4.1. Data Analysis Based on Structural Equation Modeling
This research method is based on the views of Hair et al. [
45]. PLS-SEM (partial least squares structural equation modeling) is more suitable for complex models than covariance SEM, as it can estimate the explanatory power within samples and predictive ability outside samples. It is also suitable for exploring or extending theoretical models; therefore, this study uses SmartPLS4 to construct PLS-SEM to validate hypotheses.
Considering that it may not be feasible to completely eliminate common method bias, for the data analysis stage, two methods, namely the Harman single factor test (Podsakoff, 2003) and the full collinearity assessment approach by Kock (2017), were used to assess whether there was a common method problem in the questionnaire survey data.
- (1)
Harman’s Single-Factor method
Harman’s single-factor method was used for testing. All items involved in the scale were included in the factor analysis. Through principal component analysis, a total of eight common factors with eigenvalues greater than 1 were obtained, and the cumulative variance explained was 65.713%. Among them, the contribution rate of the first unrotated factor variance was 28.664%, which was less than 50%, indicating that this study did not have a serious common method bias.
- (2)
Full Collinearity Assessment Approach
Referring to the method of Kock (2017), a method based on collinearity tests was used to identify common method biases, namely when the variance inflation factor (VIF) of endogenous variables was less than 3.3, there was no serious common method bias in the study. The maximum VIF value of the endogenous variables (Inner Model) in this study was 1.564, indicating that there was no serious common method bias in this study.
- 2.
Reliability and validity analysis
The questionnaire data were input into SPSS 24.0 for reliability analysis, obtaining Cronbach’s α coefficients of each variable, as shown in
Table 2. Cronbach’s alpha coefficients for each first-order variable were all above 0.7 (minimum value = 0.751), indicating good internal consistency and stability of the designed scale, as well as high reliability. The composite reliability (CR) indicators were all above 0.7 (minimum value = 0.857), indicating good reliability. The average variance extracted (AVE) for the construct was above 0.5 for all variables (minimum value = 0.586), indicating good convergent validity. In addition, Cronbach’s alpha coefficient for the second-order variable of miner safety behavior was 0.859, the CR was 0.840, and the AVE was 0.637, all meeting the requirements, indicating good reliability and validity of the second-order variable in this study.
In addition, the Heterotrait–Monotrait Ratio (HTMT) was employed to test discriminant validity, with results shown in
Table 3: the HTMT ratios between each construct were less than 0.85, and the confidence intervals of the HTMT statistics, tested by the Bootstrap method, did not include 1, indicating that the variables in this study can be well distinguished. The Fornell–Larcker criterion was also used to test the discriminant validity between variables. By comparing the square root of the average variance extracted (AVE) of each variable in this study, it was found that they were all greater than the correlation coefficients between variables, indicating that there was good discriminant validity among the variables in this study.
- 3.
Hypothesis testing
AMOS 24.0 was used to build an SEM model of miners’ sense of job insecurity and safety behavior during the transformation period of coal mines, as shown in
Figure 2. LJI1~LJI4, JPI1~JPI4, and II1~II4 were observed variables for job loss insecurity, job performance insecurity, and interpersonal insecurity. PSQ1~PSQ4, JC1~JC3, and SM1~SM4 were observed variables for personal security quality, job characteristics, and safe management. A test for collinearity of the model was conducted, and the analysis results are shown in
Table 4. The maximum variance inflation factor (VIF) for the predictor variables in the model was 2.437, which was less than 5, indicating no significant collinearity among the endogenous variables. From this, it can be seen that the model fits the data well and has a certain degree of adaptability.
After resampling the model using the Bootstrap method 5000 times, the results of the path significance test are shown in the table above: among them, the job loss insecurity, job performance insecurity, and interpersonal insecurity all had a significant negative impact on miners safety behavior (p < 0.05), with standardized coefficients β of −0.188, −0.152, and −0.124, respectively, thus supporting hypotheses H1, H1a, H1b, and H1c.
The results of the inspection showed that job loss insecurity, job performance insecurity, and interpersonal insecurity all had a significant negative impact on psychological resilience (p < 0.05), with standardized coefficients β of −0.173, −0.168, and −0.122, respectively. Psychological resilience had a significant positive impact on miner safety behavior (p = 0.022, β = 0.203).
- 4.
