Much of the extant leadership research is grounded in Western contexts, privileging approaches that assume universality. This is generally called an etic approach, which according to Berry (1969), is generally designed for broad applicability across cultures, assuming that a phenomenon can be understood and measured using standardised frameworks. While such frameworks provide valuable insights, they often lack contextual validity in culturally diverse settings such as South Africa (Grobler & Singh, 2018; Van der Westhuizen & Hewitt, 2021). Because of the shortcomings of the etic approach, scholars have increasingly called for research using emic research designs and methodologies, described by Pike (1967) as the quest to understand phenomena as they are experienced and expressed within specific cultural contexts, highlighting culturally embedded meanings, practices and values. This approach is supported within the leadership domain, as its application is rooted in local values, worldviews and lived experiences; all essential to better reflect the realities of a specific context and, for the purpose of this study, of African leadership (Iyioke, 2017; Punnett, 2017; Vilakati & Schurink, 2021).

The purpose of this study is to provide a nuanced, contextually relevant, valid and reliable instrument to measure African leadership, specifically from a South African perspective.

A relational-oriented leadership approach emphasises the importance of positive interpersonal relationships between leaders and followers, characterised by mutual respect, trust and social exchange (Höddinghaus et al., 2024). It further prioritises people over tasks, with a focus on collective engagement, psychological well-being and shared purpose.

The dominant relational leadership constructs (the term construct being used throughout this article to avoid an irrelevant discourse about whether it is a paradigm, theory, model or a framework), grouped together by Antonakis and Day (2018), are transformational, authentic and include servant leadership. They regard the emphasis on charisma, authenticity, vision, ethics and human connection as their shared characteristics. Short descriptions of each of these leadership constructs are provided as follows.

Transformational leadership is defined as a process of motivating followers by aligning group goals with personal development, using four key mechanisms, namely charisma, inspirational motivation, intellectual stimulation and individualised consideration (Bass, 1985; Cardona, 2000). This construct emphasises collective achievement and acknowledges followers as individuals, aligning with African leadership values that promote group solidarity and individual recognition (Grobler & Singh, 2018).

Empirical studies have linked transformational leadership to numerous positive organisational outcomes, including increased motivation, engagement and organisational fit (Bakker et al., 2023; Grobler & Grobler, 2023). Additionally, Grobler and Grobler (2023) found significant relationships between transformational leadership and variables such as psychological capital, proactive work behaviour, organisational support and reduced turnover intentions.

Authentic leadership, also regarded to be relational, is grounded in self-awareness, transparency, ethical conduct and consistency between values and behaviour (Walumbwa et al., 2008; Zehndorfer, 2013). It involves recognising and leveraging followers’ strengths and fostering trust-based relationships (Smith, 2020). Grobler and Powell (2018) conceptualise authentic leadership as the alignment of personal values with leader behaviour and decision-making based on objective, inclusive analysis.

Empirical findings support the positive effects of authentic leadership on individual and organisational outcomes. Grobler and Grobler (2024) found statistically significant associations between authentic leadership and outcomes such as employee engagement, organisational identification, psychological empowerment and lower turnover intention.

Ubuntu leadership, a distinctly African construct, embodies the Afrocentric philosophical ethos of humanness, emphasising respect, compassion, interconnectedness and collective well-being (Bhengu, 1996; Sindane, 2011). Msila (2022) and Powell (2023) identify UL behaviours such as fostering teamwork, encouraging shared learning, nurturing personal and professional growth and promoting ethical engagement grounded in dignity and mutual respect.

From an emic perspective, UL is not only a normative framework but also a practical, culturally embedded leadership construct. Grobler and Singh (2018) position it as a contextual expression of Afrocentric leadership that privileges human connection over authority or material gain. This is echoed by Bent-Goodley et al. (2017), who also call on Afrocentric scholars to centre the voices and lived experiences of African people in generating leadership knowledge.

Organisational leadership was studied by Grobler and Enslin (2025) from an emic perspective, where they identified emotional awareness (also a core element of authentic leadership; Walumbwa et al. [2008]), as the primary driver of effective OL. Secondary drivers include leadership style, personal characteristics, organisational culture, communication and vision, all of which relate to interpersonal exchange and relational processes. These elements contribute to outcomes such as leader support and team dynamics, which in turn facilitate collective goal attainment. Organisational leadership thus reflects a relationally embedded approach that integrates individual, group and systemic dimensions, with the primary strategic outcome being the successful execution of organisational objectives.

