“…Early studies in the going-concern opinion literature used discriminant analysis to model the decision (e.g., Altman & McGough, 1974;McKee, 1976). However, because going-concern prediction cannot rely on the contradictory assumptions of a multivariate normal distribution of explanatory variables and equal covariance metrics between two groups (Carson et al, 2013), models for predicting goingconcern opinions shifted from discriminant analysis to logit analysis (e.g., Menon & Schwartz, 1987;Harris & Harris, 1990), probit analysis (e.g., Koh & Brown, 1991), neural networks (e.g., Serrano-Cinca, 1996), decision tree (Koh & Low, 2004), support vector machine (Martens et al, 2008), and also a machine-learning random forest model (e.g., Hsu & Lee, 2020). For instance, Bellovary et al (2007) identified twenty-seven statistical models developed to predict issuance of a goingconcern opinion, including multivariate discriminant analysis (MDA).…”