Multivariate modeling of organizational dynamics and faculty well-being in higher education Authors: Cristhian George Alave Mamani, Esteban Tocto-Cano, Ronny Ivan Gonzales Medina, Javier Linkolk López-Gonzales, Josue Edison Turpo-Chaparro Published: 2026 ABSTRACT AbstractBackgroundLeadership styles are increasingly recognised as strategic organizational resources for faculty well-being, yet the evidence remains fragmented: prior studies have examined leadership, workplace harassment—systematic, recurrent negative acts that constrain the target’s capacity for self-defence—and psychological resources in isolation, and rarely within a single explanatory model. Whether these processes operate equivalently for male and female faculty has been largely untested, despite its direct relevance for equitable university management.AimTo test an integrated structural model relating transformational and servant leadership to workplace harassment, teacher self-efficacy, and life satisfaction among Peruvian university teachers; to formally test the mediating role of self-efficacy against directly estimated residual effects; and to examine whether the structural relations differ by gender.DesignA quantitative, non-experimental, cross-sectional study with explanatory scope.Participants506 university teachers from Peru (49.6% men, 50.4% women; 61.66% from public institutions).MethodsStructural Equation Modelling with robust WLSMV estimation on ordinal indicators, including directly estimated residual paths from both leadership styles to life satisfaction. Indirect effects were evaluated with the product-of-coefficients method and bootstrap bias-corrected accelerated confidence intervals (B=5000; ML sensitivity analysis). Measurement invariance across gender was tested with a configural–metric–scalar (threshold) sequence under WLSMV, and structural gender differences were assessed with a formal likelihood-ratio comparison of free versus constrained multigroup models, followed by Wald tests of path differences. Robustness to age, university type, and region was verified through a covariate-adjusted sensitivity model.ResultsBoth transformational (β=−0.123, p0.01) and servant (β=−0.186, p0.001) leadership were negatively associated with workplace harassment. Servant leadership showed the strongest effect on self-efficacy (β=0.505, p0.001), which in turn predicted life satisfaction (β=0.429, p0.001; R2=0.393). Servant leadership exhibited complementary partial mediation: both its indirect effect via self-efficacy (βstd=0.217, BCa CI [0.146,0.354]) and its direct residual effect (β=0.251, BCa CI [0.198,0.597]) were significant. Transformational leadership showed a small indirect-only effect (βstd=0.033, BCa CI [0.003,0.101]) with a non-significant direct path. Serial paths through harassment were not significant. Measurement invariance across gender was supported (ΔCFI ≤0.003 at every step), licensing structural comparison; full structural equivalence was rejected (Δχ2(8)=17.86, p=0.022), with Wald tests localising the differences exclusively to the transformational-leadership paths: its protective effect against harassment operated among women (β=−0.238, p0.001) but not men (β=−0.022, n.s.), whereas its effect on self-efficacy operated among men (β=0.157, p0.05) but not women (β=−0.024, n.s.). All conclusions were robust to age, university type, and region (all |Δβ|≤0.012).ConclusionServant leadership functions as a gender-equivalent organisational resource for faculty well-being, whereas transformational leadership operates through gender-specific pathways—protecting women against workplace harassment and strengthening men’s efficacy beliefs. These findings provide differentiated, data-driven guidance for university management aligned with Sustainable Development Goals 5 (gender equality) and 16 (just institutions), indicating that leadership development must be gender-responsive rather than uniform.Keywords: Transformational leadership, Servant leadership, Teacher self-efficacy, Workplace bullying, Life satisfaction, SEM, Measurement invariance, Faculty well-being Introduction Leadership constitutes a strategic mechanism through which universities strengthen faculty well-being and prevent psychosocial risks. Universities and higher education institutions implement programmes to enhance faculty well-being and organisational climate because these factors directly influence student performance and the mental health and sustainability of academic work [1]. Within this framework, leadership structures interpersonal relationships, promotes healthy work environments, and reinforces psychological resources, thereby mitigating psychosocial risks [2]. Transformational and servant leadership, in particular, improve the teaching experience and reduce workplace harassment [3, 4]. These leadership approaches therefore not only enhance academic management but also consolidate organisational conditions that sustain teachers’ overall well-being. Workplace harassment constitutes a critical psychosocial risk in university teaching that undermines both individual well-being and institutional performance. It is defined as systematic interpersonal aggression perpetrated by superiors or peers, whereby the target is exposed to recurrent negative acts that constrain the capacity for self-defence [5]. In academic settings, its prevalence is linked to hierarchical and competitive organisational cultures, which exacerbate faculty mental health problems, hinder career development, and compromise educational quality [6]. Identifying the organisational conditions that mitigate this risk is therefore essential for fostering healthy university environments and safeguarding academic sustainability. Transformational and servant leadership enhance life satisfaction by strengthening psychological resources and fostering supportive organisational climates in higher education. Life satisfaction represents a global indicator of subjective well-being and refers to the cognitive evaluation individuals make of their overall quality of life according to self-defined criteria [7]. In the university context, transformational leadership facilitates change management and improves teaching performance [8], while promoting academic innovation and strengthening educational management quality [9, 10]. Complementarily, servant leadership promotes job satisfaction, affective commitment, and professional performance [11] through the integration of emotional, relational, and value-oriented dimensions that emphasise empowerment and teamwork [12]. Both leadership styles have been shown to exert a positive influence