Abstract
Background: The Short Servant Leadership Scale (ELSVA; Spanish: Escala de Liderazgo de Servicio, Versión Abreviada) assesses self-reported servant-leadership values, but its cross-country measurement equivalence remains uncertain. Objective: To examine the internal structure, reliability, and ordinal measurement invariance of the ELSVA in 684 higher education students from Peru (n = 350) and Ecuador (n = 334). Methods: The 14 ordered items were evaluated using WLSMV confirmatory factor analysis, alternative factor specifications, and multigroup invariance tests, supplemented by sex- and age-related sensitivity analyses. Results: Residual covariances for LS4–LS5 and LS12–LS13 improved model fit, although robust fit remained suboptimal (Peru: CFI = 0.895, RMSEA = 0.110; Ecuador: CFI = 0.926, RMSEA = 0.114). Average variance extracted was 0.526 and 0.656, and model-based composite reliability was 0.914 and 0.953, respectively. Threshold equality across countries was rejected (adjusted Δχ2[56] = 144.61, p < 0.001); additional equal-loading restrictions were not rejected, conditional on threshold equality (Δχ2[13] = 17.93, p = 0.160). Conclusions: ELSVA responses describe interpretable leadership-related values in these samples; however, structural misfit and noninvariant thresholds do not support country-level mean comparisons or inferences about observed leadership behavior.
1. Introduction
In recent decades, interest has grown substantially in prosocial forms of leadership that are less centered on authority and individual benefit and more oriented toward collaboration, interpersonal development, and collective well-being (Eva et al., 2019; Spears, 2005). This approach is especially relevant in higher education, where professional training also encompasses ethical, collaborative, and socially responsible competencies (Villar-Guevara et al., 2025). Servant leadership is grounded in a fundamental disposition to serve and prioritizes followers’ needs and the well-being of individuals, organizations, and communities (Greenleaf, 1970; Eva et al., 2019).
It differs from transformational leadership because it regards followers’ development as an end in itself rather than as a means of achieving organizational objectives (van Dierendonck, 2011); from authentic leadership because it links self-awareness and self-regulation with genuine altruistic motivation (Avolio & Gardner, 2005); and from ethical leadership because it incorporates responsible stewardship, long-term development, and attention to multiple stakeholder groups (Brown & Treviño, 2006).
Among higher education students, the relevance of servant leadership does not depend on holding a formal leadership position, as curricular and extracurricular experiences provide opportunities to influence others’ participation and learning while directing that influence toward shared goals (Skalicky et al., 2020). Empirical evidence supports its importance: servant leadership is associated with self-efficacy and attitudes and intentions toward social entrepreneurship among young people (Rivera et al., 2017, 2018), perceived employability (Yue et al., 2024), and cooperative learning and satisfaction with one’s studies (Villar-Guevara et al., 2025).
The conceptual development of servant leadership has led to instruments with different scopes. These include the Servant Leadership Questionnaire (Barbuto & Wheeler, 2006), the SL-28 (Liden et al., 2008), the Servant Leadership Behaviour Scale (Sendjaya et al., 2008), and the Servant Leadership Survey (van Dierendonck & Nuijten, 2011), which differ in dimensions and sources of information (Sun et al., 2024). The 14-item Short Servant Leadership Scale is known by its Spanish acronym ELSVA (Escala de Liderazgo de Servicio, Versión Abreviada; Rivera et al., 2017). It is relevant to educational contexts because it asks about individual dispositions without requiring a formal leadership position or identifiable followers.
The ELSVA comprises 14 statements derived from the Servant Leadership Profile of Page and Wong (2000), subsequently examined by Dennis and Winston (2003) and adapted for Spanish-speaking adolescents and young adults by Rivera et al. (2017). The statements refer to keeping commitments and moral courage (LS1–LS2), service without reward and personal sacrifice (LS3–LS5), listening and developing others (LS6–LS7), purpose and goal setting (LS8–LS10), role modeling (LS11), appreciation of team members and their differences (LS12–LS13), and openness to criticism (LS14). These are substantive content areas within the short measure, not independently validated subscales in the present data. The exact administered Spanish items and English translations for readers are presented in Appendix A.
To the best of our knowledge, published structural evidence has not established that responses to the ELSVA are psychometrically equivalent across Latin American national contexts. Cross-cultural leadership research suggests that the meaning attributed to hierarchy, shared decision-making, empowerment, and responsibility to others can depend on local norms (Dickson et al., 2012; Mittal & Dorfman, 2012). Potentially context-sensitive content includes personal sacrifice (LS4–LS5), acceptance of others’ views (LS14), and inclusive teamwork (LS12–LS13); this is a theoretical concern rather than evidence that either country possesses a particular cultural characteristic. Indeed, cross-cultural equivalence was only partially supported for a different servant leadership instrument across eight countries (van Dierendonck et al., 2017). These observations motivate examination of item thresholds as well as factor loadings, rather than assuming that linguistic similarity guarantees measurement equivalence.
