Chapter 4: Heteroscedasticity in Organizational Research

Variance plays an important role in theory and research in human resource management and related fields. Variance refers to the dispersion of scores or residuals around a mean or, more generally, a predicted value (Salkind, 2007, 2010). In the general linear model, the mean square error provides an estimate of the dispersion of population error variance (Fox, 2016). As mean square error decreases, in general, the variability in the population error also decreases. In general linear models, it is assumed that the population error variance is constant across cases (i.e., observations in a sample). This assumption is known as homoscedasticity, or homogeneity of variance (Fox, 2016; King, Rosopa, & Minium, 2018; Rencher, 2000). When the homoscedasticity assumption is violated, it is referred to as heteroscedasticity, or heterogeneity of variance (Fox, 2016; Rosopa, Schaffer, & Schroeder, 2013). When heteroscedasticity is present in the general linear model, this results in incorrect standard errors, which can lead to biased Type I error rates and reduced statistical power (Box, 1954; DeShon & Alexander, 1996; White, 1980; Wilcox, 1997). This can threaten the statistical conclusion validity of a study (Shadish, Cook, & Campbell, 2002). Notably, heteroscedasticity has been found in a variety of organizational and psychological research contexts (Aguinis & Pierce, 1998; Antonakis & Dietz, 2011; Ostroff & Fulmer, 2014), thereby prompting research regarding best practices for detecting changes in residual variance and mitigating its negative effects (Rosopa et al., 2013).