Emotional intelligence is conceptualised differently across various measures, that is, different models are used in its measurement. These models include (1) ability-based EI measures, like the Mayer-Salovey-Caruso Emotional Intelligence Test (MSCEIT) (Mayer et al., 2003), (2) self-report (or peer-report) EI measures based on the same theoretical model as the MSCEIT (Mayer et al., 2003) and (3) measures of mixed EI models, which extend beyond the theoretical model included in the aforementioned two categories (Ashkanasy & Daus, 2005; Dasborough et al., 2021). Such mixed EI measures are considered assorted, because they include various items phrased similarly to those used in measuring personality and behavioural preferences (Ashkanasy & Daus, 2005). In contrast to the ability-based EI measured by the MSCEIT (Mayer et al., 2003), the self-report EI measures from the two latter models are often termed ‘trait-based EI’, as it has to do with individuals’ perceptions of their own emotional skills (Joseph et al., 2015; Petrides et al., 2016). All three of these models include scientifically sound EI-related constructs that could greatly contribute to what we know about work performance (Dasborough et al., 2021; Joseph et al., 2015).

Within the categories of ability-based EI and trait/mixed EI, the two theoretical models that have received a lot of attention when exploring the relationships between EI and job performance were the models that underpin the MSCEIT (Mayer et al., 2003) and the Bar-On Emotional Quotient Inventory (Joseph et al., 2015). This article focuses on the revised version of the latter model, namely the EQ-i 2.0 model (Wiechorek, 2011).

The EQ-i 2.0 model was established based on 25 years of research on the different aspects that constitute EI, including how these interrelate. The EQ-i 2.0 model has a multidimensional structure that provides an indication of a person’s total EI, which is contextualised by the different facets believed to underlie EI. The total EI score reflects a ‘snapshot’ of a person’s overall EI and can be defined as (Wiechorek, 2011):

This broad definition also speaks to the different facets included in the EQ-i 2.0, which encompass 15 constructs that can be collapsed into five comparable categories or composites. A visual depiction of the theoretical structure of the EQ-i 2.0, including the proposed general factor, is presented in Figure 1.

The meaning of the composites and subscales’ measures, as demonstrated in Figure 1, is described in the Wiechorek (2011) user’s handbook. The first three subscales, noted below, collectively quantify an individual’s Self-perception, which describes how people see themselves:

The second, the Self-expression composite scale, considers how people express their inner perception of themselves, which is jointly portrayed through the subscales noted below:

The Interpersonal composite scale considers the nature of the relationships that people build with others. The subscales that collectively inform this composite are:

The Decision making composite scale considers how people use emotional information to effectively make decisions, which is jointly established through:

The Stress management composite scale measures peoples’ ability to deal with stressors, utilising multiple coping strategies and showing resilience despite setbacks and incorporates:

The scales, measured at every level of the EQ-i 2.0, were designed with a specific function in mind. The subscales serve to provide a more foundational understanding of employees’ relative strengths and weaknesses, which is valuable for development purposes. Compared to a general factor, subscales are qualitatively more meaningful during psychometric feedback. Psychometric feedback on subscales provides the opportunity to suggest actionable steps that employees could take to increase their overall EI at work. An EI total, or even a composite EI score, might be perceived as too ambiguous and less meaningful from a development perspective. By contrast, the composite and overall dimensions of the structure might provide more encompassing, and therefore also more consistent, dimensions that can be utilised for selection purposes (Wiechorek, 2011).

A study conducted by Van Zyl (2014) supported the existence of the composites, as set out in Figure 1. Van Zyl (2014) was, however, unable to find support for a general factor of EI among the subscales of the EQ-i 2.0. However, Van Zyl (2014) did not inspect the recently suggested sequence for the inspection of hierarchical structure (Credé & Harms, 2015), which was explored in the current study. The present researchers were particularly interested in the general factor of EI as based on subscales of the EQ-i 2.0 and specified a bifactor model.

Van Lill and Taylor’s (2022) framework underlying the Individual Work Performance Review (IWPR) was utilised to conceptualise and measure performance in the present study. Five broad performance dimensions are differentiated by Van Lill and Taylor (2022), including in-role, extra-role, adaptive, leadership and counterproductive performance. According to Van Lill and Taylor (2022, pp. 3–5):

Each of the five broad performance dimensions is represented by four narrow performance dimensions, as shown in Figure 2. Evidence in support of the five factors and definitions of the narrow dimensions can be obtained from Van Lill and Taylor (2022). As portrayed in Figure 2, it is theorised that a general factor stands at the apex of all the performance dimensions identified in the IWPR. Van Lill and Van der Vaart (2022) found that a general factor explained a similar amount of variance in South Africa as that reported by Viswesvaran et al.’s (2005) meta-analytical study. The present study focused on the predictive validity of total EI for general and broad dimensions of individual work performance.