Mediation effect test
Using the bias-corrected nonparametric percentile bootstrap method to estimate the 95% confidence interval of the mediated effect obtained by the coefficient product method, if the 95% confidence interval did not include 0, it indicated that the mediated effect was significant. As shown in
Table 5, the analysis results indicated that for the mediating effect of psychological resilience on the relationship between job loss insecurity and miner safety behavior, the 95% confidence interval was [−0.080, −0.007], which did not include 0. This suggested that there was a significant mediating effect of psychological resilience on the relationship between job loss insecurity and miner safety behavior in this study, with a standardized effect size of −0.035. Hypothesis H2a was thus supported. Similarly, it could be inferred that there was a significant mediating effect of psychological resilience on the relationship between job performance insecurity, interpersonal insecurity, and miner safety behavior, with 95% confidence intervals of [−0.075, −0.008] and [−0.058, −0.005], respectively. The standardized mediating effect sizes were −0.034 and −0.025. Hypotheses H2b and H2c were thereby supported.
- 5.
Testing the moderating effect of team safety climate
The results of the model analysis show that the interaction between team safety climate and job loss insecurity (TSC × JLI) had a significant positive impact on psychological resilience (
p = 0.002, β = 0.129). This means that the team safety climate in this study had a significant positive moderating effect on the relationship between job loss insecurity and psychological resilience, supporting hypothesis H3a. To better illustrate the moderating effect, following the suggestion of Aiken and West (1991), the moderation effect decomposition diagram is shown in
Figure 3a. Similarly, it can be inferred that the interaction between team safety climate and job performance insecurity (TSC × JPI) had a significant positive impact on psychological resilience (
p < 0.001, β = 0.157), indicating that team safety climate in this study had a significant positive moderating effect on the relationship between job performance insecurity and psychological resilience, supporting hypothesis H3b. The moderation effect decomposition diagram is shown in
Figure 3b. Team safety climate had a significant positive moderating effect on the relationship between job performance insecurity and miner safety behavior (
p = 0.020, β = 0.141), supporting hypothesis H4b. The moderation effect decomposition diagram is shown in
Figure 3c.
In addition, the results showed that the team safety climate had no significant moderating effect on the relationship between interpersonal insecurity and psychological resilience (p = 0.244); the team safety climate had no significant moderating effect on the relationship between job loss insecurity and miner safety climate (p = 0.705); the team safety climate had no significant moderating effect on the relationship between interpersonal insecurity and miner safety climate (p = 0.897); hypotheses H3c, H4a, and H4c were not supported.
Using the PROCESS program in SmartPLS 4.0, the moderating effect of team safety climate on the mediating effect of psychological resilience on the relationship between job loss insecurity, job performance insecurity, interpersonal insecurity, and miner safety behavior was analyzed. As shown in
Table 6, the 95% confidence interval for the moderation-mediated effect coefficient of team safety climate on the relationship between job loss insecurity and miner safety behavior through psychological resilience was [0.006, 0.063], which did not include 0. This indicates that team safety climate had a significant moderating effect on the relationship between job loss insecurity and miner safety behavior through psychological resilience. Hypothesis H5a was supported. Similarly, it can be known that there was a significant moderating effect of team safety climate on the mediating role of psychological resilience between job performance insecurity and miner safety behavior (95% = [0.008, 0.075], IMM = 0.033), supporting hypothesis H5b; the moderating effect of team safety climate on the mediating role of psychological resilience between interpersonal insecurity and miner safety behavior was not significant (95% = [−0.004, 0.036]), the confidence interval included 0, rejecting hypothesis H5c.
4.2. Data Analysis Based on Fuzzy Set Qualitative Comparative Analysis (fs QCA)
With limitations, traditional structural equation models cannot reveal the multiple concurrent causal relationships of numerous influencing factors on miner safety behavior and their asymmetry in influencing miner safety behavior, as well as the causal complexity issues of multiple equivalent schemes for high miner safety behavior formation. Given the comprehensiveness and complexity of the influencing factors of miner safety behavior and its own complex characteristics, it is necessary to use non-heap research methods to explore the complex impact relationships among the influencing factors of miner safety behavior. Therefore, this article further analyzes the complex relationship affecting miner safety behavior using fuzzy set qualitative comparative analysis (fs QCA).
As the kernel of the fsQCA method is Boolean operations, uncalibrated raw data cannot be directly subjected to Boolean operations. Therefore, in the study, the direct calibration method was used first, and the sample statistics of the antecedent conditions and results were calibrated using the 95-50-5 calibration method; that is, the 95th percentile, median, and 5th percentile were set as the complete membership, crossover point, and completely non-membership calibration anchors. In addition, to avoid cases being deleted due to the exact 0.50 membership in the antecedent conditions or consequent set, a constant of 0.001 was added to all conditions and results with a membership of 0.50 [
37,
38] below. The calibration anchor points and descriptive statistics of the target set are shown in
Table 7.
