Organisational leadership (OL) is a critical driver of organisational success, shaping performance, culture and employee engagement in dynamic and context-specific environments (Northouse, 2021; Yukl & Gardner, 2020). Despite extensive global research, the application of leadership theories in African contexts, also in South Africa, remains underexplored, with a heavy reliance on Western-centric frameworks that often fail to capture the unique socio-cultural, economic and political dynamics (Banutu-Gomez Fanneh et al., 2023; Enslin & Grobler, 2025). These universal models risk oversimplifying leadership by neglecting local values and practices, necessitating context-specific (emic) approaches that integrate African perspectives, such as Ubuntu, to accommodate the cultural diversity (Abdulai, 2021; Grobler & Singh, 2018; Mbandlwa & Fagbadebo, 2020), thus acknowledging the uniqueness of the context when studying leadership. This study’s specific context, unlike developed approaches that prioritise tasks, promotes ethical and inclusive practices in organisations, resonating with collectivist and relational leadership styles and offering improved employee engagement along with broader global relevance.

According to Creighton-Randall (2024) and Vika and Pule (2025), organisations across the globe are increasingly valuing environments built on mutual respect, empathy and shared responsibility to boost performance and strengthen social bonds. As global challenges grow more intricate, human-centred leadership provides a powerful path to ethical, inclusive and effective practices across the world. The plea for context-specific leadership research extends beyond Africa; Johns (2024) advocates for context-specific leadership and urges a departure from blindly relying on universal or decontextualised frameworks. This shift is crucial for both academic theory and practical leadership development (Cikara et al., 2022; Johns, 2017), and, related to this study, the accurate measurement of leadership (Grobler & Sibanda, 2024; Grobler & Singh, 2018).

This study is focused on the development and validation of a scale that measures OL, and is conceptually based on (1) a thorough systematic literature review (Enslin & Grobler, 2025) and (2) a contextual (emic) investigation into OL, using interactive qualitative analyses (IQA) (Grobler & Enslin, in press). The IQA was purposefully used by Grobler and Enslin (in press) as a technique that is based on participation and collaboration, intended to explore local values, worldviews and lived experiences of organisational leaders as well as ‘the consumers’ of leadership. Their findings reflect the contextual realities of OL, supported by various scholars from the African continent, for example, Iyioke (2017), Punnett (2017), as well as Vilakati and Schurink (2021).

The results of the statistical procedures discussed in the previous section are now reported in this section. To determine the construct validity and construct equivalence of the OLS, an EFA was conducted. The EFA was first conducted on 32 items of the OLS with all being positively worded in Study 1 and then on a combination of 19 positive and negative worded items in Study 2.

The relatively large sample size contributed to an adequate variable-to-respondent ratio (80:1 and 86:1 in Study 1 and Study 2, respectively). It must be noted that Study 1 examined the 32-item OLS instrument, while Study 2 examined the refined 19-item version. The values of the KMO measure of sampling adequacy were 0.98 and 0.95, above the recommended value of 0.60, and Bartlett’s test of sphericity was significant (p < 0.001), and χ² (171) = 37515.05 and χ² (171) = 15984.85, supporting the strategy to conduct an EFA in both studies.

The Guttman–Kaiser K1 rule was used in conjunction with the scree plot to determine the number of factors. The results of the Kaiser’s criterion, as determined by the PCA, are reported in Table 2.

The factor solution with eigenvalues larger than one, reported in Table 2, consists of one factor consisting of 19 items, and two factors for Study 1 and Study 2, respectively. The total variance explained exceeds the critical value of 50%. Cattell’s scree test, which is aimed at retaining the components (factors) before the break (elbow rule), was performed, and the results are reported in Figure 1.

The interpretation of the scree plots, presented in Figure 1, suggests the presence of a single dominant factor in Study 1, as indicated by the clear inflection point (elbow) after the first component. In contrast, the scree plot for Study 2 suggests the presence of two distinct factors.

An HTMT value of 0.46 (< 0.85, Kline, 2011) was reported in Study 2, indicating that discriminant validity is established and that the two factors are distinctively different from one another. To further assess construct validity of the instruments, CFAs were conducted for both studies. The results reported in Table 3 include key model fit indices, namely the CFI, RMSEA, Chi-square (χ²) and the ratio of the change in Chi-square to the change in degrees of freedom (Δχ²/Δdf). Study 1 consists of 19 items, all positively worded, and Study 2, with the same items, except that 6 have been changed to be negatively worded, yielding a two-factor factor structure.