Peru and Ecuador share Spanish as the language of administration in this study and are geographically proximate, yet similar wording need not produce identical response-category use. The present study therefore evaluated the ELSVA’s internal structure, reliability, and ordinal invariance across samples, with sensitivity analyses addressing differences in their sex and age composition. No directional country difference in the latent trait was inferred or prespecified; the substantive question was which comparisons the evidence permits. Because these are self-reports, the study concerns endorsed leadership-related values, not empirically classified leadership profiles or observed behavior.
2. Materials and Methods
2.1. Design and Participants
A cross-sectional instrumental study was conducted (Ato et al., 2013). Participants were selected using nonprobability purposive sampling based on institutional access and voluntary participation (Otzen & Manterola, 2017). Each national sample contained more than 300 participants, which supports estimation of the planned factor models (Kline, 2016), but is not an effect-specific statistical-power calculation for detecting threshold or loading noninvariance. Inferences from nonsignificant difference tests are therefore interpreted cautiously, rather than as proof of equivalence.
The sample comprised 684 higher education students: Peru (n = 350, 76.9% women, 23.1% men, age M = 20.95 years, SD = 2.60, range = 18–34) and Ecuador (n = 334, 62.9% women, 37.1% men, age M = 22.95 years, SD = 4.82, range = 18–40). Table 1 presents the sociodemographic characteristics by country. Inclusion criteria were: (1) being an active undergraduate student at a higher education institution, (2) being 18 years of age or older, and (3) voluntarily agreeing to participate by providing informed consent. The exclusion criterion was having more than 10% missing data on the ELSVA items. After these criteria were applied, no missing data were identified across the 14 items analyzed. The available dataset did not record institution type (public/private) or participants’ city or region of origin; therefore, these variables are neither reported nor interpreted.
Table 1.
Sociodemographic characteristics of participants by country.
2.2. Instrument
The Short Servant Leadership Scale (ELSVA; Escala de Liderazgo de Servicio, Versión Abreviada; Rivera et al., 2017) comprises 14 positively worded statements with seven ordered response categories (1 = strongly disagree; 7 = strongly agree). Following verification with the field coordinators responsible for data collection in both countries, it was confirmed that the same Spanish wording of all 14 items and the same seven response options were administered in Peru and Ecuador, without country-specific changes to the instrument. The English translations in Appendix A are provided solely for reader comprehension and were not administered. Analytical codes LS1–LS14 correspond in order to the source items ELSVA1–ELSVA14. LS4 and LS5 address sacrificial service; LS12 and LS13 concern inclusion and constructive attention to group differences. Scores indicate self-reported endorsement, not directly observed servant-leadership behavior.
2.3. Procedure
The study followed the ethical principles of the Declaration of Helsinki (World Medical Association, 2024). The research originated as a faculty project affiliated with Universidad César Vallejo in Peru and was reviewed by its Research Ethics Committee of the School of Psychology before data collection (favorable decision dated 17 October 2025; Report No. 01770-2025/CEI-PSI). The approved binational project was entitled “Validación transcultural de la Escala Abreviada de Liderazgo de Servicio (ELSVA) en Jóvenes de Perú y Ecuador” [Cross-Cultural Validation of the Short Servant Leadership Scale (ELSVA) among Young People in Peru and Ecuador]. Because the project was institutionally administered by the Peruvian university and its approved scope explicitly included both countries, ethical review was conducted through that university’s committee for the binational protocol. Data were collected in person in Ecuador from 24 October to 28 November 2025 and in Peru from March to July 2026, in both instances after the favorable decision. Prior to participation, students were informed about the study aims and procedures, voluntary participation, confidentiality, and their right to withdraw without consequences; all provided informed consent. Responses were recorded using identification codes, and access to the collected data was restricted to the research team.
2.4. Data Analysis
Item means, standard deviations, skewness, and kurtosis were calculated by country; Mardia’s test was used to assess multivariate normality. Given seven ordered response categories, asymmetric item distributions, and evidence of multivariate nonnormality, confirmatory analyses treated all 14 indicators as ordered categorical. Estimation used weighted least squares with mean- and variance-adjusted statistics (WLSMV), theta parameterization, and ordered indicators in lavaan. Robust CFI, TLI, RMSEA (with 90% confidence intervals), and SRMR were considered jointly; conventional fit cutoffs (e.g., CFI/TLI near 0.95 and RMSEA near 0.06; Hu & Bentler, 1999) were regarded as contextual heuristics, not as mechanical decision rules established specifically for these ordinal models. Robust and scaled fit indices were identified separately because they are not interchangeable. Nested WLSMV models were compared using adjusted difference tests rather than direct subtraction of chi-square statistics. Continuous-indicator MLR estimates were retained only as sensitivity checks.