In summary, relational leadership constructs, including the four used in this study, share a common focus on values such as trust, community, ethical conduct and mutual development. Ubuntu and OL offer a culturally grounded, context-sensitive framework that affirms the humanistic foundations of African organisational life (Bolden & Kirk, 2011; Zondo, 2022), and African leadership, as discussed next.

African leadership, as conceptualised by Toendepi and Cele (2024), is grounded in humanism, collectivism and communal responsibility. It prioritises shared vision, collaboration, respect and dignity. These are all values rooted in the Ubuntu philosophy, which views individuals as fundamentally interconnected. Similarly, Mbigi (2005) emphasises dialogue, collective experience and critical engagement as central to African leadership. These humanistic and relational principles contrast with individualistic, task-driven models often associated with Western leadership theories.

There is a clear convergence of broad themes across the leadership constructs included in this study. Despite originating from different cultural and theoretical backgrounds, transformational, authentic, Ubuntu, OL and collectively African models share communalities around relationality, humanism and developmental orientation. All of them suggest a common concern for the ethical, emotional and communal dimensions of leadership. It is therefore hypothesised that African leadership encompasses key elements of both emic and etic paradigms, reflecting a hybrid relational model that bridges culturally specific values with broader leadership theories.

The core element of the four leadership constructs is the emphasis on interpersonal relationships. Transformational leadership fosters trust and inspiration through individualised consideration (Bass, 1985), while authentic leadership builds relationships through transparency and value congruence (Walumbwa et al., 2008). Ubuntu and African leadership similarly prioritise interconnectedness, mutual respect and community cohesion (Sindane, 2011; Toendepi & Cele, 2024), aligning with servant leadership’s focus on serving others and empowering followers.

These constructs further centre around ethical behaviour and specifically dignified treatment of individuals. Authentic leadership is further grounded in ethical decision-making and consistency between values and actions (Zehndorfer, 2013), a principle echoed in UL’s emphasis on humanness and respect (Bhengu, 1996). African leadership, in turn, integrates these values into a broader communal ethic that honours shared goals over individual gain (Mbigi, 2005).

Leadership across all four constructs is further seen as a developmental process. Transformational leadership seeks to elevate followers through intellectual stimulation and personal recognition (Bass & Avolio, 1994). Authentic leadership promotes follower growth by acknowledging their strengths (Smith, 2020), and UL encourages personal and team development through shared learning and inclusive dialogue (Powell, 2023). This developmental focus is also central to OL, which integrates emotional awareness, team dynamics and vision to achieve strategic outcomes (Grobler & Enslin, 2025).

Emotional awareness and trust are common drivers of effective leadership. Organisational leadership identifies emotional awareness as its primary foundation (Grobler & Enslin, 2025), consistent with both authentic and transformational leadership. Ubuntu leadership similarly draws on emotional intelligence through compassion, empathy and understanding.

Cultural integration and contextual sensitivity are also central themes across the leadership constructs. While the etically developed constructs of transformational and authentic leadership are often framed as universal, their primary focus resonates with the culturally grounded, emic constructs of Ubuntu, organisational and ultimately African leadership. The shared elements include trust, dignity, ethical behaviour and a people-centred approach. This overlap suggests a hybrid model of leadership in the African contexts, integrating both indigenous (emic) and globally recognised (etic) perspectives to form a more context-sensitive and inclusive understanding of leadership (Bolden & Kirk, 2011; Grobler & Singh, 2018).

To explore the underlying factor structure combining all 67 items from the four instruments, an EFA was conducted. The item-to-respondent recommended minimum ratio is 1:20 according to Meyers et al. (2016). Sampling adequacy was further confirmed through the Kaiser–Meyer–Olkin (KMO) measure, with a critical value of > 0.70 and Bartlett’s test of sphericity (p < 0.01) (Hair et al., 2019). An oblimin rotation was applied to allow factors to correlate, facilitating the interpretation of interrelated constructs.

Decisions on factor retention were guided by the Kaiser criterion (eigenvalues > 1), with a minimum cumulative explained variance of 50% (Hair et al., 2019). The number of factors retained was further based on the visual inspection of the scree plot and the Monte Carlo principal component analysis (PCA) for parallel analysis.

