A note on regression diagnostics for generalized estimating equations: Empirical study on environmental disclosure determinants

Abstract

The aim of this paper is to describe and illustrate the application of generalized estimation equations and several diagnostic measures. The principal idea behind generalized estimating equations is to generalize and extend the usual likelihood score equation for a generalized linear model by including the covariance matrix of the clustered responses. The advantage of generalized estimating equations is that we do not need to specify the whole response distribution, only the mean structure and, with the aim to increase efficiency, the covariance structure consisting of a working correlation matrix along with the variance function defining the mean-variance relationship. The paper investigates, from a methodological point of view, various measures for the identification of the strength of association between a response variable and covariates including the coefficient of determination based on Wald Statistics, and the pseudo-coefficient of determination based on a quasi-likelihood method. Moreover, diagnostic measures for checking the adequacy of the generalized estimating equations method are considered and applied to a dataset to assess the impact of governance factors on environmental policy disclosure. The case study presents one of the most comprehensive applications of Generalized Estimating Equations regression diagnostics in the economics literature and is a novelty in the analysis of the Environmental Social and Governance disclosure determinants in the Non-Financial Industry.

Keywords:

• Diagnostic measures
• deletion diagnostics
• coefficient of determination
• generalized estimating equations
• marginal mean regression model
• environmental disclosure

Disclosure statement

No potential conflict of interest was reported by the author.