The evidence suggests that the criterion validity of EI is replicable in the South African context (Nel & De Villiers, 2004; Sloan & Geldenhuys, 2021). However, it is less clear what specific work-related behaviours EI predicts when compared to more established international scientific literature. Performance is a multidimensional construct, and Van Lill and Taylor (2022) suggest a five-factor model for individual behaviours at work, namely in-role, extra-role, adaptive, leadership and counterproductive performance. The only local evidence that differentiated between performance dimensions was the study conducted by Sloan and Geldenhuys (2021), which focused on both in-role and extra-role performance as work-related outcomes.

Meta-analytical evidence to date suggests that EI is predictive of in-role performance, also referred to as ‘task performance’ (Joseph et al., 2015; Sackett et al., 2021). Emotional self-regulation is more frequently recognised as a core part of functioning in social enterprises and, therefore, the ability to succeed at essential tasks (Joseph et al., 2015).

There is also evidence in support of the relationship between EI and extra-role performance. Employees with higher EI might be more empathetic and prosocial, which could, in turn, lead to greater displays of extra-role performance or actions aimed at doing more than what is required by their job descriptions (Miao et al., 2017).

Emotional intelligence might also assist employees in better coping with negative emotions that arise from interpersonal strain or frustrating tasks. Consequently, they are less inclined to engage in deviant intrapersonal or interpersonal behaviours that undermine collective goals (Miao et al., 2017).

It further appears that EI could assist individuals in coping with the strain associated with the complexities of change at work. In this respect, EI could help individuals to downregulate negative emotions in response to uncertainty and help them increase positive feelings, in order to stay focussed on solutions in response to change. Emotional intelligence is, therefore, also argued to be related to adaptive performance (Yang et al., 2022).

Finally, EI might translate into greater self-confidence, self-awareness and empathy, which are essential components of interpersonal influence and, therefore, leadership performance (Harms & Credé, 2010). An overview of the evidence presented suggests that EI is likely to be related to all five performance dimensions, namely in-role, extra-role, adaptive, leadership and counterproductive performance.

The EQ-i 2.0 has been used across multiple industries in the United States of America, with accumulating evidence of the instrument’s utility in differentiating between high- and low-performing individuals (Stein & Book, 2011). However, limited research has been conducted on the predictive validity of the general factor of the EQ-i 2.0 in South Africa, which was one of the areas of focus of the current study.

The objective of this study was to determine the structural and criterion validity of a general dimension of EI in the EQ-i 2.0 assessment. Based on the current evidence reported in the present study, the following hypotheses were formulated:

For Study 1, a sample of 16 581 working adults living in Southern Africa was obtained via an online platform. The mean age of the respondents was 37.94 years (standard deviation [SD] = 8.86). Most respondents self-identified as male (n = 9427; 57%), followed by females (n = 7154; 43%). The sample further included individuals who indicated their ethnicity as follows: black (n = 6755; 41%), white (n = 4915; 30%), coloured (mixed ancestry; n = 1675; 10%) and Indian or Asian (n = 1838; 11%). The researchers computed the power for the test model (degrees of freedom [df]= 6667), based on the computer software developed by Preacher and Coffman (2006). The models returned a value of unity, which suggested that an incorrect model would be correctly rejected (α = 0.05; null root mean square error of approximation [RMSEA] = 0.05; alternative RMSEA = 0.08).

For Study 2, a total of 108 performance ratings of South African employees, who were also administered the EQ-i 2.0, were completed by managers in two participating organisations, selected via a census or stratified sampling strategy. The sample represented the finance and professional services sectors. The mean age of employees was 38.88 years (SD = 7.78 years). Most of the employees self-identified as white (n = 65; 60%), followed by black African (n = 18; 17%), Indian (n = 16; 15%), coloured (individuals of mixed ancestry; n = 8; 7%) and Asian (1; 1%). The sample comprised more women (n = 65; 60%) than men (n = 43; 40%). Most of the employees were registered professionals (n = 53; 49%), followed by low-level managers (25; 23%), mid-level managers (n = 16; 15%), skilled employees (n = 12; 11%) and top-level managers (2; 2%). The present researchers inspected the statistical power required for linear bivariate regression by using G*Power (Faul et al., 2007). The calculation suggested that 64 participants should be sufficient (α = 0.05; Power = 0.80) to detect a slope of 0.30, per prior meta-analytical validity estimates reported (Sackett et al., 2021). The present sample was roughly double the recommended size based on this calculation.