The theoretically established single-factor specification was examined first (M1). A parsimonious content-informed alternative (M2) additionally specified residual covariances LS4–LS5 (service and sacrifice) and LS12–LS13 (valuing team members and constructive use of their differences). These modifications were first explored in Peru and subsequently evaluated without adding new parameters in Ecuador. Further exploratory specifications introduced one additional residual covariance to M2 (M3: LS9–LS10; M4: LS8–LS9; M5: LS1–LS2). A bifactor sensitivity model included a general factor plus service (LS3–LS5) and vision (LS8–LS10) factors; a two-factor alternative was also examined. The initial modification process was data-informed, and the Ecuadorian analysis should be regarded as a sample-specific robustness check, not external validation. Model comparison therefore also considered admissibility, interpretable content, and reliability of specific factors rather than favoring maximal fit alone.
Ordinal multigroup analysis of M2 compared a configural model with successive threshold-equality, additional loading-equality, and item unique-variance-equality constraints (partial residual invariance), using the Wu–Estabrook identification framework in semTools. These tests differ from the intercept-based metric/scalar sequence used with continuous indicators. Adjusted WLSMV chi-square difference tests and consistently labeled scaled indices were reported for nested comparisons. The model included residual covariances LS4–LS5 and LS12–LS13, which remained freely estimated across countries throughout the reported sequence. Thus, the final step tested equality of item unique variances, rather than equality of the full residual covariance matrix. Exploratory partial-threshold sensitivity analyses released the thresholds of LS4, LS5, LS12, and LS13; because those models involved an alternative identification strategy and produced near-singular parameter-covariance warnings, partial invariance was not claimed. Pooled sex-group ordinal invariance tests and country-specific MIMIC models assessing associations with sex and age were sensitivity analyses, not formal item-level differential-item-functioning (DIF) tests.
Reliability was evaluated in the ordinal single-factor M1 and residual-covariance M2 models using model-based composite reliability (semTools::compRelSEM). Average variance extracted (AVE) was calculated with semTools::AVE and model-implied observed-variable variances (obs.var = FALSE). These statistics were compared across M1 and M2 because the introduction of local residual dependence can change their interpretation, even when global fit improves. Item-level descriptive distributions and standardized factor loadings were used to characterize item functioning. Coefficient alpha and corrected item–total correlations, which describe reliability and item associations at the raw-score level, were not available in the validated supplementary outputs and are therefore not presented as though independently verified. The exploratory bifactor models were also evaluated using general- and specific-factor reliability. All model-based indices remain conditional on the fitted structure and do not establish criterion, discriminant, or external convergent validity. In the absence of external correlates, validity claims were restricted to evidence based on internal structure and score interpretation (American Educational Research Association et al., 2014).
As an exploratory check of the operating characteristics of the ordinal invariance tests, parametric Monte Carlo diagnostics were run using the observed country sample sizes (Peru, n = 350; Ecuador, n = 334) and 200 replications under threshold-equality and threshold-noninvariance scenarios, together with conditional residual-variance scenarios. The primary simulation preserved seven ordered categories, but frequent empty categories made many replications non-estimable. A separate five-category diagnostic combined original response categories 1–3 into one lowest category to investigate numerical stability; the primary analyses retained the original seven categories throughout. This secondary exercise used approximate observed or pooled response margins and model-implied correlations. Before interpreting alternative-scenario rejection frequencies as power, null-scenario calibration and the fraction of usable replications were inspected. The conditional residual-variance contrasts presupposed threshold and loading equality, although country threshold equality had already failed in the original data. For these reasons, neither simulation provided a defensible formal estimate of power for the seven-category design. Statistical inferences were instead evaluated quantitatively against the observed adjusted difference tests, their degrees of freedom, and the conflicting approximate-fit indices.
All analyses and diagnostic simulations were conducted in R (version 4.4.2; R Core Team, 2024) for the documented supplementary runs, with lavaan (version 0.6-19) and semTools (version 0.5-6) used for the ordinal analyses and reliability evaluation (Rosseel, 2012). Continuous-indicator MLR analyses were retained as supplementary checks rather than as the basis of ordinal-invariance claims. Model adequacy was assessed jointly from fit indices, convergence, admissibility, parameter estimates, local item dependence, and theoretical interpretability; the Monte Carlo diagnostics were not used to overturn poor baseline fit or failed threshold invariance.
3. Results
3.1. Preliminary Item Analyses
The sample included 350 students from Peru and 334 from Ecuador (Table 1). Women represented 76.9% and 62.9% of the respective samples, while mean ages were 20.95 and 22.95 years. Table 2 reports the original item descriptives. Average endorsement was generally above the scale midpoint; in Peru, LS12 and LS13 had the highest means (6.02 and 6.00), whereas LS4 and LS5 had the lowest (4.95 and 4.99). In Ecuador, LS8 (5.62) and LS6 (5.61) were among the highest and LS4 (4.96) was the lowest. These descriptive contrasts concern response tendencies within nonrepresentative samples, not national differences in leadership ability. Mardia’s tests indicated multivariate nonnormality in both groups (Peru: skewness χ2[560] = 2712.93, p < 0.001; kurtosis z = 48.83; Ecuador: skewness χ2[560] = 4371.29, p < 0.001; kurtosis z = 97.65; both p < 0.001). The ordinal estimation approach was therefore used for primary inference.