Consequently, a CFA was conducted to validate the internal structure of the scale derived from the EFA for both Studies^1&2. Study 1 consisted of the confirmation of the unidimensional model, while Study 2 assessed four competing models. These statistics fit these models, namely a unidimensional (one-factor) model, a first-order three-factor model, a second-order three-factor model and a bifactor model, as reported in Table 3.

The CFA in Study 1 confirmed the unidimensional factor model (all 19, unidirectional items) of the OLS (χ²/df [184] = 21.75, CFI = 0.96, RMSEA = 0.065). The analysis of Study 2 was more complicated by the inclusion of multidirectional items, as it impacted on the factor structure. A first-order CFA with a two-factor solution was conducted to validate the measurement model and yielded acceptable fit statistics (χ²/df [146] = 6.48, CFI = 0.95, RMSEA = 0.058). A comparative analysis of the model fit indices revealed that the first-order factor model exhibited superior fit compared to the unidimensional model (Chi-square difference between these two models was 1322.71 [Δχ²= 2511.69–888.98] with a corresponding difference of 1 degree of freedom [152−151]). This was followed by attempts to introduce a second-order factor, but this resulted in estimation errors, and it was therefore decided to conduct bifactor analysis, as it would assess the possibility of an overall OL factor.

The bifactor model demonstrated the best fit (χ²/df [133] = 4.81, CFI = 0.95, RMSEA = 0.058). The Chi-square difference between the first-order and bifactor models was 249.12 (Δχ²= 888.98–639.86) with a corresponding difference of 18 degrees of freedom (151−133). In this model, all 19 items were set to load concurrently on a general factor, representing overall OL, and on two orthogonal group factors, namely constructive and destructive OL, respectively. This structure enabled the separation of item variance, firstly, into shared variance from the general factor (OL) and unique variance from each specific sub-factor, providing a clearer and more detailed understanding of OL by capturing its multidimensional nature.

A further examination of OL’s dimensionality (as a bifactor model) was conducted by assessing the strength of a general versus sub-factors, using various indices. The results are reported in Table 4.

The omega (ω) estimates for the general factor, representing OL, was exceptionally high at 0.95, indicating a highly reliable multidimensional composite. This suggests that the general factor captures most of the systematic variance in unit-weighted total scale scores. Sub-factor scores, comprising destructive and constructive OL, demonstrated slightly lower reliability, with ωS ranging from 0.84 to 0.95 (mean ωS = 0.90, SD = 0.06). This indicates that sub-factor reliability is adequate but less robust than the general factor, as expected because of item count differences (Reise et al., 2013). The variance attributed to multidimensionality from sub-factors is lower, reinforcing the general factor’s dominance.

The omega hierarchical (ωH) value of 0.84, exceeding the 0.80 threshold (Reise et al., 2013), further supports the general factor that accounts for the most reliable variance. After controlling for the general factor’s variance, the omega hierarchical subscale (ωHS) for sub-factors averaged 0.33 (SD = 0.29), suggesting that much of the reliable variance in sub-factor scores derives from the general factor (OL) rather than the sub-factors. While sub-factor scores remain sufficiently reliable for basic research, their low ωHS relative to ωS warrants cautious interpretation.

The FD value for the general factor was 0.96, indicating that observed individual differences in factor scores closely reflect true differences on the factor. Sub-factor FD values were lower (0.89 and 0.82; mean = 0.86, SD = 0.04), influenced by fewer items per sub-factor (Rodriguez et al., 2016). The H values for sub-factors averaged 0.70 (SD = 0.07). The FD and H values confirm the general factor’s dominance over sub-factors, but support the utility of the sub-factors for research purposes.

Finally, to evaluate potential bias from fitting multidimensional data to a unidimensional model, the ECV was assessed. Although the ECV value of 0.66 is slightly below the 0.70 threshold (Rodriguez et al., 2016), it still supports the use of OL as a general factor.

The descriptive statistics, the internal consistency as well as the convergent and discriminant validity of OL as a unidimensional construct (Study 1) and a two-factor with a general factor (bifactor model) (Study 2) are reported in Table 5.

The unidimensional OL construct in Study 1 exhibited a relatively high mean score (M = 3.73, SD = 0.82), while the general factor showed a slightly lower mean (M = 3.60, SD = 0.69), possibly because of the inclusion of negative items in Study 2. Variables based on positively worded items (OL and constructive OL) displayed negative skewness, indicating distributions with fewer small values and a left-tailed tendency. Conversely, destructive OL showed a positive skewness value of 0.26. Kurtosis values ranged from −0.61 to 0.60. As the skewness and kurtosis values for all factors remained below the critical thresholds of 2.00 and 7.00, respectively, the data can be considered normally distributed.