on life satisfaction by reinforcing well-being, trust, and motivational processes [12, 13]. An important gap persists in the literature regarding the integrated examination of leadership, workplace harassment, self-efficacy, and teacher well-being within a single explanatory framework, particularly in the Peruvian higher education context. Although prior studies have documented pairwise associations among these variables, this evidence remains fragmented and has rarely assessed them simultaneously within a comprehensive structural model. A second, largely unaddressed question concerns gender: whether the organisational mechanisms linking leadership to faculty well-being operate equivalently for male and female academics. Given the documented gender asymmetries in exposure to academic workplace hostility [6], assuming equivalence without testing it risks gender-blind management prescriptions. The present study addresses both gaps by estimating an integrated structural model and by formally testing its measurement and structural invariance across gender. Social cognitive theory and the Job Demands–Resources (JD–R) model conceptualise self-efficacy as a central mechanism linking environmental conditions to individual outcomes [14, 15]. From a social cognitive perspective, efficacy beliefs regulate motivation, affect, and performance by shaping how individuals interpret and respond to contextual demands [14]. In organisational settings, leadership styles function as salient social influences that can strengthen or weaken these efficacy beliefs through modelling, feedback, and resource provision [16]. Within the JD–R framework, leadership operates as a work resource that mitigates harmful job demands, such as workplace harassment, while enhancing personal resources, including self-efficacy, thereby fostering higher levels of life satisfaction [13]. In the structural model, self-efficacy is an endogenous mediator: it receives direct effects from both leadership styles and from workplace harassment, and it transmits these influences to life satisfaction. Crucially, the model also estimates the direct residual paths from each leadership style to life satisfaction, so that the presence and type of mediation (partial versus indirect-only) is empirically tested rather than imposed by model specification [17]. Because all structural paths are strictly unidirectional and the model contains no feedback loops, the model is recursive and identified without additional restrictions. Finally, the article is structured into six sections. Section 2 reviews the related literature and derives the hypotheses ; Section 3 details the methodology; Section 4 reports the results; Section 5 discusses the findings; and Section 6 presents the conclusions derived from the empirical analysis. Related works and hypothesis development This section reviews the literature hypothesis by hypothesis, making explicit how social cognitive theory [14] and the JD–R model [15] inform each predicted relationship, and closes by delimiting which hypotheses replicate prior findings and which constitute novel contributions. These two frameworks were preferred over alternatives such as Conservation of Resources or Social Exchange theory because, in combination, they specify both the psychological mechanism of the mediator (efficacy beliefs as self-regulatory cognitions; social cognitive theory) and the organisational architecture of the predictors (leadership as a job resource that simultaneously buffers demands and builds personal resources; JD–R). The integrated model extends both frameworks in two ways: it treats two distinct leadership styles as separable job resources with potentially different operating channels, and it models gender explicitly as a boundary condition rather than assuming universal mechanisms. Leadership and workplace harassment (H1) Within the JD–R framework, leadership is a job resource that shapes the organisational conditions under which hostile behaviours emerge or are deterred. Empirical evidence indicates that servant leadership is associated with a lower prevalence of hostile workplace behaviours, underscoring its role in preventing harassment dynamics [3]. Leadership also moderates the relationship between stress and exposure to harassment, demonstrating that effective management practices attenuate the organisational conditions that facilitate such behaviours [18]. Transformational leadership similarly exerts a preventive function among university faculty and operates as a protective organisational resource [4]. Accordingly, H1 posits that transformational (H1a) and servant (H1b) leadership exert a negative influence on workplace harassment. Leadership and teacher self-efficacy (H2) Social cognitive theory identifies vicarious experience, verbal persuasion, and mastery support as primary sources of efficacy beliefs [14]; leaders are the most salient organisational providers of all three [16]. Empirical work consistently links teacher self-efficacy to leadership practices, indicating that organisational environments can strengthen or undermine teachers’ beliefs in their professional capabilities [19], and that supportive, inspirational management predicts higher job satisfaction and more positive perceptions of the work climate [20–22]. Accordingly, H2 proposes that transformational (H2a) and servant (H2b) leadership positively influence teacher self-efficacy. Workplace harassment and self-efficacy (H3) Sustained exposure to hostile acts erodes perceptions of competence and control: longitudinal evidence shows that harassment predicts deteriorating mental health, insomnia symptoms, and heightened psychological distress [5]. From the JD–R perspective, harassment is a job demand that depletes personal resources [15]. H3 therefore states that workplace harassment negatively influences teacher self-efficacy. Self-efficacy and life satisfaction (H4) Efficacy beliefs foster more favourable cognitive evaluations of overall well-being [14]. Research on occupational well-being demonstrates that self-efficacy enhances intrinsic motivation, facilitates the adoption of innovative practices, and positively influences work-related well-being [23]. H4 posits that teacher self-efficacy positively predicts life satisfaction [7]. The mediating role of self-efficacy (H5) Integrating H1–H4, self-efficacy is the psychological conduit through which organisational resources reach subjective well-being [13, 15]. H5 posits that teacher self-efficacy mediates the relationship between leadership styles and life satisfaction (H5a: transformational; H5b: servant). Unlike prior