Dutton et al. (2014) argued that compassion is generative, and small increases in an individual's compassionate behavior toward others has great benefits for their own psychological and physiological wellbeing (Gilbert, 2019; Di Bello et al., 2020; Quaglia et al., 2021). These benefits have been widely researched, including increased positive and decreased negative affect, better relationships with others, decreased feelings of depression, anxiety, and stress, and improved coping mechanisms (Galante et al., 2014; Leaviss and Uttley, 2015). Interventions focused on increasing compassionate responding in organizations have shown similar benefits for those who receive compassion.

When compassionate responding is modeled by managers and encouraged in organizations, it has been shown to improve employee emotional and psychological wellbeing (Worline and Dutton, 2017). This includes reducing employees' anxiety and increasing positive moods (Chu, 2016), improving social wellbeing by improving prosocial behaviors (Runyan et al., 2019), improving employee relationships, with more cooperation, trust, and higher-quality communication and interactions (Barnes et al., 2007; Dutton et al., 2007), and reducing job burnout (Eldor, 2017). Fredrickson et al. (2008) concluded that compassion training improved employees' wellbeing at an emotional, psychological, and social level by improving their positive emotions, sense of meaning and purpose, and social interactions with others. Fredrickson (2001) broaden-and-build theory provides the best explanation for the positive relationship between compassion and wellbeing. The theory argues that positive emotions such as joy, love, and compassion lead to a broader and more flexible approach to life, which, over time, allows people to develop longer-lasting and more functional personal and social resources, which leads to greater wellbeing in the future (Kiken and Fredrickson, 2017).

Most of these studies have focused on how employees can benefit from the compassion of their managers (Fredrickson et al., 2008; Paakkanen et al., 2020). Few studies have examined whether managers benefit from giving and receiving compassion in the workplace. It is an important area of study for several reasons. Firstly, many managers find their leadership roles challenging and stressful (Lanaj et al., 2021; Morterson and Gardner, 2022), and demonstrating that increasing their compassion improves their own wellbeing would be an advantage. Secondly, since managers are in a higher power position in organizations, they may be less compassionate toward subordinates (Van Kleef et al., 2009). They may, therefore, be more in need of compassion training than other groups of employees. Thirdly, training in mindfulness and compassion has been shown to increase ethical behavior (Ozawa-de Silva et al., 2012; Kalafatoglu and Turgut, 2017), something that has been noted as lacking in many managers and business environments (Lu et al., 2018) and that is also especially important for wellbeing and prosperity (Leah, 2017). Compassion transcends the usual boundaries of power and hierarchy that exist in organizations and may present the key to transforming leader and manager behavior. Managers greatly influence their employees' behavior, and their compassionate responses lead to more compassionate behavior being displayed throughout the organization (Dutton et al., 2014). Finally, compassion can be trained, and increases in compassionate responding have shown significant benefits in a short space of time (Jazaieri et al., 2012).

The studies described above suggest that increasing compassionate behavior in organizations may be vital for improving the flourishing of managers. Flourishing refers to a state in which individuals experience all the different facets of their lives to be good, including their work (Johnson and Van der Weele, 2022). Van der Weele et al. (2019) conceptualized flourishing in terms of six dimensions: happiness, mental and physical health, meaning and purpose, financial security, close social relationships, and character. Research shows that flourishing employees do better in their jobs, are less likely to leave, tend to be more engaged in their work, display higher levels of organizational citizenship behavior, take fewer days of sick leave, and have more positive relationships with colleagues (Bono et al., 2012; Rothmann, 2013; Colbert et al., 2016; Redelinghuys et al., 2019). Previous research has shown that there is a significant relationship between self-compassion and flourishing (Akin and Akin, 2015; Fong and Loi, 2016; Verma and Tiwari, 2017; Ferguson et al., 2022), as well as between prosocial behaviors and flourishing in employees (Butt et al., 2020; Dutton et al., 2020; Kaabomeir et al., 2021), while a recent study found a link between receiving compassion and flourishing among students (Chan et al., 2022). Therefore, based on previous studies in other contexts, associations between compassion (for and from others) and flourishing may be expected.

