**3. Method**

#### *3.1. Research Framework*

Figure 1 shows the conceptual model of this study. First, it posits that prosocial motivation has direct and negative effects on financial satisfaction (H1a) and health satisfaction (H2a). Second, it suggests that prosocial motivation is related to exit intentions via financial satisfaction (H1b) and health satisfaction (H2b). The first path (H1b) predicts that prosocial motivation attenuates financial satisfaction and this, in turn, reinforces exit intention; the second path (H2b) predicts that prosocial motivation diminishes health satisfaction and this, in turn, escalates exit intention. Third, the relationship between prosocial motivation and financial satisfaction is stronger for male entrepreneurs than for female entrepreneurs (H3a), and the relationship between prosocial motivation and health satisfaction is stronger for male entrepreneurs than for female entrepreneurs (H3b).

**Figure 1.** Conceptual model.

#### *3.2. Sample and Procedure*

Via two large-scale colloquiums for entrepreneurs, organized by the All-China Federation of Industry and Commerce (ACFIC) in July and August of 2021, we identified the social entrepreneurs attending the colloquium and surveyed them for this study. The ACFIC is China's largest semi-official organization, consisting of business owners in diverse industries.

Considering that the respondents of this study were from China and the instruments used in the questionnaire were originally developed in English by prior researchers, we used the approach suggested by Brislin [114] for translating them into Chinese. After the translation was completed, the questionnaire was sent to experts in the field of social enterprise/entrepreneurship for their review. Afterward, a pilot test (on a sample of 100 respondents) was conducted. The Cronbach's alpha value was over 0.70, indicating the acceptable reliability suggested by Nunnally [115].

Data were gathered during the colloquiums in July (location: Jinan city, Shandong province, China) and August (location: Qingdao city, Shandong province, China), 2021. An invitation (on paper), including a QR code linking to the online questionnaire, was sent to the entrepreneurs (founders or CEOs) participating in the colloquiums.

A total of 450 founders or CEOs accepted our invitation and the response rate exceeded 80%, which is similar to the response rate of prior research [116]. After removing unusable data with missing or problematic values, the sample size was 317 (172 males, 145 females). Table 1 shows an overview of the sample demographics.


**Table 1.** Sample demographics.

For PLS-SEM analyses, Barclay [117] suggested the minimum sample size should be at least 10 times the maximum number of structural paths directed to a construct. The construct with the most paths in our model was the exit intention variable, which had only two paths. Thus, a minimum sample size of 20 was required to validate our model and this study's sample size (317) was highly sufficient.

Social entrepreneurs were identified with the question below employed by the Global Entrepreneurship Monitor (GEM):

"Are you, alone or with others, currently trying to start or currently owning and managing any kind of activity, organization or initiative that has a particularly social, environmental or community objective? This might include providing services or training to socially deprived or disabled persons, using profits for socially oriented purposes, organizing self-help groups for community action, etc."

Respondents choosing "no" were identified as conventional or regular entrepreneurs and excluded from this research; while those choosing "yes" were defined as social entrepreneurs and included in this research [118]. This method has also been deployed by prior studies of social entrepreneurs [35,119].

#### *3.3. Variables and Measurements*

Dependent variable: Exit intention was measured with three items developed by Pollack, Vanepps, and Hayes [52]. The items were rated on a Likert 7-point scale ranging from 1 = strongly disagree to 7 = strongly agree. The Cronbach's alpha for this scale was 0.927.

Independent variable: Prosocial motivation was measured with four items developed by Adam and Grant [27,120]. The items were rated on a 7-point scale ranging from 1 (strongly disagree) to 7 (strongly agree). The Cronbach's alpha for this scale was 0.850.

Mediating variable: Financial satisfaction was measured using the one-item measure developed by Fors, Johansson Sevä, and Gärling [70]. Participants indicated their satisfaction with their "private financial situation" on a 7-point scale ranging from 1 = extremely dissatisfied to 7 = extremely satisfied.

Mediating variable: Health satisfaction was measured by requesting the respondents to report the current state of their health [86,121] on a 7-point scale ranging from 1 = completely dissatisfied to 7 = completely satisfied.

Moderating variable: Gender. Male respondents were coded as "1" and female respondents were coded as "2".

A summary of the operational definitions is presented in Table 2. Moreover, the English version of the items has been appended (Appendix A).

