*3.4. Measures and Covariates*

Socio-demographic and other mental health related data were collected through a selfadministered survey. The variables used in to measure burnout were: emotional exhaustion, cynicism, and low effectiveness [8]. All items were scored on a 7-point Likert scale. The variables used to measure QWL were: supervisor support, co-worker support, good work environment, professional respect, work–life balance, and skills development [4]. QWL was then generated as a new variable, computed by using the six prior variables, attributing the value 1 if the levels of agreement were positive and 0 otherwise. Regarding the mediators, the variables of contribution to productivity and appropriate salary were used. The former was scored on a 7-point Likert scale, ranging from 1 if the worker felt they contribute to the organization's productivity, 0 otherwise, while the latter ranged from 1 to 7, 1 being if the workers did not agree and 7 if they totally agreed.

#### *3.5. Data Analysis*

The data collected from Google® Forms were exported to an Excel spreadsheet, and all statistical analyses were carried out using STATA Statistics (version 14.1). Variable characterization was performed by means of absolute and relative frequencies, means and standard deviations (SDs).

Two mediation analyses were performed using two models. Model 1 used the intermediate variable (contribution to productivity), and Model 2 used two intermediate variables (contribution to productivity and appropriate salary). These mediators intend to explain how or why a set of independent variables influences an outcome (here the QWL). To do so, structural equation modeling (SEM) was used, as it is a powerful multivariate technique, which makes use of a conceptual model, path diagram, and system of linked regression-style equations in order to capture complex and dynamic relationships within an agglomerate of both observed and unobserved variables. This technique allows a reciprocal role played by a variable and enables the inference of causal relationships. According to Gunzler et al. (2013), the option for the SEM framework in a mediation analysis is advantageous when the model comprises latent variables such as quality of life or stress, as it will make interpretation and estimation easier [61]. As SEM was created partly to test complex mediation models in a single analysis, this technique simplifies the testing of mediation hypotheses. Moreover, the option here is also justified as this work extends the mediation process to multiple independent variables.