Cronbach’s alpha coefficients (α) were calculated to assess the internal consistency or reliability of the factors, with a critical value of α > 0.70 (Nunnally & Bernstein, 1994). Convergent validity of the measurement items was evaluated using composite reliability (CR) and average variance extracted (AVE), with acceptable thresholds of CR > 0.70 and AVE > 0.50, respectively. Discriminant validity was assessed by comparing the AVE with the maximum shared variance (MSV) and the average shared variance (ASV). Evidence of discriminant validity was established when both MSV and ASV values were lower than the corresponding AVE for each construct (Hair et al., 2019).

Four conceptual models were tested using CFA within Amos, employing maximum likelihood estimation:

Model^a: unidimensional (one-factor) model, positing a single overarching construct.

Model^b: first-order model, where observed variables (items) load directly onto latent variables (first-order factors).

Model^c: second-order model, where the first-order latent factors load on second-order latent factor.

Model^d: bifactor model, in which a general factor alongside orthogonal subfactors account for residual variance specific to each of the latent or first-order factors.

Model fit was assessed using multiple indices, namely chi-square goodness-of-fit (χ², with p < 0.05 indicating model misfit), comparative fit index (CFI), Tucker–Lewis index (TLI), root mean square error of approximation (RMSEA) and standardised root mean square residual (SRMR). Following Byrne (2016), the following criteria were applied as guidelines for acceptable fit: CFI ≥ 0.90, TLI ≥ 0.95, RMSEA ≤ 0.05, SRMR ≤ 0.08 and χ²/degrees of freedom (df) ratio less than 5.00, acknowledging the sensitivity of Chi-square to large sample sizes. In addition, the Akaike information criterion (AIC) was used for comparative model evaluation, with the model exhibiting the lowest AIC value considered to be of superior fit.

To explore the dimensionality of the overarching construct, a bifactor model was applied. Several indices were used to assess the strength of the general and group factors, determining whether a distinct first-order group-factor structure exists or if a unidimensional model is more appropriate (Rodriguez et al., 2016).

These indices include Omega (ω), which is a model-based reliability estimate for the total factor; Omega Subscale (ωS), indicating the reliability for specific subfactors; Omega Hierarchical (ωH), which represents the proportion of systematic variance in total scores explained by the general factor; Omega Hierarchical Subscale (ωHS), which is an indication of the proportion of reliable variance in subfactor scores after accounting for the general factor; Factor Determinacy (FD), indicating how well estimated factor scores represent the true latent variables; Construct Replicability (H), reflecting the correlation between a factor and an optimally weighted item composite; Explained Common Variance (ECV), which represents the proportion of reliable variance attributable to the general factor; and lastly, the PUC, which is the proportion of item correlations influenced solely by the general factor.

An ωH value above 0.80 typically supports the use of a unidimensional model. For subfactors, ωHS quantifies the unique reliable variance after removing the influence of the general factor. When ω is high and the difference between ω and ωH is small, it suggests that the general factor accounts for most of the systematic variance, and multidimensionality resulting from content domains is minimal. Conversely, a large gap between ω and ωH indicates that multidimensionality may complicate the interpretation of the total score.

Percentage of Uncontaminated Correlations and FD serve distinct but complementary roles in evaluating model quality. Percentage of Uncontaminated Correlations help the researcher to decide if total scores are valid, with a high PUC (> 0.80) supporting unidimensionality. The FD assesses the reliability of factor score estimates, with a value > 0.90 indicating factor scores that are sufficiently reliable for further analysis (Gorsuch, 1983). It is, however, important to observe that both FD and H are influenced by the number of items defining a factor and the magnitude of their loadings. A small number of items with high loadings can inflate these indices.

Lastly, ECV, as further index used in the bifactor analysis, is a valuable tool for Structural Equation Modeling (SEM) researchers deciding how to specify their measurement models, specifically the degree of unidimensionality. An ECV > 0.70 is an indication that the general factor is strong enough and that a unidimensional model may yield unbiased estimates of the general factor loadings, even in multidimensional data (Stucky & Edelen, 2014).

Ethical approval for this study was granted by the institutional research ethics committee (No. 2022_SBL_AC_004_CA), with full adherence to the institution’s ethical guidelines. Permission was obtained from participating organisations, informed consent was secured from all participants prior to data collection, and permission was granted by all parties (including the MBA and MBL students as co-researchers) to use the data for further, advanced research purposes by the primary researcher (the author of this article). This research did not receive any specific funding from public, commercial or non-profit agencies.

