Method

Results

Regarding the dimensionality results, the KMO statistic yielded an adequate value of .85. Also, Bartlett's test of sphericity resulted significant (p = .000), confirming that the correlation matrix is significantly different from the identity matrix. The recommended number of dimensions according to parallel analysis, scree plot analysis and eigenvalues rule was 9. In addition, the EFA conducted with the first subsample (n₁ = 349) produced a pattern matrix with a nine-factor solution with three to four items on each subscale, as ^{Donaldson & Donaldson (2021b)} obtained. In this sense, all items had loadings between .88 and .52, except item 14, which had a loading of .40. However, all cross loadings were less than .20. To summarize, the performed EFA revealed a nine-factor structure that explained the 63% of the variance.

Regarding the assess of validity evidence based on internal structure, three structural equation models have been tested with a comparison of nested model fit (Figure 1). None of the models showed Heywood cases, as there were no negative error variances and no R² statistic greater than 1.

Figure 1. Best-Fit Models for the PF-W Scale: a) Nine-Factor Model. b) Higher Order Model. 

The first model tested the solution of nine correlated factors. This model suggests that PF-W is composed of nine multidimensional factors that are related to each other but are not causally determined by a higher order construct. The second model tested the bifactor structure, in order to assess the essential unidimensionality. Finally, the third model corresponds to ESEM structure. Table 1 shows the results of the fit indices obtained for each model tested.

As can be seen, the three models showed adequate fit indices. That said, the bifactor and ESEM structures obtained the best fit indices, with the most notable increase in the RMSEA, CFI and TLI values. Regarding the bifactor structure, given the considerations of ^{Rodriguez et al. (2016)} on the usefulness of the ECV, PUC and ωh in assess essential unidimensionality, these indices were calculated to test the adequacy of the bifactor model. In this case, ECV was .34, while PUC was .92 and ωh was .76. Even if the ECV is relatively modest, a high PUC indicates that the model will not be biased by specifying a bifactor structure (^{Rodriguez et al., 2016}). Therefore, the calculated indices support the use of an overall factor in the PF-W.

Then, given that the three models were nested, a comparison of the three CFA models was performed by the RMSA_D index, in order to know which model fits significantly better. Thus, the results (Table 1) showed that the nine-factor structure fitted the data worse than the ESEM model (RMSEA_D = .081); however, the fit of bifactor model was very similar to the fit of ESEM model (RMSEA_D = .043), even though the χ² difference test was significant (χ²diff = 248.234, p = .000). This suggests that bifactor and ESEM models have a similar fit and both are better than the nine-factor model at characterizing the structure of the PF-W scale.

Finally, a multigroup measurement invariance analysis (MGMI) was performed. For the present study, the educational level variable was explored, which included five categories: Basic education, secondary education, professional education, university degree and postgraduate education. Thus, a multigroup invariance analysis was performed to verify that the educational level of the participants did not influence the way in which the elements of the PF-W were interpreted. In order to achieve enough statistical power for the analysis, the educational level variable was recoded to obtain three categories: Basic or secondary education, professional education, and university education.

The fit indices of the bifactor model calculated for the three groups separately are as follows: Basic education (χ²₍₃₄₈₎ = 433.066, p = .000; χ²/df = 1.244; RMSEA = .037, 90% CI [.024, .048]; SRMR = .052; CFI = .972; TLI = .967), professional education (χ²₍₃₄₈₎ = 539.356, p = .000; χ²/df = 1.560; RMSEA = .049, 90% CI [.041, .057]; SRMR = .057; CFI = .950; TLI = .941) and university education (χ²₍₃₄₈₎ = 526.324, p = .000; χ²/df = 1.512; RMSEA = .042, 90% CI [.035, .049]; SRMR = .050; CFI = .963; TLI = .957). On the other hand, the fit indices of the ESEM model calculated for the three groups separately are as follows: Basic education (χ²₍₁₈₁₎ = 215.854, p = .000; χ²/df = 1.192; RMSEA = .036, 90% CI [.028, .044]; SRMR = .022; CFI = .986; TLI = .964), professional education (χ²₍₁₈₁₎ = 302.827, p = .000; χ²/df = 1.366; RMSEA = .054, 90% CI [.043, .063]; SRMR = .042; CFI = .968; TLI = .930) and university education (χ²₍₁₈₁₎ = 247.194, p = .000; χ²/df = 1.653; RMSEA = .036, 90% CI [.024, .046]; SRMR = .018; CFI = .986; TLI = .969).