specifications that omit direct leadership–satisfaction paths—thereby forcing full mediation by construction—the present model estimates these residual direct effects and classifies the mediation type empirically [17]. Gender as a boundary condition (RQ1) Exposure to academic workplace hostility is gender-asymmetric [6], and leadership perceptions may likewise differ across gender groups. Whether the structural mechanisms linking leadership, harassment, self-efficacy, and satisfaction are equivalent for male and female faculty is an open empirical question with direct implications for equitable management. RQ1 therefore asks: are the structural relations in the proposed model invariant across gender? Answering RQ1 requires first establishing measurement invariance and then formally comparing free and constrained multigroup structural models [24, 25]. Positioning and novelty Hypotheses H1, H2, and H4 replicate relationships documented in other settings and extend their empirical validation to the Peruvian higher education context [3, 4, 19, 23]. The novel contributions of this study are threefold: (i) the simultaneous estimation of all relationships within a single integrated structural model, including the previously untested serial pathway leadership → harassment → self-efficacy → satisfaction (H3, H5); (ii) the formal, specification-based test of mediation type with directly estimated residual paths; and (iii) the first formal test, to our knowledge, of measurement and structural gender invariance for this model in the Latin American higher education context, implemented with WLSMV estimation for ordinal indicators (RQ1). Whereas invariance testing is well established globally, typically with continuous indicators, its threshold-based application to a fully ordinal integrated model of faculty well-being remains rare. Methodology Figure 1 outlines the methodological framework of the study, organised into four interconnected components: research design, participants, measurements, and data analysis. Methodological framework (redesigned for clarity). Research design This study adopts a quantitative, non-experimental, cross-sectional design with explanatory scope and employs Structural Equation Modelling (SEM) to test the proposed hypotheses. Because the instrument items are measured using Likert-type ordinal scales, the analysis uses the Weighted Least Squares Mean and Variance Adjusted (WLSMV) estimator, which relies on polychoric correlations to estimate associations among continuous latent variables underlying ordinal indicators. This estimator corrects the chi-square statistic through mean and variance adjustments, thereby producing robust parameter estimates under violations of multivariate normality. Flora and Curran [26] demonstrate that the Maximum Likelihood estimator yields biased parameters, underestimated standard errors, and distorted fit indices when applied to ordinal variables with few response categories. Similarly, Li [27] shows that WLSMV achieves greater parametric accuracy and more stable standard errors under conditions of non-normality and asymmetry. Role of self-efficacy in the structural model Self-efficacy is specified as an endogenous mediator: it receives direct paths from Transformational Leadership, Servant Leadership, and Workplace Harassment, and it predicts Life Satisfaction. The model additionally estimates the direct residual paths from both leadership styles to Life Satisfaction, so that partial versus indirect-only mediation is an empirical outcome rather than a modelling assumption [17]. As established in the Introduction, this recursive specification is identified without further restrictions. Participants The sample comprised 506 Peruvian university teachers, recruited via a non-probability convenience sampling method. Inclusion criteria required participants to be actively engaged in university-level teaching at the time of the study. The sample exhibited a balanced gender distribution (49.6% men, 50.4% women). Most participants were affiliated with public universities (61.66%), whereas 38.34% worked in private institutions. Regarding geographical distribution, the coastal region predominated (60.87%), followed by the highlands (31.82%) and the jungle (7.31%). Participants’ ages ranged from 18 to 75 years (M=44.2, SD=11.8), reflecting substantial variability and enabling the inclusion of professional experiences across different stages of the academic career trajectory. Measures The study assesses five latent constructs through multidimensional instruments composed of ordinal-response items; for every measure, higher scores indicate a greater presence of the construct. Transformational Leadership, based on [28], is measured with four items on a five-point scale (0–4) (α=0.912, ω=0.954); Servant Leadership, grounded in [29], is assessed using fourteen items on a seven-point scale (1–7) (α=0.908, ω=0.952). Teacher Self-Efficacy, following [30], is operationalised through ten items on a seven-point scale (0–6) (α=0.931, ω=0.966); Life Satisfaction, based on [31], is measured with five items on a five-point scale (1–5) (α=0.849, ω=0.922). Workplace Harassment, according to [32], is evaluated using twenty items on a five-point scale (0–4), with higher scores indicating greater exposure to hostile acts (α=0.979, ω=0.990). All original factor structures were retained without modification. Data analysis procedure Data analyses were performed using R software (version 4.5.2) equipped with the lavaan (version 0.6–21), semTools, psych, and semPlot packages. The analytical sequence comprised: (1) descriptive statistics; (2) normality assessment via the Anderson–Darling test; (3) outlier identification via box plots; (4) Spearman rank-order correlations; (5) reliability estimation (α and ω); (6) Confirmatory Factor Analysis (CFA) with WLSMV; and (7) SEM estimation with WLSMV, including the direct residual paths from both leadership styles to life satisfaction. The scaled chi-square statistic and its degrees of freedom are reported for both the CFA and the structural model, alongside CFI, TLI, RMSEA, and SRMR. Indirect, direct, and total effects were quantified using the product-of-coefficients method within the WLSMV framework and corroborated by bootstrap bias-corrected accelerated (BCa) confidence intervals (B=5000) under Maximum Likelihood estimation as a sensitivity analysis [26]. Mediation type was classified following Zhao, Lynch and Chen [17]: a significant indirect effect accompanied by a significant direct residual effect indicates complementary (partial) mediation, whereas a significant indirect effect