Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Evaluating the higher-order structure of the Profile of Emotional Competence (PEC): Confirmatory factor analysis and Bayesian structural equation modeling

  • Yuki Nozaki ,

    Roles Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Visualization, Writing – original draft, Writing – review & editing

    y_nozaki@konan-u.ac.jp

    Affiliation Department of Human Science, Faculty of Letters, Konan University, Kobe, Japan

  • Alicia Puente-Martínez,

    Roles Data curation, Funding acquisition, Investigation, Methodology, Writing – review & editing

    Affiliation Department of Social Psychology and Methodology of Behavior Sciences, University of the Basque Country, Lejona, Spain

  • Moïra Mikolajczak

    Roles Conceptualization, Data curation, Investigation, Methodology, Supervision, Writing – review & editing

    Affiliation Department of Psychology, Université catholique de Louvain, Louvain-la-Neuve, Belgium

Evaluating the higher-order structure of the Profile of Emotional Competence (PEC): Confirmatory factor analysis and Bayesian structural equation modeling

  • Yuki Nozaki, 
  • Alicia Puente-Martínez, 
  • Moïra Mikolajczak
PLOS
x

Abstract

Emotional competence (EC) reflects individual differences in the identification, comprehension, expression, regulation, and utilization of one’s own and others’ emotions. EC can be operationalized using the Profile of Emotional Competence (PEC). This scale measures each of the five core emotional competences (identification, comprehension, expression, regulation, and utilization), separately for one’s own and others’ emotions. However, the higher-order structure of the PEC has not yet been systematically examined. This study aimed to fill this gap using four different samples (French-speaking Belgian, Dutch-speaking Belgian, Spanish, and Japanese). Confirmatory factor analyses and Bayesian structural equation modeling revealed that a structure with two second-order factors (intrapersonal and interpersonal EC) and with residual correlations among the types of competence (identification, comprehension, expression, regulation, and utilization) fitted the data better than alternative models. The findings emphasize the importance of distinguishing between intrapersonal and interpersonal domains in EC, offer a better framework for differentiating among individuals with different EC profiles, and provide exciting perspectives for future research.

Introduction

Individuals differ in the extent to which they can appropriately identify, understand, express, regulate, and utilize their own and others’ emotions. The concept of “emotional competence” (EC)––alternatively labeled “emotional intelligence” (EI)––has been proposed to account for this idea. Although the term EC was originally proposed to account for these individual differences [1], the term EI was later proposed and became much more popular. However, we prefer the term EC to EI because recent meta-analysis shows that they can be improved via relatively short trainings, unlike intelligence [2]. Given this line of research, we will use the term EC hereafter as a synonym of EI, in accordance with previous research [38].

Whether called EC or EI, the nature of these emotion-related differences has long been a source of debate among researchers. Some authors view them as the result of differences in abilities [9], others personality [10] and still others as the result of a mix of both [11]. The tripartite model proposed by Mikolajczak, Petrides [12] integrates these different conceptions by considering that people can difference in emotion-related knowledge, abilities and traits. The knowledge level refers to what people know about emotions and emotionally competent behaviors (e.g., Do I know which emotional expressions are constructive in a given social situation?). The ability level refers to the ability to apply this knowledge in a real-world situation (e.g., Am I able to express my emotions constructively in a given social situation?). The trait level refers to emotion-related dispositions, namely, the propensity to behave in a certain way in emotional situations (e.g., Do I typically express my emotions in a constructive manner in social situations?). These three levels of emotion-related individual differences are loosely connected [13]. In the current paper, we focus on the trait level typically measured through self-report questionnaires [14] because the trait-level is more strongly associated with adjustment than the ability-level is [1519].

Previous research has shown that the trait level of EI/EC is positively associated with better adjustment, such as more adaptive emotion regulation [20], greater subjective well-being [18], better mental and psychical health [16, 21], higher academic performance [22], higher job satisfaction [23, 24], less counterproductive work behavior [17] and greater romantic relationship satisfaction [25]. These relationships remain significant after controlling for personality or cognitive ability [26, 27].

