Participants
A non-probabilistic sample was composed of 400 Brazilian workers (55.8% female), 18 to 68 years of age (M = 35.6; SD = 10.5). Individuals were from the Southeast (68.5%), South (19.5%), Northeast (4.8%), Central-west (5.5%), and North (1.7%) of the country. In addition, they were attending or had completed postgraduate studies (54.0%), undergraduate (37.8%), high school (7.5%), or elementary education (0.7%). The professionals held different positions in which they had been for a mean of 64.8 months (SD = 82.9; approximately 5.4 years) and worked a mean of 33.1 h (SD = 17.3) per week in private (62%) or public (32.5%) institutions or autonomously (5.5%), in person (59.7%) or remotely (40.3%).
Instruments
The Rigid and Flexible Persistence Scale (Vallerand et al., 2022
)
Initially constructed in English, eight items constitute the scale that assesses persistence in two dimensions specific to the activity, in this case work: rigid persistence (e.g., I am willing to do anything to reach the top at work/Estou disposto a fazer qualquer coisa para chegar ao topo no trabalho.) and flexible persistence (e.g., I work hard at my work goals, but other things matter as well/Eu trabalho duro nos objetivos dos meu trabalho, mas outras coisas também são importantes.). The instrument is of the self-report type, in which participants must indicate how much they agree with each statement using a seven-point Likert-type scale, in which 1 corresponds to “do not agree at all” and 7 to “very strongly agree”. Results of both exploratory and confirmatory factor analyses supported the dual structure of the scale reflecting rigid and flexible persistence. The internal consistencies of the dimensions of the original scale (Vallerand et al., 2022, study 1) showed adequate reliability, indicated by Cronbach’s alpha coefficient, 0.85 for rigid persistence and 0.76 for flexible persistence.
The Passion Scale (Vallerand et al., 2003
)
The Passion Scale has been fully validated and shown to possess high levels of validity and reliability (see Marsh et al. 2013; Vallerand and Rahimi, in press; Vallerand et al., 2003). Adapted to the Brazilian context by Peixoto et al. (2019), the scale was used to measure passion for work. It consists of 12 items, six for harmonious passion (e.g. My work is in harmony with other activities in my life/Meus trabalho está em harmonia com as outras atividades em minha vida.) and six for obsessive passion (e.g. I have difficulties controlling my urge to work/Tenho dificuldade em controlar minha necessidade em desempenhar o meu trabalho.) answered on a Likert-type scale ranging from 1 (strongly disagree) to 7 (strongly agree). Internal consistency showed adequate reliability both dimensions, harmonious passion (α = 0.90) and obsessive passion (α = 0.86). The two-factor model showed adequate fit for instrument’s internal structure (χ2/df) = 3.96, TLI = 0.94, CFI = 0.96, RMSEA = 0.08, AIC = 23,992.86, BIC = 24,225.10).
Occupational Self-efficacy Scale–Short form (OSS-SF; Rigotti et al. 2008)
The OSS-SF refers to the reduced form of the OSS (Schyns and von Collani, 2002). Translated and adapted to the Brazilian context by Damásio et al. (2014), it was named Escala de Autoeficácia Ocupacional – Versão Reduzida. The instrument consists of six items (e.g. I meet the goals that I set for myself in my job/Eu alcanço as metas que eu estabeleço para mim mesmo em meu trabalho.), which are answered on a Likert-type scale, ranging from 1 (strongly disagree) to 6 (strongly agree). The single-factor structure was demonstrated by means of confirmatory factor analysis (χ2/df = 3.45, SRMR = 0.05, RMSEA = 0.09, CFI = 0.98, TLI = 0.95, CAIC = 461.669), and internal consistency showed adequate reliability (α = 0.78).
Procedures
Cross-cultural adaptation
With the authorization of the author of the original version, Robert J. Vallerand, the RFPS was independently translated from English into Portuguese by four bilingual professionals (namely, two psychologists, a professor and an English teacher). Subsequently, a synthesis version was obtained by an expert committee composed of four researchers specializing in psychological assessment and psychometrics. In a subsequent step the instrument was analyzed based on four criteria: clarity of language, practical relevance, theoretical relevance, and theoretical dimensions by three judges that were experts in Organizational and Work Psychology. A group composed of five professionals, two doctors, a salesperson, a seamstress and a process engineer (60% male), aged between 25 and 37 years (M = 30; SD = 4.58) was accessed to evaluate the scale content intelligibility. None of the participants had difficulty interpreting the instructions or items of the instrument since the RFPS had clearly and objectively drafted items. Finally, the Brazilian version of the scale was back-translated into English by a native translator, which made it possible to verify the equivalence between the content of the original and the Brazilian version.