with a non-significant direct effect indicates indirect-only (full) mediation. Transformational (LT) and Servant (LS) leadership are evaluated as separate predictors throughout. Measurement invariance across gender was examined with a three-stage sequence (configural → metric → scalar) using the WLSMV estimator; because the indicators are ordinal, the scalar step constrains item thresholds to equality across groups [33]. To ensure stable threshold estimation, sparse response categories—observed in 10 of the 53 items when split by gender—were collapsed across groups prior to model fitting. Invariance decisions combine the scaled Δχ2 test with approximate fit criteria (ΔCFI ≤0.010) [24, 25], given the well-documented oversensitivity of the chi-square difference in large models. Structural equivalence across gender was then tested formally by comparing a multigroup model with free structural paths against a model constraining all structural paths to equality (scaled likelihood-ratio test), followed by Wald tests of each individual path difference. Because the omnibus likelihood-ratio test must reject equivalence before any individual comparison is interpreted, the Wald tests operate as a protected testing procedure; no additional multiplicity correction was applied, and the individual differences are accordingly interpreted as exploratory. Finally, a covariate-adjusted sensitivity model verified the robustness of all structural paths to age, university type (public/private), and region (coast/highlands/jungle). The complete lavaan syntax for every model is provided as supplementary material. Results Descriptive statistics Table 1 shows that the five latent constructs present descriptive profiles consistent with a predominantly favourable organisational context, although with non-normal distributions that justify the adoption of robust estimation procedures. Table 1Descriptive statistics of latent variables.Variable M SD SkewnessKurtosisTransformational Leadership2.370.97 −0.33 −0.30 Servant Leadership5.840.84 −1.70 4.70 Teacher Self-Efficacy4.811.05 −1.11 1.06 Workplace Harassment0.760.89 1.77 2.87 Life Satisfaction4.130.71 −0.90 1.09 Descriptive statistics of latent variables. Servant Leadership (M=5.84) exhibits the highest mean, while Workplace Harassment has the lowest (M=0.76), suggesting a generally positive organisational environment. The Anderson–Darling test confirms significant departures from normality across all constructs (p0.001), corroborating the observed skewness and elevated kurtosis values and precluding the use of estimators that assume multivariate normality. Outlier analysis Figure 2 presents boxplots for the five constructs. Outliers were most evident in Workplace Harassment: 37 cases (7.3%) exceeded the upper Tukey fence and 10 cases lay more than three standard deviations above the mean (maximum z=3.66), consistent with the construct’s positively skewed distribution. No cases were removed, given the adequate sample size (N=506) and the use of robust WLSMV estimation on polychoric correlations. A sensitivity re-estimation of the structural model excluding the 10 most extreme harassment scores preserved every substantive conclusion—no path changed sign and no significant path lost significance—while the harassment-related coefficients grew somewhat larger in magnitude and the harassment → self-efficacy path attained significance (β=−0.110, p0.05). The full-sample estimates are therefore, if anything, conservative: a small number of respondents reporting extreme harassment alongside preserved efficacy attenuates, rather than inflates, the harassment effects (see Section 5). Box plots of latent constructs. Correlation analysis Spearman rank-order correlations were computed in response to the non-normal distribution of the data. All associations were statistically significant (p0.001). The strongest correlations emerged between Servant Leadership and Teacher Self-Efficacy (ρ=0.476), and between Teacher Self-Efficacy and Life Satisfaction (ρ=0.491), highlighting the central role of psychological resources in linking leadership practices to well-being outcomes (Fig. 3). Spearman’s correlation matrix. Reliability analysis Table 2 reports the reliability indices for the evaluated constructs. All values exceed the recommended threshold of 0.70, demonstrating adequate internal consistency. Workplace Harassment attains the highest reliability (α=0.979; ω=0.990), while Life Satisfaction presents satisfactory indices (α=0.849; ω=0.922). Table 2Reliability indicesConstruct α ω Transformational Leadership0.9120.954Servant Leadership0.9080.952Teacher Self-Efficacy0.9310.966Workplace Harassment0.9790.990Life Satisfaction0.8490.922 Reliability indices Confirmatory factor analysis The measurement model demonstrates excellent fit to the empirical data (χscaled2(1315)=2697.4, p0.001; CFIscaled=0.982; TLIscaled=0.981; RMSEAscaled=0.046, 90% CI [0.043,0.048]; SRMR=0.056). All standardised factor loadings were statistically significant (p0.001), ranging from λ=0.417 (ls_14) to λ=0.945 (ac_8), thereby confirming the structural validity of the proposed constructs. Table 3 presents the complete standardised factor loadings for all 53 indicators (Fig. 4). Table 3Standardised factor loadings from the confirmatory factor analysis (WLSMV; N=506).FactorItem βstd SE z LTlt_10.8550.01556.28lt_20.9070.01277.04lt_30.9050.01179.20lt_40.8750.01559.06LSls_10.6780.02725.54ls_20.6990.02428.93ls_30.6910.02527.93ls_40.5980.02821.55ls_50.7120.02331.08ls_60.7920.02040.31ls_70.8120.02040.35ls_80.8030.02040.92ls_90.8420.01846.29ls_100.7920.02040.57ls_110.7130.02429.46ls_120.8230.01944.28ls_130.7870.01940.75ls_140.4170.03611.64AEae_10.5570.02621.51ae_20.8170.01649.74ae_30.8970.01275.27ae_40.8730.01369.07ae_50.8780.01366.11ae_60.8540.01460.94ae_70.8940.00999.18ae_80.8890.01180.36ae_90.8520.01459.50ae_100.8350.01651.43ACac_10.7570.02038.46ac_20.8200.01845.98ac_30.8190.01652.17ac_40.9080.00998.27ac_50.8960.01177.90ac_60.9140.00997.31ac_70.9120.01091.12ac_80.9450.008120.30ac_90.9050.01181.37ac_100.9170.009101.54ac_110.9370.007135.42ac_120.9200.01090.98ac_130.9210.01187.46ac_140.9150.01370.32ac_150.9170.01089.62ac_160.9380.01095.44ac_170.9410.01098.50ac_180.9140.01087.62ac_190.8800.01272.34ac_200.8620.02043.07SVsv_10.8600.02337.21sv_20.8350.02041.89sv_30.8610.02043.12sv_40.7840.02630.26sv_50.6780.02923.77Allp0.001. Standardised factor loadings from the confirmatory factor analysis (WLSMV; N=506). Allp0.001. Structural model with standardised coefficients (WLSMV estimator; N = 506). Solid arrows: hypothesised paths; dashed arcs: direct residual paths from each leadership