Cronbach’s alpha coefficients for all variables across both studies are deemed acceptable (α ≥ 0.70), indicating satisfactory internal consistency (Hair et al., 2019). Convergent validity was established through Composite Reliability (CR) and Average Variance Extracted (AVE). Discriminant validity was assessed and confirmed by comparing AVE with MSV, with satisfactory results indicated by MSV being less than AVE and ASV also being less than AVE (Hair et al., 2019).

The subsequent step in the scale development involved examining measurement invariance, which was conducted in Study 1. Measurement invariance ensures that the OLS measures the same construct in the same way across different groups (Davidov et al., 2014), in this case, the private and public sectors. A random sample of 500 participants from each group (private and public sectors) was used to analyse invariance. The analysis employed a criterion of a change of −0.01 in the CFI, with Chen (2007) recommending the use of alternative fit indices such as ΔRMSEA and ΔSRMR. The critical values for these indices were set at 0.015 and 0.030 for metric invariance and scalar invariance, respectively.

Results supported configural invariance, as indicated by acceptable model fit statistics when the groups were estimated freely, without constraints. Specifically, the changes in fit indices were ΔCFI = 0.01, ΔRMSEA = 0.017(marginally exceeds the 0.015 criterion), and no change was observed in SRMR.

Further assessment of measurement invariance from both a metric and scalar perspective was conducted, and the detailed results are presented in Table 6.

The results reported in Table 6 indicate that metric and scalar invariance is supported, as the delta values of the fit indices between unconstrained and fully constrained models meet the recommended values. The instrument can therefore be used within the private and public sectors as it measures OL consistently in the same way across the sectors.

The fourth research objective is to compare the two versions, focusing on the impact of the directional change of six items in Study 2. These items were merely transformed from being positively to negatively worded, without changing the essence of the item. The comparison was performed on the item level (see Table 7) and on the construct level, as reported in Table 8. The items that were changed are marked with an asterisk (*).

The comparison of the two versions of the OLS indicates that the most significant differences are reported on four of the reversed items, namely items 4 (∆ = 38.1%), 14 (∆ = 20.1%), 16 (∆ = 19.6%) and 18 (∆ = 14.8%).

Only 14% of the respondents responded overall negatively on OL in Study 1 compared to the 18% in Study 2, resulting in the positive response being 4.58% less in Study 2. Presented in a different manner, the difference between the negative and positive responses is presented as a ratio of 1:4.95 and 1.3.55 for Study 1 and Study 2, respectively.

In summary, participants in Study 1 (one-directional) perceived their leaders more positively, with the largest differences emerging in reverse-coded items (Study 2) related to empathy, diversity, emotional safety and belonging.

To assess the impact of incorporating multidirectional items on the relationship between latent constructs and external criterion variables, as postulated by Podsakoff et al. (2003), a preliminary (albeit limited) analysis was conducted to examine the relationship between OL as measured by the one-directional instrument and the OL bifactor derived from the multidirectional instrument.

To explore this concern, basic correlational analyses were conducted with related leadership constructs that are theoretically expected to be associated. The results are presented in Table 8.

It is apparent that there is a degree of convergence between OL (both variations) and the cognate leadership constructs. Comparing the two versions indicates that the differences in the correlations are minor, with only the correlations between transformational leadership and the OL bifactor being significantly stronger than the correlation with OL in Study 1 (r = 0.77 and r = 0.85; z-score of −7.44). The inclusion of the multidirectional items, in this case, did not have an impact on the relationship with external variables. No evidence could be found that the change in the direction of the items lead to weaker correlations between the latened variables and external criterion variables (Podsakoff et al., 2003).

Organisational leadership is regarded as pivotal in driving organisational success by shaping performance, culture and employee engagement within dynamic, context-specific settings. However, global leadership research, predominantly rooted in Western frameworks, often overlooks the unique socio-cultural, economic and political dynamics of African contexts, including SA. These universal models risk oversimplifying leadership by neglecting local values such as Ubuntu, which emphasises interdependence and community. The limited application of context-specific (emic) approaches in African leadership has led to the need for frameworks that integrate African perspectives to address cultural diversity and contextual challenges effectively.

This study advances context-specific leadership research through the development and validation of an OLS – more specifically, two equivalent versions thereof. The approach to developing the instrument is regarded as emic in nature, as it was based on a systematic literature review and IQA (Enslin & Grobler, 2025; Grobler & Enslin, in press). By engaging leaders and their followers to explore local values, worldviews and lived experiences, the IQA fostered a participatory approach to understanding OL’s contextual realities. This approach is necessitated because of the uniqueness of leadership in Africa, which is grounded in principles like Ubuntu, prioritises collective decision-making, fairness and social cohesion, and aligns with servant and authentic leadership models. This contrasts with Western models’ focus on individualism and competition, emphasising the necessity of emic methodologies to develop culturally relevant leadership instruments and counter the limitations of decontextualised, universal frameworks in diverse settings.