Table 2.
Descriptive statistics for ELSVA items by country.
3.2. Confirmatory Factor Analysis
In the primary WLSMV analysis, the original one-factor model (M1) showed inadequate robust fit in Peru (CFI = 0.835, TLI = 0.805, RMSEA = 0.136, SRMR = 0.070) and Ecuador (CFI = 0.873, TLI = 0.849, RMSEA = 0.147, SRMR = 0.051). Allowing the content-related residual covariances LS4–LS5 and LS12–LS13 (M2) improved fit in both countries (Peru: robust CFI = 0.895, TLI = 0.873, RMSEA = 0.110, SRMR = 0.057; Ecuador: robust CFI = 0.926, TLI = 0.910, RMSEA = 0.114, SRMR = 0.040). Thus, relative to M1, robust CFI increased by 0.060 in Peru and 0.053 in Ecuador, while robust RMSEA decreased by 0.026 and 0.033, respectively. Nevertheless, the remaining misfit—particularly RMSEA above 0.10—limits the strength of any unidimensionality claim. Exploratory residual-covariance models M3–M5 offered additional but post hoc gains. The bifactor specification had the highest robust CFI among admissible candidate models (Peru = 0.935; Ecuador = 0.943), yet its specific factors showed weak reliability. The two-factor solutions were inadmissible and are not interpreted (Table 3). M2 was retained as a parsimonious, substantively intelligible working model for probing comparability, not as a demonstration of adequate absolute fit or an independently replicated factor structure.
Table 3.
Robust WLSMV fit indices for candidate factor models.
Under the ordinal M2 specification, standardized general-factor loadings were positive in both samples, ranging from 0.550 to 0.827 in Peru and from 0.733 to 0.873 in Ecuador (Table 4). In Peru, LS4 and LS5 had the smallest loadings (0.550 and 0.559), consistent with their shared service-specific residual content; in Ecuador, LS4 and LS11 had the smallest loadings (0.733 and 0.734). These loadings indicate the degree to which item responses covary with the general self-reported orientation under the fitted model. Their magnitude cannot establish that the corresponding behaviors are enacted in everyday settings.
Table 4.
Standardized general-factor loadings for M2 by country.
3.3. Measurement Invariance
For the WLSMV M2 country comparison, the configural model had robust CFI = 0.912 and robust RMSEA = 0.112, indicating appreciable baseline misfit. Equality of response thresholds was rejected relative to configural invariance (adjusted Δχ2[56] = 144.61, p < 0.001; Table 5). Conditional on the unsupported threshold-equality model, additional equality of loadings was not rejected (adjusted Δχ2[13] = 17.93, p = 0.160), whereas additional equality of item unique variances (partial residual invariance) was rejected (adjusted Δχ2[14] = 55.79, p < 0.001). Consequently, a nonsignificant increment for loadings must not be construed as evidence of full ordinal invariance or as permission to compare latent means. Although the scaled CFI rose from 0.961 (configural) to 0.966 (thresholds plus loadings) and the scaled RMSEA decreased from 0.109 to 0.085, the adjusted nested comparisons directly test equality constraints and gave a different pattern. Differences in scaled approximate-fit indices cannot negate significant adjusted difference tests, particularly given the poor robust configural baseline. Scaled and robust indices were kept separate; robust CFI/RMSEA were unavailable for the more constrained models and were not mixed into change scores.
Table 5.
Ordinal WLSMV measurement invariance of M2 between Peru and Ecuador.
In exploratory partial-threshold sensitivity models, freeing thresholds for LS5, LS12, or LS13 improved fit relative to full threshold equality, whereas freeing LS4 alone did not reach conventional significance. Even after jointly freeing LS4, LS5, LS12, and LS13, the resulting model remained distinguishable from the configural baseline (adjusted Δχ2[45] = 116.33, p < 0.001). Near-singular parameter-covariance warnings and an alternative identification strategy precluded a claim of established partial invariance. In the exploratory bifactor model, the service-specific reliability estimates were 0.296 in Peru and 0.152 in Ecuador, and the vision-specific estimates were 0.212 and 0.087, respectively. These low specific-factor values do not support separately scored service or vision subscales.