An increase in compassionate behavior in organizations also has significant benefits at an organizational level (Lilius et al., 2008; Dutton et al., 2014). Studies have linked compassion at work with improved financial performance and employee and customer retention (Van Dierendonck, 2011; Moon et al., 2014; Worline and Dutton, 2017). Melwani et al. (2012) found that when managers displayed compassionate behavior, their employees perceived them as having greater leadership capabilities and intelligence, which meant they were more successful leaders. In a longitudinal study, Eldor (2017) showed that increasing the compassionate behavior of supervisors resulted in significant improvements in employee behavior, including their work engagement, organizational citizenship behavior, and relationships with clients. Barsade and O'Neill (2014) found that employees in compassionate organizations experienced increased job satisfaction, better teamwork, less absenteeism, and less emotional exhaustion. Specifically, in relation to this study, compassionate managers were more sensitive to others' needs, and when they displayed compassionate and caring behavior, their employees felt a stronger attachment and were more committed to the organization and less likely to leave (Grant et al., 2008; Lilius et al., 2008; Van Dierendonck, 2011).

Studies have shown that compassion plays a fundamental role in improving social interactions (Neff and Beretvas, 2013; Yarnell and Neff, 2013), and positive relationships at work have been shown to lead to greater organizational commitment (McCormick and Donohue, 2019; López-Ibort et al., 2020). The link between compassion and commitment is sometimes explained through social exchange theory (SET), which acknowledges the unwritten rules that govern human relations. When employees experience compassion from their colleagues and managers, it usually results in feelings of gratitude and obligation, so that the employee wants to reciprocate the behavior (Saks, 2006). In this way, employees who receive compassion feel more committed to, and have a stronger bond with, the organization (Kim et al., 2017). Consequently, they are prepared to go above and beyond their duties to meet their work objectives and those of the organization (Saks, 2006). A large body of research focuses on the antecedents and consequences of organizational commitment, which has been linked with improved employee work behavior, job performance and motivation (Meyer et al., 2002; Chen et al., 2006; Rafiei et al., 2014), improved organizational performance (Rashid et al., 2003; Fornes et al., 2008; Alfalla-Luque et al., 2015), and reductions in employee turnover (Joo and Park, 2010; Coetzee and Baker, 2015; Jano et al., 2019).

It is therefore important to establish whether the compassion managers feel for others or experience from others in the organization has an impact on their organizational commitment. However, studies focusing on the links between compassion and organizational commitment have tended to incorporate other aspects of organizational behavior, such as corporate social responsibility and ethics (Moon et al., 2016; Kumasey et al., 2017); or they focus on the effect of managers' compassionate behavior on the commitment of their subordinates (Boyatzis et al., 2006, 2013; Goleman et al., 2013). No studies were found to examine the links between managers' compassionate behavior and their organizational commitment, which may be another valuable benefit of cultivating compassionate organizations. Therefore, it was hypothesized in the current study that managers who feel more compassion toward their colleagues and experience more compassion from their colleagues, would feel a greater commitment to their organization.

Women have shown higher levels of compassion for others than men across several studies and using multiple different measures of compassion (Strauss et al., 2016; Sousa et al., 2017). Women have also reported higher levels of empathy, which is a core element of compassion (Schieman and Van Gundy, 2000; Toussaint and Webb, 2005; Marigoudar and Kamble, 2014). Interestingly, however, not all compassion studies have found gender differences, and there is some evidence that women may be motivated to report higher levels of compassion, because they believe it is expected of them (Klein and Hodges, 2001). However, these differences have not been explored in a work context, which may provide additional insights for management development and training (Dodson and Heng, 2021).

