*2.4. Data Analysis*

To identify the four wellbeing patterns, we performed cluster analyses. One of the advantages of using cluster analysis is that unlike other methods that emphasize the relationship among variables, clustering involves sorting cases or variables according to their similarity in one or more dimensions and producing groups that maximize within-group similarity and minimize between-group similarity [44]. Therefore, to identify the four groups of wellbeing patterns, the 783 employees were clustered based on their individual levels of job satisfaction and mental health, applying a two-step cluster analysis procedure. Loglikelihood measured the distance between job satisfaction and mental health. The clustering criterion was Schwarz's Bayesian Criterion (BIC). Finally, to balance the distribution of responses on the job satisfaction and mental health variables, we standardized these two variables to *Z*-scores (*M* = 0, *SD* = 1) before performing the cluster analyses.

To test our hypotheses, we employed discriminant analysis to test the unique differentiating role of stress (role ambiguity, role conflict, and role overload), job importance, and overqualification across the four patterns. We conducted a stepwise solution to remove variables that did not make a unique contribution to the predictive and discriminatory function at a probability of 0.05 or less. The stepwise criterion was minimization of Wilks' lambda. Similar studies analyzing wellbeing profiles [45] applied discriminant analysis considering Wilks' lambda stepwise minimization criteria, as we did in the current study. Discriminant analysis is a method used in a multi-group setting to find out if a set of independent variables (nominal and/or continuous) are related to group membership and how they are combined to better understand group differences [46]. In fact, various authors suggest the use of cluster analysis in combination with discriminant analysis for further validation of clusters [47].
