Paternalistic leadership (PL) is a culturally rooted leadership style characterized by the combination of authoritarianism, benevolence, and moral integrity (29, 41). Chinese construction sites, where hierarchical relationships are often embedded in Confucian values, foremen typically adopt PL behaviors to manage their crews (42). This leadership style establishes a familial and hierarchical dynamic in which the leader exercises authority while also demonstrating personal concern and moral example.

From a theoretical perspective, PL affects workers’ behaviors not only through direct authority but also via relational and psychological mechanisms. According to leader–member exchange (LMX) theory, high-quality leader–subordinate relationships foster trust, obligation, and mutual respect (43, 44), which in turn promote discretionary behaviors such as voluntary safety compliance. Moreover, under social exchange theory, leaders who display care and fairness motivate workers to reciprocate with positive organizational behaviors, including compliance with safety norms (45, 46).

In the context of construction safety, where formal rules alone may be insufficient to guide behavior on dynamic worksites, the relational dimensions of PL—especially benevolence and moral leadership—may inspire a stronger internalization of safety values. Workers are likely to engage in proactive safety behaviors not merely to avoid punishment but because of perceived loyalty and moral obligation to their leader (47, 48). Therefore, the follow-up hypothesis can be drawn:

Team safety climate (TSC) refers to workers’ shared perceptions about how safety is valued, prioritized, and supported within their immediate work group (49–52). In construction teams, these perceptions are primarily shaped by the observable behaviors and communication patterns of frontline supervisors (53–56). When foremen regularly stress safety, respond seriously to unsafe acts, and actively model compliance, these phenomena convey consistent cues that safety is a core team priority (13).

According to social information processing theory, individuals develop behavioral expectations and perceptions based on the social cues received from their environment (57). In construction work, which is dynamic, unpredictable, and often lacking detailed standard procedures, these signals become even more important. Workers typically observe how their foreman handles unsafe behavior, enforces rules, or responds to near-miss incidents to form a sense of how seriously safety is treated (58, 59). When these leader behaviors are consistent and supportive of safety, workers absorb this information and build a collective sense that safety is not only valued but expected (10). Paternalistic leaders, by blending authority with benevolence, play a unique role in this signaling process (30). Their concern for workers’ well-being lends credibility to safety expectations, while their moral integrity adds legitimacy to enforcement. Over time, such behaviors create a shared understanding of safety within the team. In this way, paternalistic leadership helps generate a normative climate that reinforces safety as a shared team value.

According to prior studies on safety climate (10, 60), a strong team safety climate increases employees’ awareness of occupational risks, encourages adherence to safety guidelines, and ultimately lowers the risk of workplace incidents (4, 61, 62). Additionally, an encouraging safety environment enhances job satisfaction, which in turn motivates employees to take a more active role and engage more willingly in safety initiatives (21, 63, 64), thereby improving work efficiency and enhancing overall safety performance. Hence, team safety climate positively affects workers’ safety behavior. Thus, according to the previous argument, this study presumes the following hypotheses:

Psychological safety reflects a worker’s belief that team members can express concerns, admit mistakes, and ask questions without fear of embarrassment or punishment (65, 66). On construction sites—where risks are high, mistakes can be costly, and authority structures are pronounced—this sense of interpersonal safety is essential for promoting voluntary safety behaviors such as reporting hazards, intervening in unsafe acts, and voicing suggestions (67–69).

According to psychological safety theory, interpersonal risk-taking is a function of the perceived social cost of speaking up or acting differently in a group (70). In construction teams, workers may hesitate to report problems or challenge unsafe behavior due to fear of negative consequences, especially when foremen are seen as punitive or indifferent. Paternalistic leadership helps mitigate this risk by providing a consistent sense of emotional protection. When leaders show genuine interest in workers’ well-being, handle mistakes with understanding, and demonstrate moral consistency, workers feel respected as individuals (67, 71). This perception reduces the psychological barriers to communication and engagement (72).

Research in innovation management suggests that individuals are more inclined to act autonomously and make decisions based on personal judgment in environments where psychological safety is high (73–75). Thus, construction workers are less influenced by external pressures and more inclined to engage in safe behaviors when valuing safety. Additionally, a psychologically safe atmosphere promotes mutual trust among team members, which facilitates the exchange of safety-related knowledge and skills (76–78), and safety knowledge and skills serve as fundamental prerequisites for engaging in safe behaviors. Therefore, based on the aforementioned analyses, the following hypotheses can be drawn.