style to life satisfaction. Red indicates negative effects; blue indicates positive effects. ρ denotes the latent covariance between the two leadership factors. Fit indices: χscaled2(1316)=2610.9, p0.001; CFI = 0.983; TLI = 0.982; RMSEA=0.044; SRMR=0.056. Structural equation model The structural model—which, unlike the previous version, estimates the direct residual paths from Transformational Leadership (LT) and Servant Leadership (LS) to Life Satisfaction (SV)—achieves excellent fit to the empirical data (χscaled2(1316)=2610.9, p0.001; CFIscaled=0.983; TLIscaled=0.982; RMSEAscaled=0.044; SRMR = 0.056). Table 4 presents the complete set of direct path coefficients with standardised estimates and their delta-method 95% confidence intervals computed on the standardised solution. Table 4Direct path coefficients from the structural model (WLSMV;N=506).OutcomePredictorβstdβrawSEzpCI2.5%CI97.5%ACLT−0.123−0.1270.045−2.7470.01−0.210−0.035ACLS−0.186−0.1930.038−4.9650.001−0.260−0.113AELT0.0780.0940.0441.7920.073−0.0070.163AELS0.5050.6050.03315.1330.0010.4400.571AEAC−0.055−0.0640.040−1.3800.168−0.1330.023SVAE0.4290.4590.04310.0150.0010.3450.513SVLT0.0660.0840.0481.3680.171−0.0280.160SVLS0.2510.3230.0514.9720.0010.1520.351SE,z, and 95% CIs refer to the standardised solution (delta method);βrawis the unstandardised estimate.R2: AC = 0.064; AE = 0.304; SV = 0.393. Latent covariance LT ↔ LS: ρ = .0319, p < 0.001. Unstandardised 95% CIs are provided in the supplementary material.LT, Transformational Leadership; LS, Servant Leadership; AC, Workplace Harassment; AE, Teacher Self-Efficacy; SV, Life Satisfaction. Direct path coefficients from the structural model (WLSMV;N=506). SE,z, and 95% CIs refer to the standardised solution (delta method);βrawis the unstandardised estimate. R2: AC = 0.064; AE = 0.304; SV = 0.393. Latent covariance LT ↔ LS: ρ = .0319, p < 0.001. Unstandardised 95% CIs are provided in the supplementary material. LT, Transformational Leadership; LS, Servant Leadership; AC, Workplace Harassment; AE, Teacher Self-Efficacy; SV, Life Satisfaction. A covariate-adjusted sensitivity model including age, university type (public/private), and region (coast/highlands/jungle) as exogenous predictors of harassment, self-efficacy, and life satisfaction left every substantive conclusion unchanged: all structural coefficients shifted by |Δβstd|≤0.012, and no path changed direction. Among the covariates themselves, only two effects were significant: residing in the highlands predicted higher harassment (β=0.136, p0.01) and age predicted higher life satisfaction (β=0.111, p0.01); all remaining covariate effects were non-significant. The only qualitative movement concerned the marginal LT → AE path, which reached conventional significance under covariate adjustment (β=0.085, p=0.043). This emergence is consistent with a mild suppression pattern rather than confounding: university type correlates with perceived transformational leadership (r=0.18) but is essentially unrelated to self-efficacy (r0.01), so adjusting for it removes criterion-irrelevant variance from the leadership predictor. Given the path’s failure to reach significance in the primary model and its sensitivity to specification, it is nevertheless interpreted conservatively as non-confirmed. Hypothesis testing Table 5 summarises the hypothesis tests. The threshold for confirmation of any hypothesis is a coefficient with p0.05 whose 95% confidence interval excludes zero; for indirect effects (H5), the bootstrap BCa interval is the decisive criterion, as it does not assume normality of the product term. Table 5Hypothesis testing results.HypothesisRelationshipβstdResultH1aLT→HarassmentConfirmedH1bLS→Harassment−0.186⁎⁎⁎ConfirmedH2aLT→Self-Efficacy0.078(n.s.,p=0.073)Not confirmedH2bLS→Self-Efficacy0.505⁎⁎⁎ConfirmedH3Harassment→Self-Efficacy−0.055(n.s.)Not confirmedH4Self-Efficacy→Satisfaction0.429⁎⁎⁎ConfirmedH5aLT→AE→SV (indirect)0.033†Confirmed (indirect-only)H5bLS→AE→SV (indirect)0.217⁎⁎⁎Confirmed (partial)For H1, H2 and H5, separate coefficients are reported for Transformational (LT) and Servant (LS) leadership.∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001. Threshold for confirmation of every hypothesis: p < 0.05 and a 95% confidence interval excluding zero. †H5a: delta-method p = 0.085; confirmed by the bootstrap BCa interval [0.003, 0.101] (B = 5000), which excludes zero (see Table 6); interpreted as a small indirect-only effect. H2a reached significance in the covariate-adjusted sensitivity model (β = 0.085, p < 0.05) but not in the main model, and is therefore reported conservatively as not confirmed. Hypothesis testing results. For H1, H2 and H5, separate coefficients are reported for Transformational (LT) and Servant (LS) leadership. ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001. Threshold for confirmation of every hypothesis: p < 0.05 and a 95% confidence interval excluding zero. †H5a: delta-method p = 0.085; confirmed by the bootstrap BCa interval [0.003, 0.101] (B = 5000), which excludes zero (see Table 6); interpreted as a small indirect-only effect. H2a reached significance in the covariate-adjusted sensitivity model (β = 0.085, p < 0.05) but not in the main model, and is therefore reported conservatively as not confirmed. The structural model explains 30.4% of the variance in Teacher Self-Efficacy, 6.4% in Workplace Harassment, and 39.3% in Life Satisfaction. Indirect, direct and total effects Table 6 presents all defined effects, estimated with the product-of-coefficients method under WLSMV and corroborated by bootstrap BCa confidence intervals (B=5000; ML estimator). Table 6Indirect, direct residual, and total effects.EffectβstdpBCa CI2.5%BCa CI97.5%LT→AE→SV (indirect)0.0330.0850.0030.101LS→AE→SV (indirect)0.2170.0010.1460.354LT→AC→AE→SV (serial)0.0030.244−0.0010.012LS→AC→AE→SV (serial)0.0040.180−0.0040.019LT→SV (direct residual)0.0660.171−0.0270.184LS→SV (direct residual)0.2510.0010.1980.597Total LT→SV0.1020.0580.0040.242Total LS→SV0.4720.0010.4420.799Point estimates are standardised (WLSMV);pvalues are delta-method; the BCa 95% intervals (B=5000, ML estimator) are computed on the unstandardised effects and constitute the decisive significance criterion for indirect effects.BCa intervals that exclude zero indicate statistical significance. The serial paths through workplace harassment are not significant, indicating that harassment does not constitute a viable indirect route from leadership to life satisfaction.AE, Teacher Self-Efficacy; SV, Life Satisfaction; AC, Workplace Harassment. Indirect, direct residual, and total effects. Point estimates are