3.4. Sex and Age Sensitivity Analyses
Given the different sex distributions in Peru (76.9% women) and Ecuador (62.9% women), pooled sex-group ordinal invariance was explored. The configural model had robust CFI = 0.914 and robust RMSEA = 0.111. Adjusted difference tests did not reject equality of thresholds (Δχ2[56] = 40.92, p = 0.935) or the subsequent loading constraints (Δχ2[13] = 15.42, p = 0.282). These conditional findings must be interpreted alongside sparse response categories and baseline misfit. Separate country-level MIMIC models found no significant sex associations with the general factor (Peru: standardized β = 0.063, p = 0.280; Ecuador: β = −0.088, p = 0.112). Age was associated with a small positive difference in Ecuador (β = 0.130, p = 0.019) but not in Peru (β = 0.072, p = 0.214); the Ecuador result was not adjusted for multiple testing. MIMIC regressions did not test direct item-level DIF and cannot establish causal sex or age effects.
3.5. Extracted Variance, Model-Based Reliability, and Monte Carlo Diagnostics
Table 6 summarizes the additional reliability and extracted-variance results for ordinal M1 and M2. In Peru, AVE was 0.541 in M1 and 0.526 in M2, while model-based composite reliability declined from 0.940 to 0.914. In Ecuador, AVE declined from 0.678 to 0.656 and composite reliability from 0.971 to 0.953. All values were relatively high, but they describe model-dependent shared variance and score precision rather than independent convergent validity. The reductions after introducing residual covariances also show why better-fitting models should not automatically be regarded as having more reliable general-factor scores. Item-level mean, variability, and loading information are available in Table 2 and Table 4. Global robust fit of M2 remained suboptimal in both samples (Table 3).
Table 6.
Average variance extracted and composite reliability in ordinal models.
The original seven-category Monte Carlo exercise yielded usable results in only 86/200 (43%) threshold-null replications, 20/200 (10%) threshold-alternative replications, 98/200 (49%) conditional residual-null replications, and 59/200 (29.5%) conditional residual-alternative replications. These rates of failure preclude reliable power estimates. In a separate five-category diagnostic, all 200 replications per scenario were usable. Under its threshold-equality null, 8/200 tests rejected (4%); however, its conditional residual-variance null rejected in all 200 replications, revealing severe miscalibration of that scenario. These failures and the change in response-category structure prevent interpreting the observed 100% alternative-scenario rejections as evidence of 100% statistical power (Table 7). Quantitatively, the observed seven-category threshold comparison rejected equality with 56 degrees of freedom (adjusted Δχ2 = 144.61, p < 0.001), while the conditional loading contrast did not reject equality with 13 degrees of freedom (adjusted Δχ2 = 17.93, p = 0.160). The latter finding establishes neither adequate detection probability for small departures nor equivalence. Because threshold invariance failed and robust baseline fit was weak, ΔCFI and ΔRMSEA changes were not treated as proof of invariant measurement.
Table 7.
Exploratory Monte Carlo diagnostics: estimation feasibility and rejection frequencies.
4. Discussion
The findings qualify, rather than simply confirm, the cross-national comparability of the ELSVA. Its 14 statements form a coherent general self-reported orientation toward ethical commitment, service, attentive listening, developing others, purposeful action, and inclusive teamwork. Nevertheless, the primary ordinal analysis identified residual item dependence and inadequate robust fit for the simplest specifications. The central comparative result was a rejection of equal response thresholds between Peru and Ecuador, even though additional equality constraints on loadings were not rejected after the thresholds were fixed. Thus, similar relationships of items with the general factor cannot be equated with identical use of response categories or with directly comparable latent means.
The content of the locally dependent pairs helps interpret the findings. LS4 (willingness to make personal sacrifices to serve) and LS5 (preferring to serve rather than be served) share a self-sacrificial service emphasis. LS12 (valuing each team member) and LS13 (using interpersonal differences constructively) both concern inclusive collaboration. Correlated residuals for these pairs may reflect overlapping wording or narrower common content beyond the general factor. They should not be interpreted as proof of new dimensions. LS8–LS10 similarly connect life purpose, goal setting, and effective ideas, but their conceptual proximity does not alone justify additional correlated errors. In particular, the higher fit of the bifactor model does not support independent service and vision scores because the corresponding specific-factor reliability was very low.
The response pattern also adds a substantive perspective beyond the numerical fit statistics. In both samples, LS4 (willingness to sacrifice for others) was among the least strongly endorsed items, whereas in Peru LS12–LS13 (valuing team members and making constructive use of differences) showed high endorsement. This pattern is compatible with the possibility that cooperative values are easier to endorse than demanding self-sacrificial service, two emphases recognized within servant-leadership theory (Greenleaf, 1970; van Dierendonck, 2011). It does not establish distinct behavioral leadership profiles: item responses may reflect social desirability, response-category conventions, and the characteristics of these convenience samples. In conjunction with the weaker robust fit in Peru, the results support a cautious interpretation of one broad disposition while questioning whether it captures all item content equally well in both settings.