Power distance (PD) refers to the extent to which individuals accept and expect power inequalities within social or organizational contexts (34, 79). High power distance environments typically feature strict hierarchies and limited subordinate autonomy, while low power distance cultures favor participatory leadership and egalitarian relationships (80). Meta-analytical evidence also highlights its role in shaping leadership outcomes, particularly in domains where obedience to authority directly affects rule compliance, such as occupational safety (39). Given that paternalistic leadership combines authority with benevolence (28), PD may critically shape how paternalistic leadership translates into team safety climate and psychological safety.

The effectiveness of paternalistic leadership may depend on how well its hierarchical nature aligns with the follower’s PD orientation. According to cultural congruence theory, leadership behaviors are more likely to be accepted and internalized when leaders are consistent with the subordinate’s cultural values and expectations (81). Workers high in PD are more likely to perceive paternalistic leadership as appropriate and legitimate, whereas low PD workers may interpret such behavior as over controlling or intrusive (82).

This alignment influences how workers interpret leadership signals and form collective perceptions. In the case of team safety climate, social information processing theory suggests that individual uses social cues—especially those from supervisors—to construct shared beliefs about organizational priorities (55, 83). When PD is high, workers are more attentive to leadership cues and less likely to question authority, thereby amplifying the effect of paternalistic leadership on the formation of a strong safety climate. In contrast, low PD workers may discount or reinterpret the same signals, weakening the emergence of shared norms around safety.

Similarly, in the development of workers’ psychological safety, power distance shapes how workers perceive interpersonal risks. Psychological safety theory emphasizes that individuals assess whether it is safe to speak up based on interpersonal trust and anticipated leader response (70). High PD workers are more likely to defer to leadership judgments and view benevolent authority as protective rather than punitive (29, 30). This fosters a stronger sense of psychological safety under paternalistic leadership. In contrast, low PD workers may expect autonomy and mutual dialog; if leaders exercise top-down control—even with good intentions—team members may feel constrained, thereby weakening the safety signal.

Therefore, according to the above analyses, this research proposes the following hypotheses:

In consequence, according to the aforementioned hypotheses, a conceptual model is developed for this research, which was presented in Figure 1.

This study focused on construction workers as participants and utilized a self-administered survey to gather data. The research instrument was divided into two parts. Part 1 included multiple-choice items designed to collect demographic information, such as gender, age, educational level, and year of work experience. Part 2 featured measurement scales assessing previously-highlighted variables, including FPL, TSC, WPS, PD, and WSB.

Fellows and Liu (84) argued that construction workers are often hard to access, and convenience sampling was frequently selected as the main sampling technique. To ensure effective access to the target population, data collection was carried out in three Chinese cities: Jiaozuo, Zhengzhou, and Changsha. The reasons why this research selected these three cities are that the research group can only access to workers in these cities because of the previous cooperation with some construction enterprises in these cities. This survey strategy helps us to gain a higher response rate and more reliable data. An online platform, Wenjuanxing, was chosen to host the digital version of the survey because many previous studies selected it as an online survey platform (25, 85). Initially, the questionnaire was sent to project managers who had collaborated with the research team. These individuals then forwarded the link to the workers under their supervision. To ensure respondent anonymity, surveys do not require personally identifiable information. Participants were informed that their responses were confidential and would be used solely for academic research purposes. Additionally, all surveys were completed privately by the workers without supervision from their foremen or project managers. The data collection period spanned from November 2024 to December 2024, lasting approximately 2 months. A total of 287 responses were received. After filtering out incomplete or invalid submissions, 263 usable responses were retained for further analysis. Table 1 provides an overview of the respondents’ demographic profiles.

Foremen’s paternalistic leadership was assessed using a scale developed by Cheng et al. (30), consisting of 12 items distributed across three dimensions: authoritarianism, benevolence, and moral virtue. This instrument has been extensively applied in organizational studies within the Chinese context. An example item reads: “My leader cares about my personal and professional well-being like a family elder.”

Team safety climate was evaluated with the scale introduced by Dedobbeleer and Beland (86), which contains nine items grouped into two factors: management commitment and worker participation. One sample item states: “Management considers safety to be a top priority.”

To measure workers’ psychological safety, the unidimensional scale proposed by Liang et al. (87) was employed. A representative statement from this scale is: “I feel safe expressing my opinions in front of team members.”

