*3.2. Cluster Analysis*

With five cases identified as outliers and six cases registered as missing from the system, the two-step cluster analysis efficiently and automatically formed four clusters. Following the recommendations of Aguinis, Gottfredson, and Joo [49] about the best practices for defining, identifying, and handling outliers, we defined them as cluster analysis outliers. We handled them by performing the rest of the analyses (e.g., discriminant) with and without them. We found that they were non-influential outliers because they did not significantly change the rest of our results. Figure 2 depicts the centroids (means) of each cluster, expressed in standardized scores of job satisfaction and mental health measures. The silhouette coefficient (which was approximately 0.5) suggested that a four-cluster solution had fair levels of cohesion and separation. We named the four clusters, considering the centroids of job satisfaction and mental health. Cluster 1 was called unsatisfied-unhealthy and comprised 33% of the sample (258 employees), showing the lowest means on job satisfaction (−0.62) and mental health (−0.97). Cluster 2 was called unsatisfied-healthy and comprised 26.6% of the sample (208 employees), showing low levels of job satisfaction standardized means (−0.56), but high levels of mental health (0.76). Cluster 3 was called satisfied-unhealthy and comprised 24.8% of the sample (194 employees), in this case showing high levels of job satisfaction (0.83), but low levels of mental health (−0.15). Finally, Cluster 4 was called satisfied-healthy and comprised only 14.3% of the sample (112 employees), showing the highest levels of both job satisfaction (1.11) and mental health (1.24). To test whether the clusters were significantly different from one another, we conducted an analysis of variance (ANOVA). The results suggested that there were significant differences in job satisfaction (*F*(4, 772) = 276.41, *p* < 0.01) and health (*F*(4, 772) = 477.93, *p* < 0.01) among the four patterns. Tukey post-hoc analyses also suggested that all the clusters were significantly different from each other. Together, these results reflect different patterns of the relations between job satisfaction and mental health.