**Table 2.** Operational definitions.


#### **4. Data Analysis**

To test our hypotheses, we employed consistent bootstrapped partial least square structural equation modeling (PLS-SEM) using the software SmartPLS (Version 3.3.3) [122]. Research suggests that PLS-SEM is increasingly being deployed in entrepreneurship research [123], and it is considered suitable for analyzing models with complex paths [123]. It is not limited by stringent assumptions (e.g., the multivariate normality) and sample size requirements [123].

Specifically, there were two reasons to use PLS-SEM for data analyses: first, PLS-SEM has been found to be effective in testing complex models, allowing simultaneous estimations of multiple causal relationships between variables [124], such as the ones in this study. Second, PLS-SEM is suitable for the exploratory analyses [122], such as the one in this study.

The analysis was conducted through two stages [117]. (1) The analysis of the outer model tested the reliability and validity of all latent construct measurements. (2) The analysis of the inner model assessed the relationships among the latent constructs for hypothesis testing. This sequence was to ensure that the measurement scales were valid and reliable.

#### *4.1. Outer Model Analysis*

The outer model's validity was evaluated by testing the reliability of each construct, the internal consistency of measures, and the convergen<sup>t</sup> and discriminant validities of each construct.

First, the reliability of constructs was evaluated by examining the factor loadings of each indicator. As Table 3 shows, all factor loadings (range: 0.769 to 0.962) reached the threshold value suggested by Hair et al. [125] of 0.70, implying adequate reliability.

Second, the internal consistency of the measures was examined by computing composite reliability (CR) values (Table 3). The composite reliability values were 0.899 (prosocial motivation) and 0.954 (exit intention), above the acceptable threshold value (0.80), as suggested by Fornell and Larcker [126].

Third, convergen<sup>t</sup> validity was examined by computing the average variance extracted (AVE) values. Table 3 shows the AVEs were 0.690 (prosocial motivation) and 0.873 (exit intention) above the acceptable threshold (0.50) [126], indicating sufficient convergen<sup>t</sup> validity.


**Table 3.** Reliability and AVE of the measurement model (outer model).

Note 1: PM = prosocial motivation; EI = exit intention. Note 2: Financial satisfaction is a single-item construct.Note 3: health satisfaction is a single-item construct.

Fourth, discriminant validity was tested by comparing the cross-loadings and factor loadings for each indicator (See Table 4), and the heterotrait-monotrait (HTMT) ratio of correlations (See Table 5). As shown in Table 4, the factor loading of each scale item for its assigned latent construct was higher than its loading on any other construct [122], suggesting good discriminant validity. Moreover, as shown in Table 5, the HTMT ratios of the average correlations of the indicators across constructs were all below the threshold (0.90) [127], indicating that each construct was empirically distinct from other constructs in the model and the discriminant validity was sufficient.



Note 1: PM = prosocial motivation; FS = financial satisfaction; HS = health satisfaction; EI = exit intention. Note 2: the grey cells are the factor loadings of scale items for each construct.



Note 1: PM = prosocial motivation; FS = financial satisfaction; HS = health satisfaction; EI = exit intention.

#### *4.2. Inner Model Analysis*

The inner model was assessed by computing *R*2, effect size (*f* 2), *Q*2, and path coefficients. The *R*<sup>2</sup> value of endogenous constructs is viewed as the primary criteria for assessing the quality of structural models [128]. We chose to follow the guidelines suggested by Chin [129]; the endogenous latent variables are considered reliable if their *R*<sup>2</sup> values are greater than 0.10 [130]. Meanwhile, for exploratory studies in social sciences, the *R*<sup>2</sup> value lower than 0.10 is also accepted [130]. Thus, as shown in Figure 2, the *R*<sup>2</sup> values indicate the significant explanatory power of the model.

**Figure 2.** Path coefficients and *R*<sup>2</sup> of the inner model.

Cohen's *f* 2 was used to evaluate the contribution of an exogenous variable in multiple regression models. Cohen's guidelines sugges<sup>t</sup> the following criteria for evaluating *f* 2 values: weak = 0.02; medium = 0.15, and large = 0.35 [131]. The *f* 2 value for H1a: PM → FS was 0.218, indicating a medium-level contribution of prosocial motivation to predicting financial satisfaction. In contrast, for H2a: PM → HS, *f* 2 was 0.002, indicating a negligible contribution of prosocial motivation to predicting health satisfaction.

As part of checking the predictive relevance, the *Q*2 values were also computed. The *Q*2 values for financial satisfaction, health satisfaction, and exit intention were 0.126, 0.013, and 0.341, respectively. Given that all of them were greater than zero, the explanatory constructs had adequate predictive relevance for their indicators [132].