To perform the multigroup measurement invariance analysis, the steps described by ^{Kline (2016)} were followed, calculating first the configural invariance, followed by the metric invariance and finally the scalar invariance. Such invariance analyses were performed between the bifactor and ESEM models using the reference group method (^{Little et al., 2006}). Thus, Table 2 collects the results of this analysis. As can be seen, the configural model of both bifactor and ESEM structures showed excellent fit indices. Similarly, with respect to metric invariance, the differences between the CFI, RMSEA and SRMR indices are smaller than the criterion established by ^{Chen (2007)} for considering that metric invariance is not achieved. In addition, the RMSEA_D value indicates that there are no differences between the metric and configural model in the bifactor structure (RMSEA_D = .011) and ESEM structure (RMSEA_D = .011). Therefore, the obtained results confirm the presence of metric invariance. Finally, regarding scalar invariance, it can be observed that the χ² difference test rejects this kind of invariance in the bifactor structure (χ²diff = 83.225, p = .000) and ESEM structure (χ²diff = 84.220, p = .000). However, due to the excessive sensitivity of the χ² difference test in invariance analysis (Kline, 2016), attention should be paid to the other indices. Thus, since the variations in the CFI, RMSEA and SRMS indices do not reach the limit established by Chen (2007), and the RMSEA_D values are lower than .08 for both the bifactor model (RMSEA_D = .038) and ESEM model (RMSEA_D = .039), the presence of scalar invariance in the bifactor and ESEM structures is also confirmed.

Regarding reliability analysis, as can be seen in Table 3, in general, the adapted PF-W scale has obtained acceptable and excellent internal consistency statistics, obtaining a value of .90 for the total scale. The dimensions of accomplishment and economic security showed the lowest loadings, but still sufficiently above .70.

To analyze discriminant validity, a Pearson correlation analysis was performed between the PF-W and its dimensions with life satisfaction, psychological capital, and job stress (Table 3). As expected, the PF-W scale is positively related to other measures of well-being, such as life satisfaction and psychological capital. Specifically, the results show a positive and significant relationship between the PF-W scale and life satisfaction, r(282) = .44, p < .01, and between the PF-W scale and psychological capital, r(282) = .64, p < .01. In this sense, it should be noted that the correlation indices obtained with respect to life satisfaction and psychological capital support the existence of discriminant validity, as these correlations are lower than .85 (^{Campbell & Fiske, 1959}). In other words, the correlations between the PF-W and life satisfaction (.44) and psychological capital (.64) are considerable, but not high enough to argue that they measure the same thing. In addition, the PF-W scale also obtained a negative and significant correlation with job stress, r(190) = -.49, p <.01.

Finally, to analyze criterion validity, first, correlations were calculated between the PF-W and the variables corresponding to the work outcomes assessed: JAWS, organizational OCB and TIS (Table 3). As can be seen, all measures of assessed work outcomes have high and significant relationships with the PF-W scale. Specifically, the highest correlations were obtained between PF-W and positive job-related affective well-being, r(226) = .66, p < .01, and between PF-W and turnover intentions, r(465) = -.60, p < .01. In turn, the organizational citizenship behavior also has a positive and significant relationship with the PF-W scale, r(190) = .50, p < .01. Finally, it should be noted that all the dimensions of the PF-W scale have significant correlations with the work outcomes variables, with the exception of engagement dimension, whose correlation with the job-related negative affective well-being variable was not significant, r(226) = .01, p > .01.