Cross-national interpretation is further restricted by threshold noninvariance and poor configural fit. The confirmed use of identical Spanish item wording and seven response options in both countries makes known differences in the administered instrument an unlikely explanation for the observed threshold noninvariance; nevertheless, differences in response styles, contextual interpretation, and sample composition remain plausible and cannot be disentangled here. Different response thresholds imply that students with the same latent standing may select different categories. Consequently, raw-score and latent-mean contrasts cannot be attributed unambiguously to genuine differences in servant-leadership dispositions, still less to national cultures or leadership ability. The exploratory partial-threshold models did not establish sufficient comparability to relax this restriction. This negative result remains substantively informative: shared language and geographic proximity do not guarantee response-scale equivalence for values involving service, commitment, and teamwork.
Sex-related sensitivity analyses address an important sampling difference: women constituted 76.9% of the Peruvian sample but 62.9% of the Ecuadorian sample. In pooled ordinal multigroup models, neither threshold constraints (adjusted Δχ2[56] = 40.92, p = 0.935) nor additional loading constraints (Δχ2[13] = 15.42, p = 0.282) were rejected; however, the robust configural RMSEA was 0.111, and some item-by-subgroup response categories were sparse. Country-specific MIMIC analyses showed small nonsignificant sex associations with the general factor (Peru: standardized β = 0.063, p = 0.280; Ecuador: β = −0.088, p = 0.112). The age association was small and positive in Ecuador (β = 0.130, p = 0.019), but not in Peru (β = 0.072, p = 0.214). These associations are exploratory, do not establish absence of item-level differential functioning, and should not be interpreted causally. The combined evidence does not support a strong claim of sex differences in servant leadership.
The additional reliability analyses help distinguish score precision from structural validity. Although M1 and M2 yielded AVE values above 0.50 and composite reliability above 0.90, the decrease in both indices after residual-covariance modification, together with persistent model misfit, underscores the limitations of interpreting numerical reliability as a sufficient endorsement of the scale. Exploratory Monte Carlo work likewise cautions against overconfident inference from nonsignificant invariance contrasts: the principal seven-category simulations had extensive failed replications, while a collapsed-category diagnostic exhibited invalid null behavior for conditional residual invariance. The statistically significant threshold contrast remains as evidence against exact threshold equality in these samples; by contrast, the nonsignificant additional-loading comparison cannot establish equivalence without adequate model fit and sensitivity information.
4.1. Implications
For research and education, the ELSVA can organize discussion of self-reported values relevant to student leadership, provided the limited model fit is reported and the items are not treated as validated separate subscales. Its content spans keeping commitments and moral courage (LS1–LS2), service and sacrifice (LS3–LS5), attentive listening and development of others (LS6–LS7), personal purpose and planning (LS8–LS10), role modeling (LS11), inclusive teamwork (LS12–LS13), and receptivity to criticism (LS14). The contrast between greater endorsement of some cooperative statements and lower endorsement of sacrifice may suggest themes for reflection in leadership education; it cannot determine which students actually demonstrate these behaviors or identify stable leader types.
The principal applied implication is caution in comparative interpretation. Cross-national studies should assess threshold functioning before comparing ELSVA means, and future replication should evaluate the LS4–LS5 and LS12–LS13 pairs in independent samples. Educational programs may use these statements to stimulate reflection on service, teamwork, and moral responsibility, but the present study does not demonstrate that endorsing them predicts leadership behavior, professional outcomes, or later leadership effectiveness. Additional behavioral, peer-rated, and longitudinal evidence would be needed for those conclusions.
4.2. Limitations
Several limitations qualify these results. Nonprobability sampling, unequal sex distributions (76.9% women in Peru versus 62.9% in Ecuador), and different mean ages (20.95 versus 22.95 years) restrict generalization and complicate national comparisons. Data were gathered in nonoverlapping periods (24 October–28 November 2025 in Ecuador and March–July 2026 in Peru), so cohort and period effects cannot be separated from country effects. Institution type, city, and region were not recorded. Reliance on self-report and positively worded items may introduce social-desirability and response-style biases, while external convergent, discriminant, predictive, and behavioral variables were unavailable. Despite comparatively high AVE and model-based reliability, robust structural fit was limited and the configural baseline was poor. Country threshold equality was rejected, and exploratory partial-threshold solutions produced near-singular covariance warnings. Equal residual covariances were not independently tested across groups. Invariance analyses by sex were pooled across countries; country-specific MIMIC regressions did not constitute item-level DIF testing or full invariance by age. In the additional Monte Carlo work, only 10–49% of seven-category runs were usable and the five-category residual-null comparison was demonstrably miscalibrated, precluding reliable effect-specific power estimation. Thus, neither the nonsignificant conditional-loading test nor the pooled sex-group tests can be interpreted as proof of measurement equivalence or sensitivity to small departures. Raw-score alpha and corrected item–total statistics were not independently verified in the available supplementary results and should be examined in subsequent replications. Finally, the cross-sectional design does not support causal or developmental conclusions.