A construct can only be compared between different groups or across different times after first establishing the equivalence of the construct (Putnick and Bornstein, 2016). Therefore, if measurement invariance is not established, any cross-group differences can be attributed to the different attributes of the group, which will render the research findings invalid and incorrect (Yuan and Chan, 2016). The following tests for measurement invariance were thus conducted on the CS; configural invariance looks at whether the factorial structure is the same across the groups; metric invariance looks at whether the factor loadings are the same across the groups; scalar invariance tests whether the different groups have the same low and high thresholds; and finally, strict invariance looks at whether the variance across the groups is the same, i.e., whether the groups vary in the same way (Wang and Wang, 2020). The final two tests conducted focused specifically on the latent variables in the scale as follows; latent variance-covariance compares the variance and covariance across the different gender groups, and latent mean invariance compares the means of the latent variables across gender to check that they are not statistically different (Wang and Wang, 2020).

Pommier (2010) initially developed the Compassion Scale, using a similar underlying structure and theoretical basis as the Self-Compassion Scale (SCS; Neff, 2003) with a general compassion factor and three positive and three negative subscales. The SCS is the most widely used, valid and reliable measure of self-compassion. Pommier's (2010) study showed that the CS had good reliability with an overall Cronbach's alpha of 0.9 and good fit indices (CFI = 0.98; NNFI = 0.97; SRMR = 0.05 and RMSEA = 0.06). In 2020, Pommier et al. conducted further analyses of the CS and established convergent, discriminant and construct validity. However, the scale's factor structure was adapted to a general compassion factor and four subscales, with three positive subscales (mindfulness, kindness, and common humanity) and one negative subscale (indifference). Mindfulness concerns awareness of others' pain in a non-judgmental and accepting way. Kindness refers to being caring and concerned about other people's suffering and wishing to help them. Common humanity involves recognizing that everyone suffers and has challenges in life. As a result, we are all connected in that suffering, while indifference refers to uncompassionate responding or indifference toward another's suffering (Pommier et al., 2020). The adapted scale showed good reliability, with omega values varying from 0.78 to 0.90 across samples.

It was therefore deemed important to test the factorial validity of the CS in this study, using the method recommended by Pommier et al. (2020) in which confirmatory factor analysis (CFA) and bifactor CFA models should be compared with exploratory structural equation modeling (ESEM) and bifactor ESEM. Previous studies have established the presence of construct-relevant multidimensionality in the CS (Morin et al., 2016). This is important, because Morin (2023) points out that the conditional independence assumption in CFA holds that indicators of each factor in a multidimensional scale are related only to that specific factor. Consequently, possible associations of such indicators with other factors are considered sources of measurement error, resulting in inflated estimates of factor correlations. Exploratory structural equation modeling is thus more applicable when psychometric multidimensionality exists, i.e., when factor indicators reflect more than one thing, as is the case with the CS (Morin, 2023). More specifically, Pommier et al. (2020) argued that bifactor ESEM is in line with the original theoretical conceptualization of compassion toward others, as it accounts for multidimensionality by assessing the global factor as well as the specific subscales, allowing for the cross-loading of items which are constrained to be as close to zero as possible.

Following the procedure suggested by Morin et al. (2016) and Morin (2023), the factor structure of the CS could be explored through targeted CFA, bifactor CFA, ESEM, and bifactor ESEM analyses. Testing these four models corresponds with the procedures followed by Neff et al. (2019) and Pommier et al. (2020). The CS showed good psychometric properties in previous validation studies, where the bifactor ESEM showed superior fit statistics compared to the CFA (Pommier et al., 2020).

Based on the above discussion, various research gaps were identified. First, compassion for others is important for building flourishing institutions. People view and experience compassion differently, making it difficult to measure in diverse communities. More specifically, scientific information is lacking regarding the validity, reliability and measurement invariance of the CS in organizations in non-Western contexts (Sousa et al., 2017). According to Henrich et al. (2010), behavioral scientists need to extend their research beyond Western, Educated, Industrialized, Rich and Democratic (WEIRD) societies to understand human psychology. Measuring instruments of compassion might show different psychometric properties in non-WEIRD contexts, which may indicate that such measures and the theory on which it is built, are not applicable across all contexts. Second, while compassion for and from others is essential for managers, no studies directly examined the link between compassion felt for and received from others and the flourishing of managers. Finally, more research is needed to examine whether there are differences in compassion between men and women in management positions. This would be useful information when designing management development programs.