Ethical aspects
The project was submitted to the Human Research Ethics Committee, and after approval the instruments were compiled on the Google Forms platform. Due to the COVID-19 pandemic data was collected online. The link to answer the online instruments was sent to participants from social networks. The participants who agreed to participate initially completed the consent form. Subsequently, the participants responded to the sociodemographic questionnaire, the Persistence Scale, the Passion Scale, and the Occupational Self-efficacy Scale–Short form taking approximately 15 min.
Data analysis
The content validity coefficient (CVC) was used to verify evidence of validity based on the content of the instrument, considering values above 0.70 (Hernandez-Nieto, 2002) regarding language clarity, practical relevance, and theoretical pertinence. In relation to the theoretical dimensions, the Kappa coefficient, considering values above 0.60, was used to evaluate the level of agreement among the experts (Fleiss, 1981).
The analyses directed toward evidence of validity based on internal structure, reliability and the relationship with an external variable were performed using the statistical software Factor (Lorenzo-Seva & Ferrando, 2006), Mplus 7.3 (Muthén & Muthén, 2017), and Jamovi (2020). It should be noted that the participants were randomly divided into two groups. Sample 1 (n = 150) was used to verify the factor structure through EFA, while sample 2 (n = 250) was selected for the CFA. In spite of the extensive and inconclusive debate over sample size in factor analysis, both samples could be considered adequate as they fit most of the classical rules-of-thumb suggested in the literature, as subject to item ratios of 10:1 and 10–20:1, or number of observation greater than 100 (Hair et al., 2010). However, it was decided to have a greater number of observations to carry out the CFA, which, as it is a restrictive model, allows the estimation of the loads of the items that represent the factor, forcing the factor loads of the items that were not designed to represent them to 0, which can make the model too restrictive to the point of harming the fit indicators (Marsh et al., 2020; Morin et al., 2020).
Exploratory factor analysis (EFA) was performed using the statistical software Factor, with a parallel analysis method, as an indicator for the retention of the factors, polychoric correlation, diagonally weighted least squares (DWLS) estimation, and Promin oblique rotation (Hongyu, 2018; Lim & Jahng, 2019). To verify the reliability level of the internal consistency, Cronbach’s alpha and McDonald’s omega were used (George & Mallery, 2002). According to the literature, indices greater than 0.70 are considered good indicators of reliability (Tabachnick & Fidell, 2019).
The CFA was performed using the Mplus software with the weighted least squares mean and variances adjusted (WLSMV) estimation method. The fit of the measurement model to the data was evaluated using the indices recommended by Muthén and Muthén (2017), namely, χ2/df, RMSEA, CFI, and TLI. The reference values commonly used in the specialized literature were used as adequacy parameters: χ2/df < 5, RMSEA < 0.08, CFI and TLI > 0.90 (Tabachnick & Fidell, 2019). Finally, the RFPS scale’s generalizability was assessed with participants from both samples. More precisely, using invariance testing we assessed whether the RFPS was free of measurement bias across gender and work conditions (i.e., presential and remote). Thus, fit indices are estimated for the different invariance models (configural, metric, and scalar). The configural model is evaluated with the same parameters as the AFC, the adequacy of these indices indicates the equivalence of the factor structure across groups. In order to evaluate the subsequent models, it is necessary to examine the extent to which the invariance constraint between the groups contributes to variability in the fit indices, being considered indicators of non-invariance ΔCFI ≥ − 0.01, ΔRMSEA ≥ 0.015, ΔSRMR ≥ 0.01 (or 0.03 in metric invariance) (Chen, 2007; Cheung and Rensvold, 2002).
To verify the validity based on the relationship with other variables, Pearson’s coefficient (r) was calculated, using the Jamovi software, to assess the degree of correlation between the constructs and their directions (whether positive or negative). Significance levels of p < 0.05 were considered. The magnitudes of the correlations were interpreted according to Cohen’s classification (1988): from − 0.09 to 0.09 null; from 0.10 to 0.29 small; from 0.30 to 0.49 medium and from 0.50 to 1.0 large. Finally, to evaluate the mediation model, structural equation modeling was carried out, using multiple regressions between the latent variables estimated from the items of the respective instruments. These analyses were also carried out using the Mplus 7.3 software and the goodness of fit of the data to the theoretical model was verified through the fit indices described in the CFA. To assess the significance of the total, direct, and indirect effects, a significance level of p < 0.05 was adopted, with the 95% confidence interval (CI) estimated through a bootstrap procedure, a method that allows for more robust estimates for the confidence limits of these effects, as well as for estimates of the standard errors (SE) associated with these statistics. Using the a priori sample size calculator for Structural Equation Models (Soper, 2022), that considering the number of latent (5) and observed variables included in the mediational model presented in this study (26), a power analysis of 80%, probability level of 0.05, and the effect size of 0.20 recommended a minimum sample of 376 participants (Westland, 2010), suggesting the adequacy of the available sample.