5. Conclusions
The ELSVA captures an interpretable set of self-reported values related to service, ethical commitment, attentive listening, purposeful action, and inclusive collaboration among the participating students. Descriptively, valuing cooperation was endorsed more strongly than demanding personal sacrifice, which may inform discussion of leadership-related values in higher education but does not identify observable leadership behaviors or stable leader profiles. Model-based AVE and reliability were comparatively high, yet robust fit of the tested factor structures remained suboptimal. Country threshold invariance was rejected, and the nonsignificant additional loading test was conditional on a threshold model already rejected by the data. Cross-country comparisons of observed totals or latent means therefore are not psychometrically justified. Analyses by sex and age were exploratory and do not eliminate demographic confounding or item-level differential functioning. The chief contribution is to clarify both the substantive content and the limits of the scale’s comparability in these particular samples. Future research should examine more adequate factor specifications, independently verify local dependencies, collect behavioral and external validation criteria, use more representative and temporally comparable samples, and evaluate DIF and invariance with stable statistical sensitivity.
Author Contributions
Conceptualization, G.P.-E. and W.C.M.-G.; methodology, G.P.-E. and W.C.M.-G.; software, W.C.M.-G.; validation, G.P.-E., L.M.H.-B., R.E.C.-L., L.M.-P. and W.C.M.-G.; formal analysis, W.C.M.-G.; investigation, G.P.-E., L.M.H.-B., R.E.C.-L. and L.M.-P.; resources, G.P.-E., R.E.C.-L. and W.C.M.-G.; data curation, G.P.-E. and W.C.M.-G.; writing—original draft preparation, G.P.-E. and W.C.M.-G.; writing—review and editing, G.P.-E., L.M.H.-B., R.E.C.-L., L.M.-P. and W.C.M.-G.; visualization, W.C.M.-G.; supervision, W.C.M.-G.; project administration, G.P.-E. and W.C.M.-G. All authors have read and agreed to the published version of the manuscript.
Funding
This research received no external funding.
Institutional Review Board Statement
The study was conducted in accordance with the Declaration of Helsinki and received a favorable decision from the Research Ethics Committee of the School of Psychology, Universidad César Vallejo (Report No. 01770-2025/CEI-PSI; 17 October 2025). The committee reviewed the faculty research project entitled “Validación transcultural de la Escala Abreviada de Liderazgo de Servicio (ELSVA) en Jóvenes de Perú y Ecuador”, whose approved scope encompassed research in both Ecuador and Peru. The review was administered through the Peruvian institution because the project originated under its faculty research program. Fieldwork began after approval: Ecuador, 24 October–28 November 2025; Peru, March–July 2026.
Informed Consent Statement
Informed consent was obtained from all subjects involved in the study. Participation was voluntary, and participants were informed about the study objectives and procedures, confidentiality, and their right to withdraw from the study without consequences.
Data Availability Statement
The data and analysis scripts that support the findings of this study are available from the corresponding author upon reasonable request, subject to compliance with institutional ethical guidelines and participant confidentiality protections.
Conflicts of Interest
The authors declare no conflicts of interest.
Appendix A. Item Content of the ELSVA
Table A1.
Original Spanish Item Wording and English Translation.
References
- American Educational Research Association, American Psychological Association & National Council on Measurement in Education. (2014). Standards for educational and psychological testing. American Educational Research Association. [Google Scholar]
- Ato, M., López-García, J. J., & Benavente, A. (2013). Un sistema de clasificación de los diseños de investigación en psicología. Anales de Psicología, 29(3), 1038–1059. [Google Scholar] [CrossRef] [Scilit]
- Avolio, B. J., & Gardner, W. L. (2005). Authentic leadership development: Getting to the root of positive forms of leadership. The Leadership Quarterly, 16(3), 315–338. [Google Scholar] [CrossRef] [Scilit]
- Barbuto, J. E., Jr., & Wheeler, D. W. (2006). Scale development and construct clarification of servant leadership. Group & Organization Management, 31(3), 300–326. [Google Scholar] [CrossRef] [Scilit]
- Brown, M. E., & Treviño, L. K. (2006). Ethical leadership: A review and future directions. The Leadership Quarterly, 17(6), 595–616. [Google Scholar] [CrossRef] [Scilit]
- Dennis, R. S., & Winston, B. E. (2003). A factor analysis of Page and Wong’s servant leadership instrument. Leadership & Organization Development Journal, 24(8), 455–459. [Google Scholar] [CrossRef] [Scilit]
- Dickson, M. W., Castaño, N., Magomaeva, A., & den Hartog, D. N. (2012). Conceptualizing leadership across cultures. Journal of World Business, 47(4), 483–492. [Google Scholar] [CrossRef] [Scilit]
- Eva, N., Robin, M., Sendjaya, S., van Dierendonck, D., & Liden, R. C. (2019). Servant leadership: A systematic review and call for future research. The Leadership Quarterly, 30(1), 111–132. [Google Scholar] [CrossRef] [Scilit]
- Greenleaf, R. K. (1970). The servant as leader. Center for Applied Studies. [Google Scholar]
- Hu, L.-T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55. [Google Scholar] [CrossRef] [Scilit]
- Kline, R. B. (2016). Principles and practice of structural equation modeling (4th ed.). The Guilford Press. [Google Scholar]
- Liden, R. C., Wayne, S. J., Zhao, H., & Henderson, D. (2008). Servant leadership: Development of a multidimensional measure and multi-level assessment. The Leadership Quarterly, 19(2), 161–177. [Google Scholar] [CrossRef] [Scilit]
- Mittal, R., & Dorfman, P. W. (2012). Servant leadership across cultures. Journal of World Business, 47(4), 555–570. [Google Scholar] [CrossRef] [Scilit]
- Otzen, T., & Manterola, C. (2017). Técnicas de muestreo sobre una población a estudio. International Journal of Morphology, 35(1), 227–232. [Google Scholar] [CrossRef] [Scilit]
- Page, D., & Wong, P. T. P. (2000). A conceptual framework for measuring servant-leadership. In S. B.-S. K. Adjibolosoo (Ed.), The human factor in shaping the course of history and development (pp. 69–110). University Press of America. [Google Scholar]
- R Core Team. (2024). R: A language and environment for statistical computing (Version 4.4.2) [Computer software]. R Foundation for Statistical Computing. Available online: https://www.R-project.org/ (accessed on 15 July 2026).