In order to assess the incremental and criterion validity of the PF-W scale, a series of hierarchical multiple regressions were performed to determine whether or not the four new dimensions added to the PERMA framework improved the prediction of work outcomes (Table 4). Prior to conducting each regression analysis, the corresponding assumptions were checked. First, linearity was observed using partial regression plots and a plot of studentized residuals against the predicted values. Likewise, there was homoscedasticity, assessed by visual inspection of a plot of studentized residuals versus unstandardized predicted values. The assumption of normality of the residuals was also met, assessed by a Q-Q plot. Finally, independence of residuals, assessed by a Durbin-Watson statistic, and multicollinearity, assessed by the variance inflation factor (VIF), are shown in the regression table.

First, the incremental validity of PF-W for predicting JAWS was evaluated. The first model that analyzes the influence of PF-W on JAWS (negative emotions). As can be seen, using only the five dimensions of the PERMA framework, the model is statistically significant, R² = .26, F_{(5, 218)} = 17.10, p = .000. However, adding the four new factors to the JAWS prediction (negative emotions) leads to a statistically significant increase in ΔR² of .03, ΔF_{(4, 214)} = 3.10, p = .017. Specifically, physical health (β = -.194, p = .002) acts as a negative predictor.

Finally, the last analysis on the incremental validity of PF-W was performed for predicting TIS. The data reveal that the model with the five dimensions of the PERMA framework is statistically significant, R² = .41, F_{(5, 459)} = 66.67, p = .000. Moreover, the addition of the four new factors to the TIS prediction represents a statistically significant increase in ΔR² of .03, ΔF_{(4, 455)} = 5.80, p = .000. Specifically, the economic security dimension (β = -.160, p = .000) is acting as a strong negative predictor.

On the other hand, the incremental validity of PF-W for predicting OCB was also analyzed. In this case, although the model with the five dimensions of the PERMA framework is statistically significantly, R² = .45, F_{(5, 184)} = 32.12, p = .000, the addition of the four new factors to the OCB prediction does not result in a statistically significant increase in ΔR², ΔF_{(4, 180)} = 0.65, p = .631. Only the dimensions of positive emotions (β = .401, p = .000) and relationships (β =.254, p = .000) exerted a significant influence on organizational citizenship behavior in both models.

Finally, to conclude the analysis of the predictive validity of the PF-W scale and to test its discriminant capacity with respect to life satisfaction and psychological capital, a final hierarchical multiple regression analysis was performed with TIS as the criterion variable (Table 5). As can be seen in the first model, life satisfaction and psychological capital are predictor variables of turnover intentions, R² = .23, F_{(2, 272)} = 41.54, p = .000. However, in the second model, in which life satisfaction, psychological capital and PF-W are used as predictor variables, there is a statistically significant increase in ΔR² of .10, ΔF_{(1, 271)} = 40.08, p = .000. Specifically, in the second model, PF-W becomes the main predictor variable (β = -.427, p = .000), while life satisfaction loses weight in the model (β = -.168, p = .003) and psychological capital does not act as a predictor variable (β = -.077, p = .244).

Discussion

The purpose of the present study was to validate the PF-W scale with a sample of Spanish workers. Specifically, to evaluate the internal structure of the PF-W scale, the nine-factor model determined by ^{Donaldson & Donaldson (2021b)} was tested together with the bifactor model and ESEM model. These both methods seek to overcome the limitations of CFA, explaining more sources of variability, and, therefore, being more realistic and less biased. The results have shown very positive fit indices in the three models analyzed, with those corresponding to the bifactor model and ESEM model being significantly higher than those corresponding to the nine-factor model. The results show that the addition of the four new factors to the PERMA model provide a better understanding of well-being and positive functioning in the work environment (^{Donaldson et al., 2022}).

More specifically, the results of the CFA show that all dimensions of the PF-W model added significant variance. Of the four new factors added, mindset obtained the highest factor loadings, followed by environment, physical health and economic security. Finally, the results of the multigroup measurement invariance analyses have shown that there are no differences in the interpretation of the scale according to the level of education. These results demonstrate measurement invariance across the three sociodemographic groups developed and support the PF-W scale as a useful measurement tool.