To assess the trait-level EC, Brasseur, Gregoire [28] recently developed the Profile of Emotional Competence (PEC). This scale assesses 10 core EC facets: five types of competences (emotion identification, emotion comprehension, emotion expression, emotion regulation, and emotion utilization), each comprising an intrapersonal domain (concerning one’s own emotions) and an interpersonal domain (concerning others’ emotions). These five competences derive from the four-branch model proposed by Mayer and Salovey [9]; however, they separate the identification from the expression of emotions based on research on alexithymia showing that these branches are factorially and conceptually distinct [29]. A strength of the PEC is that it can assess both intrapersonal and interpersonal domains in all five core competences. Moreover, previous research has found that it had an adequate reliability and incremental validity over the Big Five personality traits [6, 28]. Given its strengths, the PEC has been rapidly adopted in recent EC research [5, 7, 3037].

Because EC facets are positively related to each other [28], they will be hierarchically structured. Clarification of the higher-order structure of individual differences is important because it can provide a parsimonious summary of the vast complexity of human nature [38]. Given that the above 10 core EC facets are categorized into a 2 (type of target) × 5 (type of competence) framework, we can assume six possible structures. These six candidate models are depicted in Fig 1 and briefly described hereafter.

thumbnail
Fig 1. Candidate factorial models for emotional competence.

EC: emotional competence, Iden.: emotion identification, Com.: emotion comprehension, Exp.: emotion expression, Reg.: emotion regulation, Uti.: emotion utilization.

https://doi.org/10.1371/journal.pone.0225070.g001

Unidimensional structure: The core 10 EC facets form only one higher-order factor (global EC). This model will serve as a baseline for model comparison in the statistical analyses.

Target-based structure: The 10 core EC facets form two higher-order factors: intrapersonal and interpersonal EC. These factors do not distinguish between the type of competence (emotion identification, emotion comprehension, emotion expression, emotion regulation, or emotion utilization).

Competence-based structure: The 10 core EC facets form five higher-order factors (emotion identification, emotion comprehension, emotion expression, emotion regulation, and emotion utilization) that do not distinguish between intrapersonal and interpersonal competence.

Hybrid structure: Instead of a normal second-order factor model, we can use a hybrid model [39, 40]—an extension to the bifactor model to capture the 2 (type of target) × 5 (type of competence) crossed structure. The 10 core EC facets form two types of dimensions (type of target and type of competence) to yield additive effects. We provide further details and previous research applications of this model in S1 Text.

Modified target-based structure and modified competence-based structure: In the hybrid structure model, if factors are unstable, they can be replaced with residual correlations [41, 42]. Thus, we can also test a model replacing factors of competence-based structure with residual correlations in the hybrid structure (i.e., a modified target-based structure) or replacing factors of target-based structure with residual correlations in the hybrid structure (i.e., a modified competence-based structure).

The authors of the PEC [28] originally assumed 10 first-order factors and two second-order factors (intrapersonal and interpersonal EC), as corroborated by other research [6, 32]. However, to the best of our knowledge, no study has ever systematically compared this target-based structure with other theoretically plausible factor structures. Consequently, the optimal model for the PEC is still unknown. To fill this gap, we compared the fit of the six theoretically plausible models and tested the replicability/stability of the results across four different samples (French-speaking Belgian, Dutch-speaking Belgian, Spanish, and Japanese).

To evaluate the best factor structure, we followed the flowchart recently proposed by Schmitt, Sass [38]. They encourage researchers to start by conducting dimensionality analyses (e.g., parallel analysis, minimum average partial test, exploratory graph analysis); then, if theoretical candidate factor structures exist, they recommend confirmatory factor analysis (CFA). After that, if model fit is not sufficient, they recommend Bayesian structural equation modeling (BSEM) to explore the source of model misfit.

