Descriptive statistics
Table 1 showed the sociodemographic characteristics of the respondents. The mean age was 36.98 ± 9.84 years (minimum: 18 years, maximum: 73 years). 426 (31.09%) medical staffs were male and 944 (68.91%) were female. 42.77% of participants were physicians and 41.02% were nurses. The largest number of participants in the 40-49 age group, accounting for 34.45%, while the group over the age of 49 accounts for 11.31%. Most participants (77.30%) were married, 724 (52.58%) have a junior title and sixty-three percent have no night shift in their work.
Structural Equation Model Constructing and Fitting
The Hypothetical model was established, as shown in Figure 2. We have constructed four paths: (1) Path a: Path from independent variable to potential mediator variable, the path coefficient of path a represents the indirect effect of burnout to job satisfaction (Job satisfaction ¬ Burnout). (2) Path b: The path from potential mediator variable to dependent variable, the path coefficient of path b represents the indirect effect of job satisfaction to turnover intention (Turnover intention ¬ Job satisfaction). (3) Path c: the path from independent variable to dependent variable, the path coefficient of path c represents the total effect of burnout to turnover intention (Turnover intention ¬ Burnout). (4) Path c’: Under the influence of potential mediator variables, the path from the independent variable to the dependent variable, the path coefficient of path c’ represents the direct effect of burnout to turnover intention (Turnover intention’ ¬ Burnout’).
As shown in Table 2. From the results of the hypothetical model operation, we found that all fitting indices did not meet the fitting criteria, indicated that the hypothetical model was not ideal, so we revised the model. The model path was modified according to the amendment advice given by AMOS. We removed some items (A18, B1 and C4) and added lots of bidirectional arrows to make the model fitting better. The final fitting indices results were also shown in Table 2, and the revised standardized path coefficient map were displayed on Figure 3. After the adjustments, validity and reliability of the three scales remained acceptable. KMO measure , p value for batrtlett’s spherical test and Cronbach’s a for job satisfaction is 0.957, <0.01 and 0.976, respectively; For turnover intention is 0.857, <0.01 and 0.910, respectively; For burnout is 0.798, < 0.01 and 0.879, respectively.
As shown in Table 3, we estimated the significance of total effect, direct effect and indirect effect by bias-corrected approach. The results showed that the total effect was significant of independent variable (burnout) to dependent variable (turnover intention) (p<0.01), that was, the total effect of path c was statistically significant. The direct effects of path a (Job satisfaction Burnout) (p<0.01), path b (Turnover intention ¬ Job satisfaction) (p<0.01) and path c' (Turnover intention’ ¬ Burnout’) (p<0.01) were also significant. The results of indirect effect test again proved that path c' (Turnover intention’ ¬ Burnout’) was statistically significant. We concluded that this mediating effect was a partial mediating effect.
As shown in Table 4, standardized estimates and its standard errors were calculated by using bootstrap method. Standardized path coefficient of path a (Job Satisfaction ¬ Burnout) is -0.410 and its standard error (Sa) is 0.038. Standardized path coefficient of path b (Turnover intention ¬ Job satisfaction) is -0.180 and its standard error (Sb) is 0.028. Standardized path coefficient of path c’ (Turnover intention’ ¬ Burnout’) is 0.824 and its standard error (Sc’) is 0.029. Looking up the related tables, the standardized path coefficient of path c is 0.899 and its standard error (Sc) is 0.020. Soble-Z test is carried out according to the formula (see Formula 1 in the Supplmentary Files). Finally, z= 4.506. According to MacKinnon's critical value table, the result is p < 0.05, indicating that the mediation effect is significant.
Interpretation of the Revised Model
Model fit is acceptable if c2/df ≤4.0[37], GFI >0.90, AGFI >0.90, CFI >0.90, NFI >0.90, IFI >0.90[38], TLI >0.90 and RMSEA <0.05. As shown in Table 2, all fit indexes were up to standard, except for c2/df and RMSEA. Our sample size is larger than 1000, the value of c2/df is acceptable. In another study, the author pointed out that 0.05< RMSEA <0.08 was also acceptable[39]. Overall, the model fits well and the model is established.
As shown in Figure 3, the standardized path coefficient of burnout to job satisfaction is -0.41, indicates that burnout is negatively correlated with job satisfaction (p<0.01). It shows that when the other conditions are unchanged, the turnover intention decreases by 0.41 units for each additional unit of burnout. The standardized path coefficient of job satisfaction to turnover intention is -0.18, demonstrates that job satisfaction is also negatively correlated with turnover intention (p<0.01). Under the same other conditions, the turnover intention decreases by 0.18 units for each additional unit of job satisfaction. The standardized path coefficient of burnout to turnover intention is 0.83, reveals that burnout is positively correlated with turnover intention (p<0.01). That is, under the influence of job satisfaction, the turnover intention increases by 0.83 units for each additional unit of burnout.
The mediation effect is statistically significant (p<0.01), and the impact of burnout on turnover intention through the intermediary effect of job satisfaction is 0.074 (a*b= (-0.180)*-(0.410)). It manifests that when other conditions remain unchanged, the turnover intention will be indirectly increased by 0.074 units for each unit of burnout.