear Models

  2002cited by this paper
• ESTIMATION OF POPULATION TRAJECTORIES FROM COUNT DATA

  1997cited by this paper
• Randomized Quantile Residuals

  1996influential reference
• The North American Breeding Bird Survey

  1994cited by this paper
• Generalized Linear Models (2nd ed.)

  1992cited by this paper
• Regression-based specification tests for the multinomial logit model

  1987cited by this paper
• A Diagnostic Test for the Multinomial Logit Model

  1987cited by this paper
• The Statistical Analysis of Compositional Data

  1986cited by this paper
• Goodness of fit tests for the multiple logistic regression model

  1980cited by this paper
• On the compound multinomial distribution, the multivariate β-distribution, and correlations among proportions

  1962cited by this paper

• A Universal Method for Identifying and Correcting Induced Heave Error in Multi-Beam Bathymetric Surveys

  2026cites this paper
• Estimation of Parameters in Inhomogeneous Neyman‐Scott Processes Using Presence/Absence Data

  2026cites this paper
• Predatory Dogs as Drivers of Social Behavior Changes in the Central Himalayan Langur (Semnopithecus schistaceus) in Agro-Forest Landscapes

  2024cites this paper
• A multiclass logistic regression, Markov chain, and cellular automata integration for modelling urban land use change

  2024cites this paper