Item-level descriptive statistics were examined, including means, SDs, skewness and kurtosis. Item means on the 5-point Likert scale ranged from 2.63 (SD = 1.17) for OL18 (reverse-coded) to 3.44 (SD = 1.07) for Trf14. Standard deviations varied between 0.92 (AL2) and 1.23 (Trf7), indicating acceptable dispersion around the mean, as all values fell within the acceptable range (SD ≤ 2). Skewness values ranged from −1.02 (UL3) to 0.58 (OL18), while kurtosis ranged from −1.17 (TfL15) to 1.17 (AL2), all within the recommended thresholds for normality (skewness < 2.00, kurtosis < 7.00, West et al., 1995), confirming an approximately normal distribution of the data.

Initially, Pearson correlation analyses were conducted to examine relationships between the four original leadership constructs. This was deemed necessary to assess their convergence as relational leadership constructs, thus supporting the integrative strategy to be followed by this study. These results are presented in Table 1.

An examination of Table 1 reveals strong positive correlations (r > 0.70, Pallant, 2020) among all leadership constructs, providing empirical support for investigating their commonalities and overlaps, based on their hypothesised associations. This justifies the subsequent analysis aimed at exploring the feasibility of developing a unified measurement scale through EFA, CFA and bifactor modelling.

The item-to-respondent ratio of 1:24 exceeds the recommended threshold suggested by Meyers et al. (2016), supporting the adequacy of the sample size for factor analysis. Further support is provided by the KMO measure of sampling adequacy, which yielded a value of 0.98, well above the minimum acceptable level of 0.70 and Bartlett’s test of sphericity, which was significant (χ² [2211] = 68 611, p < 0.001), indicating suitability for factor extraction.

The EFA yielded a three-factor solution, consisting of 34 items, with 11 from OL, 10 from Ubuntu and authentic leadership each and 3 from transformational leadership, explaining 57% of the total variance in African leadership (surpassing the 50% benchmark; Meyers et al., 2016). The number of factors retained was corroborated by Cattell’s scree test, which identified a clear inflection point consistent with a three-factor model and confirmed by the PCA.

The first factor (from now on referred to as subfactor), labelled African Team Leadership, comprises 15 items (67% from the UL construct, followed by OL and transformational leadership with 27% and 6%, respectively) and reflects collaborative leadership behaviours, exemplified by the item: ‘My leader encourages collaboration in achieving our goals’. The second subfactor, termed Adverse African Leadership, includes 6 negatively worded items (all OL), focused on the leadership practices to avoid in the African context such as: ‘My leader does not act with integrity’. The third subfactor, Authentic African Leadership, consists of 13 items (with 77% from authentic leadership, 15% OL and 8% transformational leadership) reflecting self-awareness and genuineness, exemplified by the item: ‘My leader shows that he/she understands his/her strengths and weaknesses’.

Descriptive statistics and psychometric properties of the three subfactors and the general factor, African leadership (where appropriate), are detailed in Table 2.

The descriptive statistics presented in Table 2 indicate that African Team Leadership (M = 3.71, SD = 0.77), Authentic African Leadership (M = 3.59, SD = 0.71) and the general African Leadership factor (M = 3.63, SD = 0.67) exhibited relatively high mean scores on the 5-point Likert scale. In contrast, Adverse African Leadership, the sole negatively worded factor, reported a lower mean score (M = 2.47, SD = 0.89) (reversed, in other words, 1 represents a negative and 5 a positive score).

Skewness and kurtosis values for the three subfactors and the overall African Leadership construct ranged from −0.79 to 0.26 and −0.61 to 0.70, respectively, suggesting approximate normality as these values fall well within the acceptable thresholds for skewness and kurtosis (West at al., 1995).

Internal consistency reliability was confirmed with α of 0.95, 0.85 and 0.91, and CR coefficients of 0.95, 0.81 and 0.92 for the respective factors, all exceeding the conventional benchmark of 0.70 (Nunnally & Bernstein, 1994).

Construct validity was supported by AVE values near or above the recommended threshold of 0.50, in conjunction with high CR values, indicating that the items adequately capture the underlying latent factors. Furthermore, discriminant validity was established, evidenced by both the MSV and ASV values being lower than the AVE on all three subfactors, confirming that they are empirically distinct (Hair et al., 2019).