Power distance was captured using a five-item scale originally designed by Hofstede (88). As a unidimensional construct, it includes statements such as: “We should follow our superiors’ instructions without asking questions.”

Workers’ safety behavior was evaluated using the scale developed by Neal and Griffin (89), comprising six items across two dimensions: safety compliance and safety participation. A typical example is: “I comply with all safety rules at work.”

All aforementioned concepts were rated on a five-point Likert-type scale, ranging from 1 (strongly disagree) to 5 (strongly agree).

Several variables were selected as control variables because of their latent effects on WSB. These variables consist of workers’ gender (1 = male, 2 = female), educational level (1 = primary school or below; 2 = junior high school, 3 = senior high school or technical secondary school, 4 = university or above), and year of work experience (1 for ≤1, 2 for 2–4, 3 for 5–7, 4 for 8–10, 5 for >10).

To analyze the collected data, a range of statistical techniques was applied, including confirmatory factor analysis (CFA), structural equation modeling (SEM), and multiple regression analysis (MRA). CFA was conducted to assess the construct validity of the measurement model (90). SEM was utilized to evaluate both direct and mediating relationships (91). Specifically, an initial SEM model (SEM1) was developed to examine the impact of FPL on WSB. A second SEM model (SEM2), incorporating Bootstrap resampling techniques, was then constructed to investigate the mediating effects of TSC and WPS.

MRA was a widely-accepted statistical technique to test the moderating mechanism among variables (92, 93), thus this research performed MRA to explore the moderating effects of PD. Compared to the moderation analyses by incorporating an established latent interaction term in SEM, the moderation analysis in MRA is more convenient and easily conducted (94). Besides, the statistical approach to comprehensively test main effect, mediating effect and moderation effect by mixing SEM and MRA is well established in previous research (25, 95). Two baseline models (MRA1 and MRA2) were first established to assess the main effects among FPL, TSC, WPS, and PD. Subsequently, two extended models (MRA3 and MRA4), which incorporated interaction terms, were developed to test the moderating effect of PD on these relationships.

The collected data were imported into SPSS 23 for analysis, and Cronbach’s alpha coefficient was calculated to evaluate internal consistency. The results revealed alpha values of 0.92 for FPL, 0.93 for TSC, 0.91 for WPS, 0.89 for PD, and 0.94 for WSB—all surpassing the recommended threshold of 0.80. This suggests that the measurement instruments demonstrated strong reliability.

To examine the suitability of the data for factor analysis, the Kaiser-Meyer-Olkin (KMO) measure and Bartlett’s test of sphericity were conducted. KMO values for all variables exceeded 0.70, and Bartlett’s test yielded statistically significant results (p < 0.05), indicating that the dataset met the assumptions required for further exploratory and confirmatory analyses.

CFA was carried out using AMOS 23 to assess the construct validity of the measurement scales. Six measurement models (including the hypothesized model and 5 alternative models) were established for CFA, and the calculated results were presented in Table 2. As presented, our hypothesized model demonstrates a better fit (χ²/df = 2.21, GFI = 0.906, CFI = 0.929, RMSEA = 0.055). Convergent validity was evaluated through factor loadings, composite reliability (CR), and average variance extracted (AVE) based on the hypothesized model results. The calculating results are presented in Table 3. The measurement tools for FPL, TSC, and WPS exhibited acceptable levels of convergent validity. However, two items related to WSB were eliminated due to factor loadings below 0.70. The revised four-item version of the WSB scale subsequently passed the convergent validity criteria.

The square roots of the AVE for each construct, along with their correlations to other variables, were computed. These findings are summarized in Table 4. As illustrated, the square roots of the AVE for all variables exceed their corresponding Pearson correlation coefficients with other variables (dimensions), confirming adequate discriminant validity among the constructs.

The data were also imported into SPSS 23 to test common method variance; the Harman’s one-factor test shows that the first principal factor only explains 18.43% of the total variance (< 40%), which indicates that the common method variance is not significant.

SEM1 was established to investigate the direct impact of FPL on WSB. The model demonstrated acceptable fit based on standard fit indices: χ²/df = 2.35, GFI = 0.915, CFI = 0.943, and RMSEA = 0.053. These indicators collectively confirm that SEM1 achieved a satisfactory model fit. No post-hoc model modifications were made, as all proposed pathways were theoretically grounded and statistically significant. Moreover, modification indices did not indicate the need for additional paths to improve model fit. The results of SEM1 were presented in Figure 2.