Previous research has emphasized that the model constraints in traditional CFA are unrealistic for the study of hierarchical constructs. For example, Hopwood and Donnellan [43] found that widely used personality trait inventories (e.g., the Revised NEO Personality Inventory [44]) usually demonstrate poor model fit when their structure is evaluated with CFA. This failure is due to the inherent complexity of hierarchical constructs: In typical CFA, cross-loadings and residual correlations are presumed fixed at exact zero, but these unnecessarily strict models lead to poor model fit and substantial parameter biases for factor loadings and correlations [45, 46]. Nevertheless, freer parameters for cross-loadings and residual correlations would result in a non-identified model under the traditional CFA.

To solve this issue, Muthén and Asparouhov [46] proposed a new statistical approach, called BSEM. This approach allows simultaneous estimation of all cross-loadings and residual correlations by using approximate zero informative priors to replace the exact zeros for those loadings and correlations. By applying BSEM, researchers can investigate whether model misfit is due to small or large cross-loadings/residual correlations, missing factors, or extra factors [41]. BSEM has already been successfully applied to various existing cognitive and non-cognitive measures [41, 4652]. Thus, we apply BSEM to investigate source of model misfit if the fit of the best traditional CFA model is not sufficient.

The current research

This study aimed to evaluate the higher-order structure of the PEC using Schmitt, Sass [38]’s guidelines. As recommended in their flowchart, we started with dimensionality analyses, followed by traditional CFA and BSEM. In order to test the stability and replicability of the results, we evaluated the structure of PEC across four different language samples from Western and Eastern cultures (French-speaking Belgian, Dutch-speaking Belgian, Spanish, and Japanese).

Method

Participants and procedure

Sample A consisted of 3295 French-speaking Belgians (males = 1355, females = 1854, unanswered = 86, Mage = 53.36, SD = 14.01), who completed the French version of the PEC. Sample B consisted of 9955 Dutch-speaking Belgians (male = 3746, female = 5850, unanswered = 359, Mage = 55.62, SD = 13.34), who completed the Dutch version of the PEC. Sample A and B were derived from a part of a study conducted by the largest Mutual Benefit Society in Belgium. The data have already been used to answer other research questions (i.e., on the impact of EC on healthcare service use; [30, 37]); however, no factor analysis of EC has ever been conducted on these data. Sample C consisted of 792 Spanish people (male = 278, female = 512, unanswered = 2, Mage = 24.07, SD = 8.44), who completed the Spanish version of the PEC over the course of a university semester. The survey was conducted using SurveyMonkey, and sent via email to all students enrolled in the course. Sample D consisted of 549 Japanese people (male = 344, female = 205, Mage = 31.67, SD = 14.45), who completed the Japanese version of the PEC. They were recruited via a Japanese data collection company (Cross Marketing Inc.)

Participants in all samples answered the questionnaire online. At the beginning of the survey, they were informed about the nature of the study, including the study’s purpose, their right to withdraw from the study, and the confidentiality of their responses. After reading this material, participants provided informed consent by clicking the “accept” button to start the survey. In addition to the EC scale, participants completed other measures unrelated to the present research question. This study was approved by the ethics committees of the Université catholique de Louvain, University of the Basque Country, and Kyoto University.

Measure

EC was assessed with the PEC [28]. This scale comprises 10 first-order subscales with five items each: identification-self (e.g., I am aware of my emotions as soon as they arise), comprehension-self (e.g., As my emotions arise, I don’t understand where they come from; reversed item), expression-self (e.g., I am good at describing my feelings), regulation-self (e.g., When I am sad, I find it easy to cheer myself up), utilization-self (e.g., My emotions inform me about changes I should make in my life), identification-other (e.g., I can tell whether a person is angry, sad or happy even if they don’t talk to me), comprehension-other (e.g., Most of the time I understand why people feel the way they do), expression-other (e.g., I find it difficult to listen to people who are complaining; reversed item), regulation-other (e.g., I am good at lifting other people’s spirits), and utilization-other (e.g., If I wanted, I could easily influence other people’s emotions to achieve what I want).