This study aimed to assess the associations between compassion for others, experienced compassion, secure flourishing, and organizational commitment of managers in South Africa. Given that Pommier's Compassion Scale was not previously validated in South Africa, the study investigated its factorial validity and measurement invariance across genders.

The following hypotheses are formulated:

Hypothesis 1: Managers' compassion for others at work is associated with their secure flourishing (Hypothesis 1a) and organizational commitment (Hypothesis 1b).

Hypothesis 2: Managers' experienced compassion from others at work is associated with their secure flourishing (Hypothesis 2a), and their organizational commitment (Hypothesis 2b).

Hypothesis 3: Female managers show significantly more compassion for others than male managers.

The study participants were 390 managers registered for post-graduate management qualifications at business schools across South Africa. The sample included 207 (53.1%) males and 183 (46.9%) females. The average age of participants was 40 years, and the average number of years of total work experience was 17 years. Participants had an average of 9 years' experience in a management role, with 35% (N = 135) being less experienced (<5 years), and 31% (N = 122) being experienced (more than 11 years). Most participants (63.9%) worked in the services sector.

The Compassion Scale (CS; Pommier, 2010) was used to measure compassion for others. The scale has 16 items, with four underlying subscales: kindness (four items, e.g.: “I like to be there for others in times of difficulty”); common humanity (four items, e.g.: “Everyone feels down sometimes, it is part of being human”); mindfulness (four items, e.g.: “I pay careful attention when other people talk to me”); and indifference (four items, e.g.: “Sometimes when people talk about their problems, I feel like I do not care”). Items are rated on a five-point scale, varying from 1 (almost never) to 5 (almost always). Anchors were not provided for ratings 2, 3, and 4 in line with the approach adopted by Pommier et al. (2020). Studies in WEIRD contexts have shown good reliability, validity, and model fit for the CS, but the scale has not previously been validated in South African samples (Neff and Germer, 2013; Pommier et al., 2020). Participants were asked to rate each item in relation to how they normally behaved toward others at work, rather than thinking about their behavior generally, so that their compassion in the workplace was specified. Pommier et al. (2020) reported alpha and omega reliability coefficients varying from 0.77 to 0.90 for the general compassion scale of the CS. Furthermore, reliabilities higher than 0.70 were found for the subscales of the CS.

The Secure Flourishing Index (SF; Van der Weele et al., 2019) was used to measure flourishing. The scale has 12 items, which include two items taken from each of the six aspects of flourishing included in Van der Weele et al.'s (2019) description: happiness (e.g., “In general, how happy or unhappy do you feel?”); mental and physical health (e.g., “In general, how would you rate your physical health?”); meaning and purpose (e.g., “I understand my purpose in life”); character (e.g., “I always act to promote good in all circumstances, even in difficult and challenging situations”); close social relationships (e.g., “I am content with my friendships and relationships”); and financial stability (e.g., “How often do you worry about being able to meet normal monthly living expenses?”). Each question or statement is scored from 0 (lowest possible) to 10 (highest possible), and anchors were not provided for ratings 1 to 9 to mimic the scale of Van der Weele et al. (2019). The scale has been validated in workplace settings across several different countries and a recent study by Weziak-Bialowolska et al. (2019) showed that the scale was valid and reliable (α = 0.89).

The Organizational Commitment Scale (OCS; Saks, 2006) was used to measure organizational commitment. The scale has six items and measures the two main aspects of the construct, namely, attachment (e.g., “I feel personally attached to my work organization”) and pride (e.g., “I feel proud to be an employee of this organization”). Responses are measured on a Likert scale, ranging from 1 (strongly disagree) to 7 (strongly agree). Confirmatory factor analysis supported the construct validity of the OCS, while an alpha coefficient of 0.85 was found. The scale had good validity and reliability and has been used previously in South African studies (Swart and Rothmann, 2012; Ahuja and Gupta, 2019).