Table 5 presents the path coefficients and their significance levels within SEM1. The analysis reveals that the regression coefficient for the relationship between FPL and WSB is statistically significant at p < 0.001, suggesting a strong positive association. Consequently, Hypothesis H1 is supported.

To explore the mediating effects of TSC and WPS on the relationship between FPL and WSB, a parallel mediation structural equation model (SEM2) was developed in this research. The model exhibited an acceptable fit: χ²/df = 2.18 < 3, GFI = 0.912 > 0.9, CFI = 0.931 > 0.9, and RMSEA = 0.052 < 0.08. Table 6 displays the path coefficients and their significance levels, while Table 7 summarizes the estimates and statistical significance of the indirect effects. The results of SEM2 were presented in Figure 3.

After 5,000 bootstrap resamples, the regression coefficients for the paths from PL to TSC (0.413) and from TSC to WSB (0.473) were both highly significant (p < 0.001). These findings confirm that FPL enhances TSC, which in turn promotes safer behaviors among workers. As a result, Hypotheses H2 and H3 are validated. Moreover, these results suggest that TSC partially mediates the link between FPL and WSB, supporting Hypothesis H4. The indirect effect through this pathway is estimated at 0.195, indicating a meaningful impact.

In addition, after 5,000 bootstrap samples, the regression coefficients for the paths from FPL to WPS and from WPS to WSB were 0.391 and 0.522, respectively, both statistically significant (p < 0.001). This indicates that FPL contributes to improving WPS, which subsequently leads to enhanced WSB. Thus, Hypotheses H5 and H6 are supported. Consequently, WPS acts as a mediator in the relationship between FPL and WSB, confirming Hypothesis H7. The indirect effect via this mechanism is calculated as 0.204.

Although the indirect effect sizes (0.195 and 0.204) appear moderate, these paths represent meaningful improvements in workers’ safety behavior, particularly in settings where small behavioral shifts can translate into significant reductions in accident risk (6). These effect sizes are comparable to or exceed those reported in prior safety behavior research using similar constructs (11, 25, 95). Besides, the direct effect of FPL on WSB in SEM2 was 0.104 and remained statistically significant, suggesting that FPL also exerts a direct positive impact on WSB. A comparison of the three pathways reveals that the direct impact of FPL on WSB is smaller than its indirect effects, implying that the overall impact of leadership on safety behavior primarily operates through these two psychological mechanisms.

To examine the moderating role of PD in the relationship between FPL and TSC, two multiple regression models (MRA 1 and MRA 2) were established. In MRA 1, FPL and PD served as independent variables, with TSC as the outcome variable. Building on this, MRA 2 introduced an interaction term (i.e., FPL × PD) to assess the moderating influence.

The results from both models are presented in Table 8. A statistically significant change was found in F value (ΔF = 18.753), suggesting that MRA 2 provided a better fit, as reflected by an increase in the adjusted R² by 0.036. Notably, the coefficient for the interaction term was 0.124 and reached statistical significance, indicating that PD positively moderates the effect of FPL on TSC.

Furthermore, MRA 3 and MRA 4 were developed to investigate the moderating effect of PD on the linkage between FPL and WPS. In MRA 3, FPL and PD were treated as independent variables, with WPS serving as the dependent variable. MRA 4 extended this model by incorporating an interaction term (i.e., FPL × PD) to assess the moderating effect.

The outcomes of these two models are summarized in Table 9. The change in F value (ΔF = 16.929) was found to be statistically significant, indicating that MRA 4 offered a better model fit, as evidenced by an increase of 0.041 in the adjusted R². Additionally, the coefficient for the interaction term was 0.175, which was also statistically significant. This result suggests that PD positively moderates the relationship between FPL and WPS.

The findings of this study provide actionable guidance for improving safety outcomes on construction sites through culturally grounded leadership strategies. To maximize practical relevance, this research offers three targeted recommendations while addressing potential barriers in cross-cultural settings.

First, construction firms operating in high-risk environments should move beyond generic leadership development and introduce modular training programs focused on scenario-based paternalistic leadership (PL). Instead of emphasizing only general authority or empathy, such training should simulate on-site dilemmas—e.g., responding to unsafe behavior while maintaining relational trust—to help foremen integrate moral integrity with discipline in practice. In parallel, formal performance evaluations should include 360-degree safety behavior assessments from both supervisors and workers, ensuring that PL effectiveness is reinforced through organizational incentives rather than informal approval alone (13).