The EQ-i 2.0 assessment consists of 133 items; that is an average of eight items in each of the 15 subscales (i.e. three subscales per composite). Eight items contribute to a Well-being indicator, also referred to as the ‘happiness scale’, while seven of these items are also used as a validity measurement. The EQ-i 2.0 model comprises a 1-5-15 structure, where the 15 subscales underlie the five composites that all contribute to one total EI ‘snapshot’ (see Figure 1). A five-point Likert-type frequency scale provides the response options for each item. The response options range from 1 = Never/Rarely to 5 = Almost Always/Always, with a qualitative interpretation guide connected to the meaning of each option (Wiechorek, 2011). The internal consistency reliabilities for the South African sample on most of the EQ-i 2.0 subscales were satisfactory (α and ω ≥ 0.71). Only one subscale had an unsatisfactory internal consistency reliability coefficient of 0.66, namely independence. However, this reliability coefficient was still considered marginally acceptable (see Table 1).

The IWPR (Van Lill & Taylor, 2022) consists of 80 items (4 items for each of the 20 narrow performance dimensions) that cover five factors, namely In-role performance, Extra-role performance, Adaptive performance, Leadership performance and Counterproductive performance (Van Lill & Taylor, 2022). Per the guidelines of Aguinis (2019), each item was measured using a five-point behavioural frequency scale. Word anchors defined the extreme points of each scale, namely, 1 = Never demonstrated and 5 = Always demonstrated. Qualitative interpretation of numeric values between the extreme points is provided, to better approximate an interval rating scale, namely 2 = Rather infrequently demonstrated, 3 = Demonstrated some of the time and 4 = Quite often demonstrated. Van Lill and Taylor (2022) demonstrated the internal consistency reliability of all the narrow dimensions of the IWPR (α and ω ≥ 0.83).

The data on the EQ-i 2.0 (n = 16 581) were collected as part of several archival projects, for different client projects on the JVR Online platform. Data were collected via online assessments for either selection or development purposes. A concurrent set of data was separately collected by asking managers of the 108 employees, who simultaneously completed the EQ-i 2.0, to rate their employees’ performance. A study conducted by Van Lill and Van der Merwe (2022) revealed that employees greatly inflate self-ratings on the IWPR (Van Lill & Taylor, 2022) when compared to managerial ratings, because of leniency bias. Managers might therefore provide a more conservative and accurate estimate of work performance (Van Lill & Van der Merwe, 2022). Managerial ratings also come with the added benefit of reduced method bias, because of another rating source used in addition to self-ratings on the EQ-i 2.0 (Podsakoff et al., 2012).

The current study was low in risk, but precautions were taken to ensure that participation was voluntary and anonymous, that no harm was caused, that the questions were filled in truthfully and that informed consent was given to use the results for research purposes.

Ethical clearance to conduct this study was obtained from the University of Johannesburg Department of Industrial Psychology and People Management Research Ethics Committee (reference number: IPPM-2022-598).

The mean item score and SD for each subscale of the EQ-i 2.0, along with the alpha and omega reliability estimates and standardised inter-factor correlations, are reported in Table 1. The inter-factor correlations were obtained by conducting an oblique lower-order confirmatory factor model. The fit statistics for the oblique lower-order confirmatory factor model of the entire EQ-i 2.0 (χ² [df] = 254827.62 [6680]; CFI = 0.97; TLI = 0.97; SRMR = 0.05; RMSEA = 0.05 [0.05; 0.05]) were satisfactory (Brown, 2015). The oblique lower-order model, which fit statistics are also reported in Table 2, allows group factors to covary and enables an inspection of the inter-factor correlations between the dimensions for descriptive purposes.

The size of the relationships reported in Table 1 was mostly in the medium-to-large range, alluding to the possibility of a general factor. However, 95% of the standardised upper limit inter-factor correlations (UL) were below the cut-off recommended by Rönkkö and Cho (2020), namely UL < 0.80, suggesting a fair degree of discriminant validity at the subscale level. In a select few cases (5% of UL), insufficient evidence for discriminant validity existed. However, the lower levels of discriminant validity could be attributable to established theoretical relationships that have been reported between the scales in the past (Rönkkö & Cho, 2020; Wiechorek, 2011).

Table 1 also contains the inter-item consistency reliabilities. All the subscales obtained coefficient alpha ordinal and McDonald’s omega values (α and ω ≥ 0.71) above the recommended threshold (Cortina, 1993; Cortina et al., 2020). Only the independence subscale had an ω-value of 0.66, which was still accepted, as the α ordinal value was 0.77. These results suggest that the subscales reliably measure the respective scale constructs.