The Experienced Compassion Scale (ECS; Lilius et al., 2008) was used to measure the compassion that managers experienced in their organizations. The scale has three items (e.g., “How frequently do you experience compassion from your supervisor?”) and uses a scale varying from 1 (never) to 5 (nearly all the time). The scale showed good reliability and validity, with a Cronbach's alpha of.83 and composite reliability of 0.84 in previous studies (Lilius et al., 2008).

Ethics clearance was obtained from the EMS-REC (Economic and Management Sciences Research Ethics Committee) at the North-West University in South Africa (NWU-01309-21-A4). A purposive sampling technique was used to recruit students registered for postgraduate management qualifications at all accredited South African business schools. The business school postgraduate qualifications require work and management experience, and their students are prospective, new, or experienced managers and supervisors. In total, nine public university business schools and two private business schools participated, and further ethics clearances were obtained from each institution, where necessary. The link to the online survey was shared via the schools' communication platforms, and all students were invited to participate.

IBM SPSS 27 for Windows (IBM Corp, 2021) and Mplus 8.10 (Muthén and Muthén, 1998-2023) were used to conduct the analysis. Missing values were dealt with using the full information likelihood method (FIML; Enders, 2023). The weighted least squares mean and variance estimator (WLSMV), appropriate for analyzing categorical variables was used to investigate the CS responses (Neff et al., 2019). A robust maximum likelihood estimator (MLR) was used to analyze the SF items, which uses a 10-point scale (Wang and Wang, 2020).

In line with previous studies, the factor structure of the CS was explored through targeted CFA, bifactor CFA, ESEM, and bifactor ESEM analyses (Neff et al., 2019; Pommier et al., 2020). A code generator (developed by De Beer and Van Zyl, 2019) was used to specify the ESEM and bifactor ESEM models in Mplus 8.10. Target rotation was applied, whereby a confirmatory approach to ESEM was taken. The fit of the four models was assessed using the comparative fit index (CFI), Tucker-Lewis index (TLI), the standardized root mean residual (SRMR) and the root mean square error of approximation (RMSEA). The model fit criteria are as follows (Morin, 2023; West et al., 2023): a non-significant (p > 0.05) chi-square (χ²) value; 0.90 and 0.95 (and higher) for CFI and TLI; 0.06 (and lower) for RMSEA, and 0.08 (and lower) for SRMR. According to West et al. (2023), the cutoff values for CFI, TLI, RMSEA, and SRMR can serve as rough guidelines for the model's overall fit.

The reliability of all scales used in the study was assessed using the McDonald's omega (Rodriguez et al., 2016). The reliability indices were computed using the bifactor indices calculator (Dueber, 2017). According to Hayes and Coutts (2020), a value of 0.70 can be used as a cutoff value for scale reliability, while higher values are preferable. McDonald's Omega was used to assess the internal consistency of the bifactor ESEM model (Perreira et al., 2018; Morin, 2023; ω > 0.50).

Compared to unidimensional models, bifactor CFA and ESEM models result in lower omega values by design because variance is shared between general and specific factors. In these cases, conventional thresholds like 0.70 or 0.80 are “unrealistic, inappropriate and not suited to bifactor models” (Perreira et al., 2018, p.70; Morin, 2023), therefore an omega value of 0.50 should rather be used as a minimum cutoff (Perreira et al., 2018). After establishing the fit of the model, the quality of the measurement should also be assessed. van Zyl and ten Klooster (2022) proposed using factor loadings, ECV, and reliabilities, and advised that researchers should use their theoretical knowledge to decide which of these to apply. These criteria can range from inspecting the standardized factor loadings (e.g., λ > 0.35), the item uniqueness (e.g., residual error variances >0.10), tolerance levels for cross-loadings, and the overall R² per item. They further advised that there should be some flexibility when applying the chosen criteria, keeping in mind the study context and what the values might mean (van Zyl and ten Klooster, 2022).