standardised (WLSMV);pvalues are delta-method; the BCa 95% intervals (B=5000, ML estimator) are computed on the unstandardised effects and constitute the decisive significance criterion for indirect effects. BCa intervals that exclude zero indicate statistical significance. The serial paths through workplace harassment are not significant, indicating that harassment does not constitute a viable indirect route from leadership to life satisfaction. AE, Teacher Self-Efficacy; SV, Life Satisfaction; AC, Workplace Harassment. Mediation: analytical strategy and classification Because the model estimates the direct residual paths LT → SV and LS → SV, the type of mediation is an empirical outcome rather than a specification artifact. Following Zhao, Lynch and Chen [17], a significant indirect effect combined with a significant direct effect of the same sign indicates complementary partial mediation, whereas a significant indirect effect with a non-significant direct effect indicates indirect-only (full) mediation. Applying these criteria: (i) servant leadership displays complementary partial mediation—its indirect effect via self-efficacy (BCa [0.146,0.354]) and its direct residual effect (BCa [0.198,0.597]) are both significant and positive; and (ii) transformational leadership displays a small indirect-only effect—its indirect path is significant by the bootstrap criterion (BCa [0.003,0.101]) while its direct path is not ([−0.027,0.184]). The lower bound of the LT indirect interval lies close to zero, so this effect should be interpreted as small and detected only by the more powerful resampling criterion. Gender invariance and multigroup analysis (RQ1) Measurement invariance A three-stage sequence (configural → metric → scalar) was estimated with WLSMV; because the indicators are ordinal, the scalar step constrains item thresholds [33] (Nmen=251, Nwomen=255). Results are presented in Table 7. The scaled Δχ2 for the metric step was significant (Δχ2(48)=239.05, p0.001); however, this statistic is known to be oversensitive in large ordinal models, and the approximate fit criteria clearly support invariance: ΔCFI =0.002 and ΔSRMR =0.006, well below the ΔCFI ≤0.010 threshold [24, 25]. The scalar (threshold) step produced no deterioration whatsoever (CFI improved to 0.986; the scaled Δχ2 was negative, an admissible outcome of the Satorra–Bentler correction [34]). Measurement invariance across gender is therefore supported, licensing the structural comparison. Table 7Measurement invariance across gender (WLSMV;Nmen=251,Nwomen=255).Modelχscaled2dfCFIRMSEASRMRΔχ2(Δdf)ΔCFIConfigural3869.9726300.9850.0430.064——Metric (loadings)4139.6026780.9830.0470.070239.05(48)⁎⁎⁎0.002Scalar (thresholds)4022.6328090.9860.0410.0650(n.s.)−0.003∗∗∗p < 0.001. Invariance decisions follow the approximate fit criterion ΔCFI ≤ 0.010 [24, 25], given the oversensitivity of Δχ2 in large ordinal models. A negative scaled Δχ2 is an admissible outcome of the mean-and-variance correction [34] and indicates no deterioration in fit. Measurement invariance across gender (WLSMV;Nmen=251,Nwomen=255). ∗∗∗p < 0.001. Invariance decisions follow the approximate fit criterion ΔCFI ≤ 0.010 [24, 25], given the oversensitivity of Δχ2 in large ordinal models. A negative scaled Δχ2 is an admissible outcome of the mean-and-variance correction [34] and indicates no deterioration in fit. Structural comparison A multigroup structural model with all regression paths freely estimated in each gender group (measurement loadings constrained) was compared against a model constraining all eight structural paths to equality. The scaled likelihood-ratio test rejected full structural equivalence (Δχ2(8)=17.86, p=0.022). Following this rejection, Table 8 details the freely estimated coefficients separately for men and women, and Table 9 then formally assesses each individual path difference through Wald tests. Table 8Structural path coefficients by gender group (WLSMV multigroup SEM; free structural paths, loadings constrained).OutcomePredictorGroupβstdSEzpACLTMen−0.0220.065−0.3420.732ACLTWomen−0.2380.058−4.0060.001ACLSMen−0.2010.058−3.5790.001ACLSWomen−0.2260.047−4.8760.001AELTMen0.1570.0812.4830.05AELTWomen−0.0240.080−0.3890.697AELSMen0.5450.0788.9620.001AELSWomen0.4580.0669.3970.001AEACMen−0.0410.066−0.7810.435AEACWomen−0.0780.081−1.2990.194SVAEMen0.4560.0736.6490.001SVAEWomen0.3960.0616.9000.001SVLTMen0.1090.0771.9350.053SVLTWomen0.0850.1061.1200.263SVLSMen0.2440.0834.0390.001SVLSWomen0.2260.0973.3850.001 Structural path coefficients by gender group (WLSMV multigroup SEM; free structural paths, loadings constrained). Table 9Wald tests of gender differences in structural paths (unstandardised differences, Men – Women).Path differenceEstimateSEzpCI2.5%CI97.5%LT→AC0.2100.0872.4240.050.0400.379LS→AC0.0240.0720.3270.744−0.1180.165LT→AE0.2330.1142.0370.050.0090.457LS→AE0.0760.0870.8780.380−0.0940.246AC→AE0.0530.1040.5120.609−0.1510.258AE→SV0.0680.0850.7950.427−0.0990.235LT→SV0.0310.1300.2420.809−0.2240.287LS→SV0.0070.1230.0550.956−0.2340.248Only the two transformational-leadership paths differ significantly between genders; the remaining six paths are statistically equivalent. Given the eight simultaneous comparisons, these two localised differences should be interpreted as exploratory and warrant replication. Wald tests of gender differences in structural paths (unstandardised differences, Men – Women). Only the two transformational-leadership paths differ significantly between genders; the remaining six paths are statistically equivalent. Given the eight simultaneous comparisons, these two localised differences should be interpreted as exploratory and warrant replication. The pattern is highly localised: significant gender differences emerged exclusively in the two transformational-leadership paths—LT → workplace harassment and LT → self-efficacy—while the remaining six paths, including all servant-leadership paths and the central self-efficacy → satisfaction mechanism, are statistically equivalent across genders. The protective effect of transformational leadership against workplace harassment operates among women (β=−0.238, p0.001) but is absent among men (β=−0.022, n.s.), while its effect on self-efficacy operates among men (β=0.157, p0.05) but is absent among women (β=−0.024, n.s.). Under a Bonferroni correction for the eight comparisons (α=0.00625), neither difference would retain significance (adjusted p=0.123 and 0.334); the family-wise error rate is protected instead by the preceding omnibus test, and these localised differences are accordingly reported as exploratory findings awaiting replication. Discussion