Second, to cultivate a psychologically safe team climate that sustains safety behavior, site-level managers must institutionalize structured peer-review and “speak-up” routines. For example, weekly “safety listening huddles” can be co-led by foremen and peer champions to normalize open dialog on near-misses and unsafe acts. Unlike top-down safety audits, these grassroots safety conversations democratize risk perception while preserving PL’s role as a relational anchor. To make this effective, firms must also protect whistleblowers through formal “non-retaliation clauses” written into project protocols, particularly for subcontracted laborers who may be more vulnerable to retaliation (70).

Third, the moderating effect of power distance (PD) highlights the need for cultural calibration in multinational projects. While PL may align well with workers in East Asian or Middle Eastern settings, its hierarchical undertones may clash with egalitarian norms in European or North American teams. Multinational firms should adopt a dual-mode leadership strategy, enabling foremen to shift between culturally dominant PL and more participative, autonomy-supportive behaviors depending on team composition (40). This requires training not only in leadership style but also in cross-cultural adaptability—for example, through simulated intercultural role-play and real-time feedback tools. Additionally, project assignments should consider foremen–team cultural match as part of workforce deployment strategy (67).

However, barriers to implementing PL in multinational contexts must not be underestimated. Some foremen may lack moral legitimacy or interpersonal skills to enact PL credibly, especially when promoted based on technical seniority. Others may face language, trust, or role ambiguity issues when managing diverse crews unfamiliar with hierarchical relational norms. To address this, HR departments should offer cultural mentoring and language-pairing systems, linking foremen with multilingual safety liaisons who can mediate cultural misunderstandings while preserving leadership authority (115).

In sum, this study encourages firms to treat paternalistic leadership not merely as a trait of individual foremen, but as a strategically cultivable and context-sensitive management capability. When backed by institutional systems, cultural awareness, and credible support mechanisms, PL can serve as a practical lever to foster safer, more cohesive construction teams.

Despite this study’s theoretical and practical contributions, several limitations must be acknowledged and addressed in future research.

First, the use of a cross-sectional research design limits causal inference. Although structural equation modeling (SEM) allows examination of directional associations, it cannot fully rule out reverse causality. For example, workers with strong safety behaviors may elicit more paternalistic responses from their foremen. Future studies should consider longitudinal or experimental designs to capture the evolving relationship between leadership behaviors and safety engagement.

Second, this study relies exclusively on self-reported data collected through a single-source survey. This approach introduces potential social desirability bias, particularly in a collectivist context where workers may feel pressured to report high levels of safety compliance or favorable views of authority figures. Future research should integrate multi-source data, such as supervisor ratings, peer assessments, or behavioral safety observations, to reduce method bias and validate findings through triangulation.

Third, although this study controls for power distance (PD) as a moderator, it does not address other important individual or contextual factors, such as job stress, project complexity, or multicultural crews. Including these variables could deepen our understanding of how paternalistic leadership interacts with organizational conditions. For instance, examining whether benevolence’s safety-enhancing effect changes under time pressure versus routine work could provide a more nuanced view of leadership adaptability.

Fourth, while this study adopts an integrated view of paternalistic leadership, it does not differentiate its specific subdimensions—authoritarianism, benevolence, and moral leadership—in empirical testing. Future research should consider separating these components to explore whether benevolence enhances psychological safety while authoritarianism undermines it, especially in teams with mixed power distance values. Such research could challenge the assumption that all elements of PL function synergistically and contribute to more detailed theoretical development.

Fifth, although this study is situated in a single cultural context, future research could explore within-country variations in how paternalistic leadership operates across regions or organizational types. For instance, state-owned enterprises, private firms, and foreign-invested projects may differ in how workers respond to benevolence or authoritarian cues. Such comparative designs would help identify intra-national heterogeneity in leadership reception and cultural congruence, without assuming cross-national generalizability.

Finally, future research should move beyond confirming existing effects and consider paternalistic leadership as a dynamic, socially constructed influence. For instance, when does moral leadership become intrusive? How might generational shifts or rising individualism reshape worker expectations of foremen? At what point does care transform into control? Addressing such provocative questions would deepen our understanding of paternalism as a contested, evolving social phenomenon.