The coefficient of determination <i>R</i><sup>2</sup> and intra-class correlation coefficient from generalized linear mixed-effects models revisited and expanded.

Nakagawa S, Johnson PCD, Schielzeth H. · 2017

Read The coefficient of determination <i>R</i><sup>2</sup> and intra-class correlation coefficient from generalized linear mixed-effects models revisited and expanded. by Nakagawa S, Johnson PCD, Schielzeth H. (2017) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5636267.txt.

Introduction

Nakagawa and co-authors revisit two summary measures for mixed-effects models. Their methods paper covers the coefficient of determination, R2, and the intra-class correlation coefficient. Earlier work handled only Poisson and binomial models. This paper extends both measures to other kinds of data. The focus falls on negative binomial and gamma models, which biologists use often.

The authors explain two tools that help readers grasp the math. One is Jensen's inequality, and the other is the delta method. They also give special notes for binary and proportion data. Worked examples from ecology and evolution show the method in R. The result is a guide that works across fields and software.

Catalog record

CREATORNakagawa S, Johnson PCD, Schielzeth H.
YEAR2017
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC5636267

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The coefficient of determination R2 and intra-class correlation coefficient from generalized linear mixed-effects models revisited and expanded By Shinichi Nakagawa, Paul C D Johnson, Holger Schielzeth Published in: Journal of the Royal Society Interface DOI: 10.1098/rsif.2017.0213 ABSTRACT Abstract The coefficient of determination R2 quantifies the proportion of variance explained by a statistical model and is an important summary statistic of biological interest. However, estimating R2 for generalized linear mixed models (GLMMs) remains challenging. We have previously introduced a version of R2 that we called for Poisson and binomial GLMMs, but not for other distributional families. Similarly, we earlier discussed how to estimate intra-class correlation coefficients (ICCs) using Poisson and binomial GLMMs. In this paper, we generalize our methods to all other non-Gaussian distributions, in particular to negative binomial and gamma distributions that are commonly used for modelling biological data. While expanding our approach, we highlight two useful concepts for biologists, Jensen's inequality and the delta method, both of which help us in understanding the properties of GLMMs. Jensen's inequality has important implications for biologically meaningful interpretation of GLMMs, whereas the delta method allows a general derivation of variance associated with non-Gaussian distributions. We also discuss some special considerations for binomial GLMMs with binary or proportion data. We illustrate the implementation of our extension by worked examples from the field of ecology and evolution in the R environment. However, our method can be used across disciplines and regardless of statistical environments. Keywords: repeatability, heritability, goodness of fit, model fit, variance decomposition, reliability analysis INTRODUCTION One of the main purposes of linear modelling is to understand the sources of variation in biological data. In this context, it is not surprising that the coefficient of determination R2 is a commonly reported statistic, because it represents the proportion of variance explained by a linear model. The intra-class correlation coefficient (ICC) is a related statistic that quantifies the proportion of variance explained by a grouping (random) factor in multilevel/hierarchical data. In the field of ecology and evolution, a type of ICC is often referred to as repeatability R, where the grouping factor is often individuals that have been phenotyped repeatedly [1,2]. We have reviewed methods for estimating R2 and ICC in the past, with a particular focus on non-Gaussian response variables in the context of biological data [2,3]. These previous articles featured generalized linear mixed-effects models (GLMMs) as the most versatile engine for estimating R2 and ICC (specifically and ICCGLMM). The descriptions were initially limited to random-intercept GLMMs, but have later been extended to random-slope GLMMs [4], widening the applicability of these statistics (see also [5,6]). However, at least one important issue seems to remain. Currently, these two statistics are only described for binomial and Poisson GLMMs. Although these two types of GLMM are arguably the most popular [7], there are other families of distributions that are commonly used in biology, such as negative binomial and gamma distributions [8,9]. In this paper, we revisit and extend and ICCGLMM to more distributional families with a particular focus on negative binomial and gamma distributions. In this context, we discuss Jensen's inequality and two variants of the delta method, which are hardly known among biologists. These concepts are useful not only for generalizing our previous methods, but also for interpreting the results of GLMMs. Furthermore, we refer to some special considerations when obtaining and ICCGLMM from binomial GLMMs for binary and proportion data, which we did not discuss in the past [2,3]. We provide worked examples inspired from the field of ecology and evolution, focusing on implementation in the R environment [10] and finish by referring to two alternative approaches for obtaining R2 and ICC from GLMMs along with a cautionary note. DEFINITIONS OF , ICCGLMM AND OVERDISPERSION To start with, we present and ICCGLMM for a simple case of Gaussian error distributions based on a linear mixed-effects model (LMM, hence also referred to as and ICCLMM). Imagine a two-level dataset where the first level corresponds to observations and the second level to some grouping/clustering factor (e.g. individuals with repeated measurements) with k fixed-effect covariates. The model can be written as (referred to as Model 1): [formula: 2.1] [formula: 2.2] [formula: 2.3] where yij is the jth observation of the ith individual, is the jth value of the ith individual for the hth of k fixed-effect predictors, β0 is the (grand) intercept, βh is the regression coefficient for the hth predictor, αi is an individual-specific effect, assumed to be normally distributed in the population with the mean and variance of 0 and , ɛij is an observation-specific residual, assumed to be normally distributed in the population with mean and variance of 0 and , respectively. For this model, we can define two types of R2 as [formula: 2.4] [formula: 2.5] [formula: 2.6] where represents the marginal R2, which is the proportion of the total variance explained by the fixed effects, represents the conditional R2, which is the proportion of the variance explained by both fixed and random effects, and is the variance explained by fixed effects [11]. As marginal and conditional R2 differ only in whether the random effect variance is included in the numerator, we avoid redundancy and present equations only for marginal R2 in the following. Similarly, there are two types of ICC: [formula: 2.7] and [formula: 2.8] If no fixed effects are fitted (other than the intercept), so that ICCLMM(adj) equals ICCLMM. In such a case, the ICC should not be called ‘adjusted’ (sensu [2]). For an ICC value to be adjusted for a source of variance, that variance must be more than 0 and omitted from the ICC calculation. As the two versions of ICC differ only in whether the fixed-effect variance, calculated as in equation (2.6), is included in the denominator, we avoid redundancy and present equations only for adjusted ICC in the following. One of the main difficulties in extending R2 from LMMs to GLMMs is defining the residual variance . For binomial and Poisson GLMMs with an additive dispersion term, we have previously stated that is equivalent to where is the variance for the additive overdispersion term, and is the distribution-specific variance [2,3]. Here, overdispersion represents the excess variation relative to what is expected from a certain distribution and can be estimated by fitting an observation-level random effect (OLRE; [12,13]). Alternatively, overdispersion in GLMMs can be implemented using a multiplicative overdispersion term [14]. In such an implementation, we stated that is equivalent to where ω is a multiplicative dispersion parameter estimated from the model [2]. However, obtaining for specific distributions is not always possible, because in many families of GLMMs, (observation-level variance) cannot be clearly separated into (overdispersion variance) and (distribution-specific variance). It turns out that binomial and Poisson distributions are special cases where can be usefully calculated, because either all overdispersion is modelled by an OLRE (additive overdispersion) or by a single multiplicative overdispersion parameter (multiplicative overdispersion). This is not the case for other families. However, as we will show below, we can always obtain the GLMM version of (on the latent scale) directly. We refer to this generalized version of as ‘the observation-level variance’ here rather than the residual variance (but we keep the notation ). Note that the

Read the complete text