The findings substantiate that transformational (LT) and servant leadership (LS) operate as organisational resources within the Job Demands–Resources framework, promoting faculty well-being and mitigating psychosocial risks in Peruvian higher education. Both leadership styles significantly buffer workplace harassment, and servant leadership exerts by far the strongest effect on teacher self-efficacy. These findings align with previous literature emphasising that leadership practices grounded in compassion, inspiration, and organisational care construct protective climates that deter hostile behaviours [3, 4]. Notably, the pooled protective effect of transformational leadership on harassment, like its effect on self-efficacy discussed below, masks gender heterogeneity: the multigroup analysis (Section 4.10) shows it is driven almost entirely by the female subsample. Nevertheless, the limited variance explained in workplace harassment (R2=0.064) suggests that this risk also reflects deeply entrenched structural determinants and institutional power asymmetries not fully captured by leadership styles alone [6]; leadership development is therefore best understood not as a standalone remedy but as a strategic entry point for broader institutional dialogue on organisational culture and power structures. Two results depart from expectation and merit careful interpretation. First, sustained exposure to workplace harassment did not significantly degrade teacher self-efficacy (H3). This attenuation may stem from a ceiling effect: self-efficacy responses cluster at the high end of the 0–6 scale (M=4.81), restricting the observable variance and thereby limiting the sensitivity of the polychoric correlations to detect degradation even where present; cross-sectional constraints additionally preclude the detection of delayed or cumulative erosion. Consistent with this attenuation account, the outlier sensitivity analysis (Section 4.2) showed that the harassment → self-efficacy path attained significance once the ten most extreme harassment scores were excluded. Second, the aggregate effect of transformational leadership on self-efficacy was small and did not reach conventional significance in the main model (H2a: β=0.078, p=0.073), although it did under covariate adjustment. The multigroup analysis clarifies why: this path is gender-moderated, operating among men (β=0.157, p0.05) but not women (β=−0.024, n.s.), so the pooled estimate averages a real effect with a null one. This illustrates the analytical value of testing invariance rather than assuming it. Regarding mediation, the re-specified model—which estimates the direct residual leadership–satisfaction paths rather than omitting them—shows that servant leadership reaches life satisfaction through complementary partial mediation: self-efficacy transmits a substantial portion of its influence (indirect β=0.217) while a significant direct pathway remains (β=0.251), plausibly reflecting unmodelled mechanisms such as organisational trust or psychological safety [11, 35]. Transformational leadership, in contrast, shows only a small indirect-only effect whose bootstrap interval barely excludes zero; its contribution to life satisfaction in this context is modest. Even so, empirically discerning an exclusively indirect mechanism is itself theoretically informative: transformational leadership’s influence on life satisfaction in this setting is subtle and channelled entirely through efficacy beliefs, a precision that models imposing mediation by construction cannot deliver. The serial pathway through harassment was not significant for either style, indicating that harassment reduction, although valuable in itself, is not the conduit through which leadership enhances life satisfaction in these data. The gender analysis constitutes a central contribution of the revised study. Measurement invariance was supported at every step (ΔCFI ≤0.003), licensing structural comparison; the formal likelihood-ratio test then rejected full structural equivalence (Δχ2(8)=17.86, p=0.022), with Wald tests localising the differences exclusively to the two transformational-leadership paths. The resulting picture is theoretically coherent: servant leadership and the self-efficacy → satisfaction mechanism operate as gender-equivalent resources, whereas transformational leadership operates through gender-specific pathways—protecting women against workplace harassment, consistent with evidence that women face disproportionate exposure to academic hostility [6], and strengthening efficacy beliefs among men, possibly through identification and role-modelling processes [14, 16]. These localised differences reframe the study’s contribution to Sustainable Development Goal 5: rather than asserting that one model fits all, the evidence identifies precisely where gender-responsive leadership development is needed. Cultivating transformational practices attentive to their harassment-preventive function is especially consequential for female faculty, while servant leadership can be promoted as a universally effective resource. In practical terms, gender-responsive development could take the form of training modules on inclusive leadership communication and active-bystander intervention that explicitly target the harassment-preventive function identified here. These findings map directly onto SDG target 5.1 (ending discrimination against women) and SDG target 16.6 (effective, accountable institutions). Together with SDG 16 (just institutions), these findings offer differentiated, data-driven guidance for equitable university management and for meeting educational quality and accreditation standards (e.g., SINEACE models in Peru). Limitations and future research The study presents several methodological limitations. First, the cross-sectional design precludes temporal precedence and therefore definitive causal claims: the modelled directionality is theory-driven, but reverse or bidirectional relations—for example, faculty well-being shaping how leadership is perceived—cannot be ruled out, nor can unmeasured confounders for which no design-based control exists; reliance on self-reported ordinal measures further introduces potential common method variance. Second, the non-probability convenience sample constrains external validity in specific ways: self-selection may over-represent faculty more engaged with well-being issues, and the concentration of participants in the coastal region (60.9%) means that the institutional and cultural conditions of highland and jungle universities are under-represented; future studies should employ stratified