Regarding discriminant validity, positive, and moderate relationships were found between PF-W, life satisfaction and psychological capital, and a negative and moderate relationship between PF-W and job stress. These results are in line with that obtained by ^{Goodman et al. (2018)}, who suggested that life satisfaction and PERMA framework were defined by the same higher-order factor of well-being; likewise, other studies such as ^{Ho & Chan (2022)} found positive relationships between psychological capital and work-related affect, relationships in the work environment, meaningful work, and work achievement, which could be identified with some of the dimensions of the PF-W model.

Regarding the analysis of criterion validity, the PF-W scale has obtained significant medium-strong relationships with the work outcomes assessed in this study. The highest correlations were between the PF-W scale and JAWS and the PF-W and turnover intentions. Of the four new factors added to the PERMA framework, all of them have significant relationships with the JAWS scale and with TIS. Specifically, mindset dimension obtained the highest correlation with the JAWS scale (positive emotions) (r = .49) and with TIS (r = -.36). Considering the mindset dimension as a construct characterized by a growth mentality and a propensity to persevere in the face of setbacks, it is consistent that this factor has significant relationships with positive affective well-being related to work and with intentions to leave an organization (^{Duckworth et al., 2007}). For its part, the economic security dimension also obtained a medium and negative correlation with turnover intentions (r = -.35). In this case, economic security is largely determined by the salary that a person receives in his or her job, a factor that is crucial in determining the intentions to leave an organization (^{Belete, 2018}).

Continuing with the analysis of criterion validity, a key aspect of this section was the analysis of the differential role of the four new additional blocks to PERMA in the prediction of employee well-being and performance. First, regarding the prediction of well-being, the hierarchical multiple regression on the JAWS scale (positive and negative emotions) has shown a significant increase in R² of .03 in both cases. More specifically, the physical health dimension, together with positive emotions and relationships, are the significant predictors of negative affective well-being at work. On the other hand, the mindset dimension, together with positive emotions, engagement, and relationships, are the predictors of positive affective well-being at work. These results suggest that employees' perceptions of physical health and mindset play an important role in their experience of work-related affective well-being.

As for the prediction of performance, organizational citizenship behavior and intentions to quit were used as criterion variables. Regarding organizational citizenship behavior, hierarchical multiple regression analysis has shown that the increase in R² is not significant. In this case, the PF-W dimensions that are significant in predicting organizational citizenship behavior (positive emotions and relationships) belong to the PERMA framework, so there is apparently no influence of any of the four new factors added. The literature on organizational citizenship behavior indicates that the individual predictors of this construct are job satisfaction, organizational commitment, organizational justice, trust in leaders, and conscientiousness (^{LePine et al., 2002}). Therefore, it is consistent that the four new additional factors of the PF-W model (physical health, mindset, environment and economic security) do not exert a significant influence on organizational citizenship behavior, since the antecedent elements of this construct are more related to the dimensions of the PERMA model, such as positive emotions and relationships, factors that have been shown to exert a significant influence in the regression analysis performed.

Finally, to test the discriminant ability of the PF-W scale with respect to life satisfaction and psychological capital, a comparative analysis was conducted to assess whether the PF-W scale predicted a unique portion of variance in turnover intentions. The PF-W scale captured 10% of the variance in turnover intentions above and beyond life satisfaction and psychological capital. This is a remarkable finding and suggests that positive functioning may be a useful predictor of undesirable work outcomes.

In sum, at a theoretical level we consider that this study confirms the multidimensional framework on well-being in the workplace proposed by ^{Donaldson et al. (2020)}, in addition to providing a questionnaire with very good psychometric properties to be able to explain the results in the workplace beyond other well-validated scales.

This study is not without limitations. The first would be the use of self-reporting questionnaires. In this sense, other methods of measuring should be explored. As ^{Donaldson et al. (2020)} pointed, measuring a construct with two or more credible sources of information (like self and knowledgeable co-worker reports) would result in assessments with less error than those reported by a single source. On the other hand, it should be investigated how the different dimensions may work with different types of workers such as managers, middle managers, or staff workers.