Structural equation modeling (SEM), a powerful multivariate statistical technique, has seen significant scientific advancements that have enabled rigorous modeling of complex latent variables across a broad spectrum of disciplines in the social, behavioral, and health sciences. However, selecting an appropriate estimation approach can be challenging due to multiple assumptions and conditions, such as sample size and the number of response options in questionnaires reflecting latent structures. The Copenhagen Psychosocial Questionnaire (COPSOQ) has been used internationally to assess the impact of psychosocial risk on workers’ health and well-being. In a survey conducted using the COPSOQ instrument with employees of a Portuguese company, a small sample was obtained, and the proposed theoretical model was estimated using four distinct estimation methods: maximum likelihood robust (MLR), diagonally weighted least squares (DWLS), Bayesian estimation (BE), and the consistent partial least squares (PLSc). When comparing the results obtained, the MLR, BE, and PLSc estimators lead to similar results. This reinforces the idea that the PLSc estimator can be considered appropriate when we have ordinal (non-normal) data, small samples, and complex models. The latent construct “quality of leadership” showed statistically significant effects on the mediating constructs “justice” and “quantitative demands,” which in turn influenced “job satisfaction” and its effects on “stress.” The model estimated produces insights considered relevant regarding the health and well-being of the company’s workforce.