This study focuses on developing and validating an emic OLS, tailored to the SA context, with two equivalent versions evaluated across two independent studies. The first version (Study 1) consists of only positively worded items (unidirectional), while the second (Study 2) uses the same items but worded multidirectionally (positive and negative). The rationale for this conversion was to mitigate issues like acquiescence and common method bias, while enhancing measurement balance.

The study’s objectives are fourfold. Objective 1 was to develop and validate the unidirectional item OLS. An instrument, consisting of 32 original items, was assessed in terms of construct validity by means of an EFA. The results yielded a unidimensional factor structure, with 19 items (all positively worded), and accounted for more than 60% of the variance in OL. This unidimensional factor structure was also confirmed with a CFA, and measurement invariance was also reported, by using the sector (public and private sector) as a grouping variable (Objective 2).

Based on these 19 items, a multidirectional OLS (with six negatively worded items) was developed and validated (Objective 3). The EFA yielded a two-factor model, also confirmed by means of a CFA, but because of estimation errors, a secondary factor model could not be assessed. It was subsequently decided to conduct a bifactor analysis, which produced the best fit statistics.

The bifactor model entails that all 19 items (although multidirectional) reflect a general factor, namely OL and two distinct sub-factors (constructive and destructive OL, respectively).

The two sub-factors of OL can be best described as:

Constructive organisational leadership – Leadership that integrates emotional intelligence, ethical integrity and strategic vision to inspire, empower and unite team members towards shared organisational goals. This dimension is characterised by self-awareness and emotional regulation, transparent and empathetic communication, high ethical standards and a focus on fostering individual and collective growth through meaningful relationships, clear direction and collaborative energy. Authentic transformational leaders motivate through their actions, challenge others to grow through engaging dialogue and create a supportive, trust-based environment that drives sustainable success.

Destructive organisational leadership – Detrimental leadership approach characterised by a lack of empathy, integrity and inclusivity, which undermines team morale and organisational success. This style is marked by dismissing diverse perspectives without compassion, acting without ethical principles, neglecting to coach or mentor team members for growth, failing to foster a safe emotional environment, rejecting diversity and neglecting to cultivate a sense of belonging or unity. Toxic leaders create divisive, unsupportive and disengaged workplaces, leading to diminished trust, poor collaboration and hindered individual and team performance.

Both versions, as a unidimensional (measured by unidirectional items) factor model and as the bifactor model (measured by multidirectional items), consisting of a general factor (OL) and two sub-factors (constructive and destructive OL), reported acceptable psychometric properties, including Cronbach alpha coefficients, CR, AVE, MSV and ASV.

The convergent validity of OL, as measured by both versions of the OLS, was further assessed by comparing it statistically to cognate relational leadership constructs. This includes authentic, ethical, servant and transformational leadership. Strong positive correlations were reported (regardless of the factor permutation), supporting the notion of convergent validity.

The fourth objective was to compare the response patterns of the two versions as well as the impact the directional differences have on their relationship with cognate leadership variables (using the same assessment as with Objective 3).

Note, a major difference at the item level in the response patterns was reported, although participants in Study 1 (unidirectional items) perceived their leaders more positively, with the most significant differences emerging in the reverse-coded items in Study 2. These items were mainly related to empathy, diversity, emotional safety and belonging. The assessment of the impact of the different versions on the relationship with other external (in this case cognate) leadership constructs yielded unconvincing results.

In conclusion, this study provides two equivalent versions of an instrument to measure OL, which is context-specific and was rigorously assessed. The first version is a unidimensional, positively worded item instrument measuring OL. The second is a multidirectional item instrument, consisting of two sub-factors, namely constructive and destructive OL, with a general OL factor (in accordance with the bifactor model).

The choice between the instruments could be influenced by various aspects, such as the nature of the population and sample (level of English proficiency, preparedness to participate in the study), the organisation and the rationale for the use of the instrument, the relationship between the participants and their leaders, etc. The choice of the multidirectional item might mitigate acquiescence bias, enhance content validity, reduce the risk of common method bias and improve measurement balance. It may however reduce consistency because of participant confusion or misinterpretation because of cognitive demand, and for the researcher, it requires careful handling of reversed items, complicating the analysis.

If the priority of the application of the instrument is clarity, reliability and ease of interpretation, and the population may be sensitive to complexity, it is recommended that the unidirectional item version be used. On the contrary, if the priority is detecting and correcting for bias, and the participants can handle cognitively demanding tasks, then the multidirectional item version is more appropriate.