probability sampling across regions and institution types. Third, a measurement limitation qualifies H3: self-efficacy responses cluster near the top of the 0–6 scale (M=4.81), and this restricted upper range plausibly attenuated the observable harassment–efficacy association—an account reinforced by the outlier sensitivity analysis, in which the path attained significance; more sensitive measures, or longitudinal designs attuned to gradual decline, are needed. Fourth, the gender differences identified by the Wald tests emerge from eight simultaneous comparisons and do not survive a Bonferroni correction; they are protected only by the preceding omnibus test and must therefore be treated as exploratory findings requiring independent replication, as must the small indirect effect of transformational leadership detected only by the bootstrap criterion. We note that the discrepancy between the significant scaled Δχ2 and the supportive approximate fit criteria in the invariance analysis is not a limitation specific to this study but a well-documented property of chi-square–based difference tests in large ordinal models; reliance on approximate fit indices under these conditions is standard recommended practice [24, 25]. Future research should adopt longitudinal designs to isolate temporal effects; integrate macro-level institutional variables—for example, organisational justice perceptions or accreditation pressures—as potential moderators within multilevel SEM designs; and employ qualitative inquiry (in-depth interviews, focus groups) to illuminate why transformational leadership operates through gender-specific channels—women’s lived experience of harassment mitigation, men’s identification and role-modelling processes—and to probe the unmodelled mechanisms, such as psychological safety and organisational trust, that plausibly underlie servant leadership’s direct effect on life satisfaction. Conclusions This study establishes transformational and servant leadership as organisational resources for safeguarding faculty well-being and mitigating psychosocial hazards in higher education, within a single integrated structural model that estimates mediation empirically rather than by construction. Both leadership styles curtail workplace harassment. Servant leadership emerges as the dominant resource: it strongly cultivates teacher self-efficacy and reaches life satisfaction through complementary partial mediation, combining a substantial indirect pathway with a significant direct one. Transformational leadership contributes a small indirect-only effect on life satisfaction, and workplace harassment, while undesirable in itself, did not erode self-efficacy in this cross-sectional context. The multigroup analysis yields the study’s most policy-relevant finding: after establishing measurement invariance, full structural equivalence across gender was formally rejected, with the differences localised exclusively in the transformational-leadership paths. Transformational leadership protects women—but not men—against workplace harassment, and strengthens self-efficacy among men—but not women—whereas servant leadership and the central self-efficacy–satisfaction mechanism operate equivalently for both groups. These results provide differentiated, data-driven support for aligning university management with Sustainable Development Goals 5 (Gender Equality) and 16 (Peace, Justice, and Strong Institutions): equitable institutions require not uniform leadership prescriptions but gender-responsive ones, deploying transformational practices deliberately for their harassment-preventive value to female faculty while promoting servant leadership as a universally effective resource. From an institutional policy perspective, higher education administrations should move beyond strictly individual-level stress management and prioritise systemic, leadership-driven interventions calibrated to these differentiated mechanisms. A phased strategy is advisable: beginning with servant-leadership development, whose effects are universal across genders, and progressively incorporating gender-responsive transformational modules. Universities can further embed metrics of gender-responsive leadership into existing accreditation and quality frameworks (e.g., SINEACE in Peru), creating a tangible pathway for demonstrating progress toward SDGs 5 and 16. Given the documented organisational costs of unaddressed harassment—turnover, absenteeism, and diminished performance—such investment is plausibly cost-effective, an economic question that future programme evaluations should quantify. Future research should also track these dynamics longitudinally and probe the processes underlying the gender-specific transformational-leadership pathways identified here. Supplementary information Supplementary material Author contributions CGAM, ET-C, RIGM, JLL-G and JET-C contributed to the study conception and design, and material preparation. All authors read and approved of the final manuscript. Funding This research has no external funding. Data Availability The dataset and instrument used are available at the following \href{https://drive.google.com/file/d/1k-xynTMsOzfJQFScdBokVnFYirUxKp_t/view?usp=sharing}{link}. For detailed information about the instrument, once the article has been published, one can contact the authors. Declarations Ethics approval and consent to participate The Ethics Committee of the Graduate School of the Peruvian Union University granted formal approval (EPG number: 00281). Written informed consent was obtained virtually from each participant, in accordance with the Ethics Committee’s approval. This study was conducted in accordance with the ethical standards of the institutional research committee and the 1964 Declaration of Helsinki and its subsequent amendments. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Cristhian George Alave Mamani, Esteban Tocto-Cano and Ronny Ivan Gonzales Medina contributed equally to this work. Contributor Information Ronny Ivan Gonzales Medina, Email: rgonzalesm@ucsm.edu.pe. Josue Edison Turpo-Chaparro, Email: josuetc@upeu.edu.pe. Supplementary information The online version contains supplementary material available at https://doi.org/10.1186/s40359-026-05450-0. References Associated Data Supplementary Materials Supplementary material Data Availability Statement The dataset and instrument used are available at the following \href{https://drive.google.com/file/d/1k-xynTMsOzfJQFScdBokVnFYirUxKp_t/view?usp=sharing}{link}. For detailed information about the instrument, once the article has been published, one can contact the authors.