Furthermore, tests for measurement invariance were conducted to assess the validity of the CS for different genders. A code generator (developed by De Beer and Morin, 2022) was used to specify models to test the configural, metric, scalar, strict invariance, latent variance-covariance and latent mean invariance of the CS in Mplus 8.10. Chen (2007) advised that a change in CFI and TLI of ≤ .01, a change in RMSEA of ≤ .015 and a change in SRMR of ≤ .01 between two models would support their equivalence (Chen, 2007; Morin, 2023).

Factor scores were used to compute Pearson correlations between the variables that were included in the structural model. The structural model was specified based on the best-fitting measurement model of the CS. In addition, gender (male vs. female), experienced compassion (three items), secure flourishing (12 items), and organizational commitment were modeled in the structural model (see Figure 2).

Next, the structural model of compassion, secure flourishing, and organizational commitment was tested. The following variables were included in the analyses: the general compassion factor and the four specific compassion factors (mindfulness, kindness, indifference, and common humanity) resulting from part 1 of this study, gender, experienced compassion, secure flourishing, and organizational commitment. Table 5 shows the reliabilities and Pearson correlation coefficients between the compassion, experienced compassion, secure flourishing, and organizational commitment scales.

Table 6 shows that the omega scores for the SFI, OCS, and ECS indicated good reliability, as all were above 0.70 (Hayes and Coutts, 2020). Correlations revealed that compassion for others (GF) was not significantly correlated with the four subscales, but was significantly correlated with experienced compassion (r = 0.13) and with secure flourishing (r = 0.36). The mindfulness subscale (M) showed small but significant correlations with the three other subscales (r_K = −0.24; r_I = 0.11; r_CH = −0.13), but not with any of the other constructs. The kindness and indifference subscales showed significant negative associations with organizational commitment and secure flourishing, which was unexpected in the case of the kindness subscale. Common humanity was the only subscale to be significantly correlated to all three remaining variables, including experienced compassion (r = 0.13), organizational commitment (r = 0.19) and secure flourishing (r = 0.53). The last three constructs, namely experienced compassion, organizational commitment and secure flourishing were all moderately to strongly correlated with one another (r = 0.43 to 0.65).

The model tested for significant relationships between the following: compassion for others (general factor and four subscales) and secure flourishing; experienced compassion and secure flourishing; compassion for others (general factor and four subscales) and organizational commitment; and experienced compassion and organizational commitment. The fit statistics of the structural model were as follows: χ² = 853.55, df = 401, p < 0.001, CFI = 0.96, TLI = 0.95, RMSEA = 0.05, p = 0.105 [0.05,0.06], SRMR = 0.06). The fit indices of the structural model pointed to a good model fit, although the chi-square value was statistically significant. Figure 2 shows the structural model for testing the remaining study hypotheses

The model highlights the relationships between the study constructs. In line with Hypothesis 1a, managers' general compassion for others had a statistically significant and positive effect on their secure flourishing (p < 0.001). The mindfulness and common humanity subscales also showed statistically significant effects on secure flourishing (p = 0.001). Unexpectedly, the kindness subscale showed a weak non-significant and negative effect on secure flourishing. Consistent with Hypothesis 2a, experienced compassion was statistically significantly associated with secure flourishing (p < 0.001). These variables explained 50.2% of the variance in secure flourishing.

Managers' compassion for others had no statistically significant connection with their organizational commitment (p = 0.433), showing that Hypothesis 1b was not supported. However, the indifference and common humanity subscales had a statistically significant (but small) effect on organizational commitment. On the other hand, Hypothesis 2b was supported, because experienced compassion strongly affected organizational commitment (p < 0.001). Together, experienced compassion and gender explained 43.9% of the variance in organizational commitment. Finally, male managers showed significantly higher levels of organizational commitment than female managers (p < 0.001).