The strength and direction of the linear relationship between the three African leadership subfactors, as Pearson correlation coefficients, are presented in Table 3.

A strong positive correlation was observed between African team leadership and Authentic African leadership (r = 0.82, p < 0.001). Adverse African leadership, the sole negatively worded factor, showed expectedly significant negative correlations with both African team leadership (r = −0.51, p < 0.001) and Authentic African leadership (r = −0.43, p < 0.001).

Consequently, a CFA was conducted to validate the internal structure of the scale derived from the EFA. Four competing models were evaluated: a unidimensional (one-factor) model, a first-order three-factor model, a second-order three-factor model and a bifactor model. Model fit was assessed using multiple indices, as summarised in Table 4.

The results from the CFA, specifically the evaluation of the fit of the four competing models, are discussed in Table 4 (see Figure 1). The one-factor model, in which all 34 items were constrained to load onto a single latent factor representing overall African leadership, demonstrated the poorest fit (CFI = 0.82, RMSEA = 0.10, AIC = 2931.70). A comparative analysis of the model fit indices revealed that the second-order factor model exhibited marginally superior fit (CFI = 0.93, RMSEA = 0.05, AIC = 2192.70) relative to the first-order three-factor model (CFI = 0.92, RMSEA = 0.05, AIC = 3256.00). The Chi-square difference between these two models was 650 (Δχ² = 2700 − 2050) with a corresponding difference of 138 df (662−524), indicating a statistically significant improvement in model fit for the second-order model. This is further supported by the difference in AIC values (3256.00–2192.70 = 1063.30).

The bifactor model, however, provided the best overall fit (CFI = 0.95, RMSEA = 0.04, AIC = 2170.50). In this model, all 34 items were specified to load simultaneously onto a general factor, representing overarching African leadership, as well as onto three orthogonal group factors, namely African team leadership, Adverse African leadership and Authentic African leadership (see Figure 1). This model allowed for the partitioning of item variance into shared variance attributable to the general factor and unique variance explained by each specific factor (subfactor), supporting a more nuanced and multidimensional conceptualisation of African leadership.

The key question is whether the bifactor model provides sufficient evidence to confirm that a single-scale score adequately reflects one-dimensionality, and if it is also appropriate to report domain-specific subfactor scores. To address this, the bifactor model was evaluated using multiple indices, with results presented in Table 5.

The ω estimate for the general factor was exceptionally high at 0.97, indicating a highly reliable multidimensional composite. Subfactor scores, comprising fewer items, exhibited lower reliability, ranging from 0.85 to 0.95 (mean ωS = 0.92, SD = 0.04), suggesting that the general factor (African leadership) predominantly drives the systematic variance in unit-weighted total scale scores. Variance attributed to multidimensionality from subfactors (African team leadership, Adverse African leadership and Authentic African leadership) is thus minimal. The ωH value is > 0.80, supporting unidimensionality (Reise et al., 2013). By controlling for the general factor’s variance, the ωHS for subfactors averaged 0.24 (SD = 0.28), indicating that much of the reliable variance in subfactor scores stems from the general factor rather than unique group factors. Despite this, subfactor scores remain sufficiently reliable for basic research purposes but should be interpreted cautiously because of their low ωHS relative to ωS.

The general factor reported an FD value of 0.98, indicating that observed individual differences in factor scores strongly correspond to true differences on the factor. Subfactor FD values were, however, lower (0.73, 0.89 and 0.78; mean = 0.80, SD = 0.07), influenced by the number of items per subfactor (Rodriguez et al., 2016), yet still acceptable for research purposes. The H values for subfactors averaged 0.55 (SD = 0.16), further supporting the dominance of the general factor.

Finally, to assess potential bias when fitting multidimensional data to a unidimensional model, ECV and PUC were examined. The ECV value was 0.80 (> 0.70, Rodriguez et al., 2016), suggesting that the factor loadings from a unidimensional model closely approximate those from the general factor in a bifactor model, supporting unidimensionality. The PUC value was 0.65, slightly below the 0.70 threshold (Rodriguez et al., 2016), indicating that most correlations reflect the general factor with minimal bias, reinforcing the suitability of a unidimensional framework. It is, however, important to note that PUC is sensitive to the number of factors and fixed number of items, with the PUC value in this case lower because of the rather few factors (only three) and the number of items within a group-factor that are relatively large (the three subfactors contained 11.33 items on average (SD = 3.86), ranging from 6 to 15 items).