The fit of different factor models proposed by Credé and Harms (2015) was subsequently investigated to determine whether a general factor or alternative configurations explained the covariances between the subscale dimensions of the EQ-i 2.0. The different models are reported in Table 2.

The results reported in Table 2 indicated that the more parsimonious models, namely the orthogonal first-order and single-factor models, displayed a weaker fit to the data (Credé & Harms, 2015). By contrast, the more complex models (oblique lower-order, higher-order and bifactor models) displayed superior fit, supporting the existence of hierarchical structure in the data. The present study focussed on the manifestation of a general factor of EI and, therefore, necessitated a further inspection of the satisfactory fitting bifactor model, instead of the slightly better fitting oblique lower-order factor model.

It is recommended that bifactor statistical indices be calculated to determine the practical meaningfulness of general versus group factors in a bifactor analysis (Rodriguez et al., 2016a, 2016b), such as the explained common variance (ECV), coefficient omega hierarchical (ω_h), construct replicability (H), factor determinacy (FD), percentage uncontaminated correlations (PUC) and relative percentage bias (ARPB). Group factors are considered plausible when ω_h, H and FD² are > 0.50, 0.70 and 0.70, respectively (Dueber, 2017; Reise et al., 2013). Explained common variance for the general factor > 0.70 and PUC > 0.80 are indicative of unidimensionality (Reise et al., 2013). Relative percentage bias below 10% – 15% indicates little difference in the factor loadings between a single-factor model and the general factor in a bifactor model (Rodriguez et al., 2016a). Bifactor statistical indices were calculated using Version 0.2.0 of the Bifactor Indices Calculator package (Dueber, 2020) in R (R Core Team, 2016). The bifactor statistical indices are reported in Table 3.

The General EI factor accounted for over half of the common variance. The ECV > 0.50, including the high PUC > 0.80, suggests the presence of a strong general factor (Reise et al., 2013). These results were further supported by the ARPB value of 5%, which was well below the 10% – 15% mark, indicating no serious concern of measurement bias if the model was treated as unidimensional (Rodriguez et al., 2016a).

The subscale values all had a large difference between the ω and ω_h values, with ω_h ranging from 0.14 to 0.54. This normally suggests that the subscales mostly do not add additional unique variance, and that the factor model is unidimensional. However, Morin (2023) cautions against the use of ω and ωhs, as both tend to underestimate the reliability of group factors. The general trend of the FD and H coefficients, as well as the evidence of the discriminant validity of the group factors presented in Table 1, suggests that the subscales in the EQ-i 2.0 add additional interpretive value for development purposes (Dueber, 2017; Reise et al., 2013). The subscales also still explain the remaining 42% of the variance in the items. Consequently, at a cursory view of the hierarchical structure and without discounting the value of the subscales, it can be argued that a general EI factor exists, which supports Hypothesis 1.

Linear regressions were conducted to determine the relationship between overall EI and individual work performance. The regression coefficients of the different relationships are reported in Table 4.

In comparison to meta-analytical evidence, the validity estimate (R = 0.39) for General performance appears to be slightly higher in the present study compared to other meta-analytical estimates, namely ρ = 0.29 (Joseph et al., 2015) and ρ = 0.30 (Sackett et al., 2021). However, the estimate was still in the same direction and, similar to prior findings, moderate in size. Hypothesis 2 was, therefore, confirmed. The correlations for the broad dimensions of performance also appeared mostly moderate in size (R = 0.25 to 0.42; M = 0.35) and in the theorised directions. Therefore, Hypotheses 2A to 2E could also be confirmed. The regression coefficient for Leadership performance appeared to be pronounced, which could be attributed to the high emotional labour often associated with interpersonal influence (Glomb et al., 2004).

Mean group gender differences have been reported for EI in the past, with scholars suggesting that men might be adversely affected in selection procedures if a total score of EI is used in isolation to make a hiring decision (Joseph & Newman, 2010). Multi-group CFAs suggest that composite scales in the EQ-i 2.0 are invariant across gender groups. An inspection of mean group differences further revealed only small effect size differences (Stols & Van Lill, 2022). Further research could inspect the invariance and mean group gender differences of a general dimension of EI based on the EQ-i 2.0 in the Southern African context.

The predictive validity of total EI was mainly based on professional and managerial staff, many of whom were employed in the financial and health sector, for whom high EI might be a more pronounced occupational requirement. Most other Southern African studies appeared to have sampled managerial employees, and future studies could inspect whether these relationships are replicable for other job families in South Africa, such as skilled/semiskilled, clerical, military and law enforcement jobs.