Secondly, to further examine the necessity of job loss insecurity, job performance insecurity, interpersonal insecurity, psychological resilience, and team safety climate in explaining miner safety behavior, and then determine whether the generation of miner safety behavior depends on a single conditional variable, in the thesis, the “NCA” package in R software was called, and necessity test used two methods, ceiling regression (CR) and ceiling envelopment (CE), were used. Furthermore, the CR method was applied to analyze the bottleneck level of miner safety behavior. In addition, the necessity of individual conditions for miner safety behavior was tested for robustness using QCA4.1. Consistency could reflect the degree to which these five independent variables were necessary conditions for miner safety behavior; coverage could reflect how many samples could explain the existence of these necessary conditions. To ensure the accuracy of the empirical conclusions, this study referred to previous research and set the consistency threshold at 0.9. The necessary condition analysis results are shown in
Table 8, where “~” represents “not”.
The results in the table above show that the necessary consistency of individual conditions for high/not-high miner safety behavior in this study is generally low (all less than 0.9), and they are not necessary conditions for high/not-high miner safety behavior, which was consistent with the results of the NCA, indicating that job loss insecurity, job performance insecurity, interpersonal insecurity, psychological resilience, and team safety climate are not necessary conditions for high/not-high miner safety behavior.
In order to further analyze the explanatory power of the comprehensive correlation between multiple independent variables and miners’ safety behaviors and explore the configuration path that affects safety behaviors, the configuration that generated high/not-high miner safety behavior, as well as the synergistic relationship between the various conditional variables within the configuration were analyzed. Based on simple solutions, intermediate solutions, and complex solutions, the configuration was named and explained. Through the combination of theory and practice, it was clarified that job loss insecurity, job performance insecurity, interpersonal insecurity, and the adaptation and substitution relationship between psychological resilience and team safety climate are the causes of high/not-high miner safety behavior.
Specifically, the fsQCA method was applied to analyze the configuration of job loss insecurity, job performance insecurity, interpersonal insecurity, psychological resilience, and team safety climate that lead to high/not-high miner safety behavior. Due to the inconsistent research conclusions of the correlation between job loss insecurity, job performance security, interpersonal insecurity, psychological resilience, and team safety climate and miners’ safety behavior, it was hypothesized in the study that the job loss insecurity, job performance insecurity, interpersonal insecurity, psychological resilience, and team safety climate, individually or collectively, may constitute one of the antecedents of high/not-high miner safety behavior in counterfactual analysis.
In addition, the core and boundary conditions of each configuration (intermediate solution) were further distinguished by comparing the intermediate solution with the simplified solution. The conditions that appear in both solutions are core conditions, and the conditions that only appear in the intermediate solution are boundary conditions. The conditional variables in the configuration are expressed in three forms: present (●) or (■), missing (U), and optional (not care). There are five states of conditional variables in the configuration, namely, the presence of core conditions (●), the presence of boundary conditions (■), the absence of core conditions (⃝), the absence of boundary conditions (☐), and optional (not care). The presence or absence of these conditions has no impact on the outcome.
Configuration S1: Core driving type of work insecurity and psychological resilience. The configuration is based on the core missing conditions of job loss insecurity, job performance insecurity, interpersonal insecurity, and the core existence condition of psychological resilience. This means that under the conditions of low job loss insecurity, low job performance insecurity, low interpersonal insecurity, and high psychological resilience, high miner safety behavior can be produced.
Configuration S2: Core driving type of job insecurity and team safety climate. The configuration is based on the core missing conditions of job loss insecurity and interpersonal insecurity, the edge missing condition of job performance insecurity, and the core existence condition of team safety climate. This means that under the conditions of a high team safety climate, low job loss insecurity, and low interpersonal insecurity, accompanied by low job performance insecurity, it is possible to produce high miner safety behavior.
Configuration S3: Core driving factors of job loss insecurity, interpersonal insecurity, and team safety climate. The configuration is based on the core missing conditions of job loss insecurity and interpersonal insecurity, the core existence condition of team safety climate, and the marginal existence condition of psychological resilience; it means that under the conditions of low job loss insecurity, low interpersonal insecurity, and high team safety climate, accompanied by high psychological resilience, the result of high miner safety behavior can be produced.
As shown in
Table 9, there are four configurations that can generate non-high mining worker safety behaviors. The overall consistency of their solutions is 0.887, and the overall coverage of their solutions is 0.695.
Configuration NS1 means that under conditions of high job loss insecurity, low psychological resilience, and low team safety climate, the result of miner safety behavior is not high.
Configuration NS2 means that under conditions of high job loss insecurity, high job performance insecurity, and low team safety climate, the resulting miner safety behavior will not be high.
Configuration of NS3 means that under conditions of high job insecurity, high interpersonal insecurity, and low team safety climate, the resulting safety behavior of miners will not be high.
Configuration NS4 means that under conditions of high job loss insecurity, high job performance insecurity, high interpersonal insecurity, and low psychological resilience, the resulting miner safety behavior will not be high.