Goodness of Fit (GoF) (0 < GoF <1) is another indicator of a PLS-SEM model's quality [133]. The GOF is calculated as:

$$\text{GoF} = \sqrt{\overline{\overline{\text{commonality} \times \overline{R^2}}}} \times \overline{R^2} = 0.794$$

The GoF values of 0.10, 0.25, and 0.36 are defined as small, medium, and large effect sizes, respectively [134]. The GoF value for the proposed model was 0.79, indicating a large effect size. Based on the above findings, it can be concluded that the proposed model has a good overall fit.

Next, we examined the structural relationships in the proposed model. Figure 2 reports the results of the algorithm and bootstrapping tests (based on 5000 samples), including the path coefficients (β), *t*-values, and retention or rejection of each hypothesis. Purvis et al. [135] suggested that bootstrapping is an effective procedure to evaluate the significance of each path coefficient. Figure 2 presents the bootstrapping validation outcomes. H1a (predicting prosocial motivation and negatively related to financial satisfaction) was supported (PM → FS: β = −0.368, *t*-value = 8.091, *p* < 0.001). H2a (predicting prosocial motivation and negatively related to health satisfaction) was not supported (PM → HS: β = −0.035, *t*-value = 0.632).

#### *4.3. Mediation Effects*

The Sobel test and variance accounted for (VAF) index were employed [136] to examine the mediation hypotheses (H1b and H2b). Per Sobel's test (See Table 6) [136], the mediation by financial satisfaction was significant (absolute Z value = 4.173, *p* < 0.01); whereas the mediation by health satisfaction was not significant (absolute Z value = 0.363, *p* > 0.05).

The method of variance accounted for (VAF) suggested by Hair Jr., Hult, Ringle, and Sarstedt [122] was used to determine the strength of the indirect effects (i.e., mediation effect) in relation to the total effect (i.e., direct effect plus indirect effect). The recommended VAF cutoff values for determining mediation effects are as follows: full mediation >80%, partial mediation ≤80%, and no mediation <20% [122]. Table 6 shows that financial

Supported

**Table 6.** Test of mediation effect. **Original Sample (O) Standard Error (STERR)** *t* **(|O/STERR|)** PM → FS −0.368 0.062 5.903 - FS → EI −0.398 0.067 5.900 - PM → HS 0.035 0.074 0.467 - HS → EI 0.032 0.056 0.577 - PM → EI 0.366 0.051 7.232 - PM → FS → EI - PM → HS → EI Total indirect effect Indirect effect 0.146 - 0.001 0.515 Sobel Z Test 4.173 - 0.363 - VAF 0.285 - 0.002 0.287

> -

 since the VAF was less than 20%.

Note 1: PM = prosocial motivation; JS = job satisfaction; WB = work burnout; WA = work anxiety; EI = exit intention. Note 2: number of bootstrap samples = 5000.

 NO

motivation–exit

 the mediation effects of

 H2b, concerning  intention relation,

#### *4.4. Moderation Effects*

 YES

satisfaction

supporting

health  was a partial mediator in the prosocial

satisfaction,

 H1b. However, hypothesis

 was not supported

 hypothesis

The multiple group analysis procedure (PLS-MGA) in SmartPLS (Version 3.3.3) was used to examine if the path coefficients [137] for males and females (1 = male, and 2 = female) differed significantly. PLS-MGA was conducted with a bootstrapped sample of 5000 cases. This analysis allowed us to see which path was distinct, how different the paths were, and whether there was a difference in the path direction. The results are presented in Table 7.

The results of the PLS-MGA indicate that the path between prosocial motivation and financial satisfaction was significantly stronger for males than for females, with a coefficient difference of 0.234 (*p* = 0.003). Therefore, H3a was supported. However, there was no statistically significant difference between males and females in the path coefficients between prosocial motivation and health satisfaction. Accordingly, H3b was not supported.


**Table 7.** Results of the multi-group analysis.

Note 1: PM = prosocial motivation; FS = financial satisfaction; HS = health satisfaction; EI = exit intention. Note 2: β = path coefficient; CI = 95% Confidence interval. Note 3: *f* 2 = size effect: 0.02 < *f* 2 < 0.15 (small effect size); 0.15 < *f* 2 < 0.35 (medium effect size); *f* 2 > 0.35 (large effect size).