- Rivera, R. G., Santos, D., & Martín-Fernández, M. (2017). Spanish validation of the Servant Leadership Short Scale for young adults and adolescents. Revista de Psicología Social, 32(2), 395–423. [Google Scholar] [CrossRef] [Scilit]
- Rivera, R. G., Santos, D., & Martín-Fernández, M. (2018). Predicting attitudes and behavioural intentions towards social entrepreneurship: The role of servant leadership in young people. International Journal of Social Psychology, 33(3), 650–681. [Google Scholar] [CrossRef] [Scilit]
- Rosseel, Y. (2012). lavaan: An R package for structural equation modeling. Journal of Statistical Software, 48(2), 1–36. [Google Scholar] [CrossRef] [Scilit]
- Satorra, A., & Bentler, P. M. (2001). A scaled difference chi-square test statistic for moment structure analysis. Psychometrika, 66(4), 507–514. [Google Scholar] [CrossRef] [Scilit]
- Sendjaya, S., Sarros, J. C., & Santora, J. C. (2008). Defining and measuring servant leadership behaviour in organizations. Journal of Management Studies, 45(2), 402–424. [Google Scholar] [CrossRef] [Scilit]
- Skalicky, J., Warr Pedersen, K., van der Meer, J., Fuglsang, S., Dawson, P., & Stewart, S. (2020). A framework for developing and supporting student leadership in higher education. Studies in Higher Education, 45(1), 100–116. [Google Scholar] [CrossRef] [Scilit]
- Spears, L. C. (2005). The understanding and practice of servant-leadership. International Journal of Servant-Leadership, 1(1), 29–45. [Google Scholar] [CrossRef] [Scilit]
- Sun, P. Y. T., Anderson, M. H., & Gang, H. (2024). Determining the hierarchical structure and nature of servant leadership. Journal of Business and Psychology, 39, 715–734. [Google Scholar] [CrossRef] [Scilit]
- van Dierendonck, D. (2011). Servant leadership: A review and synthesis. Journal of Management, 37(4), 1228–1261. [Google Scholar] [CrossRef] [Scilit]
- van Dierendonck, D., & Nuijten, I. (2011). The servant leadership survey: Development and validation of a multidimensional measure. Journal of Business and Psychology, 26(3), 249–267. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- van Dierendonck, D., Sousa, M., Gunnarsdóttir, S., Bobbio, A., Hakanen, J., Pircher Verdorfer, A., Duyan, E. C., & Rodríguez-Carvajal, R. (2017). The cross-cultural invariance of the Servant Leadership Survey: A comparative study across eight countries. Administrative Sciences, 7(2), 8. [Google Scholar] [CrossRef] [Scilit]
- Villar-Guevara, M., García-Salirrosas, E. E., & Fernández-Mallma, I. (2025). Teachers’ effectiveness, cooperative learning, and servant leadership: Examining the mediating role of satisfaction with studies in a Latin American university sample. Frontiers in Education, 10, 1666870. [Google Scholar] [CrossRef] [Scilit]
- World Medical Association. (2024). WMA Declaration of Helsinki—Ethical principles for medical research involving human participants. World Medical Association. Available online: https://www.wma.net/policies-post/wma-declaration-of-helsinki/ (accessed on 15 October 2025).
- Yue, X., Zou, L., Ye, Y., & Cai, T. (2024). The impact of undergraduates’ servant leadership on their employability. SAGE Open, 14(2), 21582440241260936. [Google Scholar] [CrossRef] [Scilit]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.