All translated measures were created via a back-translation procedure. Participants in samples A, B and D rated each item on a 5-point scale, whereas, participants in sample C rated it on a 7-point scale, because this sample were relatively homogeneous (i.e., everyone was a student). To increase the potential to detect true variation, the number of response options was increased [53]. Importantly, this modification did not affect our main results, because we found similar factor structure across all samples, as described in the results section.

Statistical analyses

First, we conducted dimensionality analyses based on the first-order facet scores, using the exploratory graph analysis [54]. This method has been shown to be superior to other traditional dimensionality analysis methods such as the parallel analysis or the minimum average partial test [54, 55]. Exploratory graph analysis with a triangulated maximally filtered graph was conducted using the EGA 0.4 package [56] in R 3.5.0 [57].

Next, we implemented CFA to compare the fit of possible factor structure models (Fig 1). All CFA were conducted with Mplus Version 8.2 [58]. Since the normalized estimate of Mardia’s coefficient indicated that multivariate normality was violated, we applied a robust maximum likelihood (MLR) estimator and the Satorra–Bentler scaled χ2. There is a controversy as to whether MLR or weighted least squares mean- and variance-adjusted (WLSMV) estimation is superior when multivariate normality is violated [59]. However, neither the Akaike Information Criterion (AIC) nor the Bayesian information criterion (BIC) can be computed with WLSMV, while both can with MLR. Because AIC and BIC are frequently used for model comparison, we used MLR in this study. To help parameter estimation, we constrained paths from the same second-order factors with only two indicators (i.e., competence-based structure, hybrid structure, and modified competence-based structure), as in previous studies [60]. We used AIC and BIC for model comparison; lower BIC and AIC suggest better model fit. Moreover, we used the comparative fit index (CFI; a value ≥ .90 suggests acceptable fit), standardized root mean square residual (SRMR; a value ≤ .08 suggests acceptable fit), and root mean square error of approximation (RMSEA; a value ≤ .08 suggests acceptable fit) to evaluate overall model fit [61, 62]. Missing values (only 0.013%) were handled by full information maximum likelihood estimation (software default settings).

If the fit indices of the best-selected model are not sufficient in CFA with MLR, Schmitt, Sass [38] recommend BSEM to explore source of model misfit. Here, the BSEM models were estimated using the Bayes estimator with a series of prior specifications for cross-loadings and residual correlations with the standardized item scores. All BSEM were conducted with Mplus Version 8.2 [58]. For metrics, we fixed one relatively stable first-order factor loading per factor and set variances of second-order factors at one. First, BSEM models specified noninformative priors for the hypothesized factor loadings, but did not estimated cross-loadings and residual correlations. Next, we specified small-variance informative priors for the cross-loadings, choosing normal prior distributions N (0, 0.01) yielding 95% small cross-loading bounds of ±0.20 [46]. Finally, we added informative Inverse Wishart (dD,d) priors for the residual variances/covariances [41], where D refers to the residual variance/covariance of the Bayesian CFA models and d refers to the degrees of freedom. We used d = 1000 as a starting value; then, we conducted the sequence of sensitivity analyses described in Asparouhov, Muthén [41]. If convergence was fast but model fit was unacceptable (PPp < .05), the next step reduced d (e.g., -100) and repeated the analyses. If slow or no convergence happened, the next step increased d (e.g., + 100) and again repeated the analyses. This sensitivity analysis procedure was intended to change the variance of the small priors to monitor the distance between the data and the model. As explained in Asparouhov and Muthén [63], “In this process no particular prior variance is preferred, rather, the prior variance is adjusted gradually to maintain identifiability of the model while resolving model fit and separating parameters that have minor deviations from zero from substantively important misspecifications” (p. 2).

The BSEM estimation was run with three independent Markov chain Monte Carlo chains using the Gibbs sampler [41, 46], with 150,000 iterations (of which the first 75,000 were discarded as the burn-in phase). No thinning was conducted. Model convergence was monitored by potential scale reduction (a value ≤ 1.10 suggests convergence) and visually checking trace plots. Model fit was evaluated using the posterior predictive p-value (PPp) with associated 95% confidence interval; a PPp < .05 and a positive 95% lower limit imply a poor model fit. The deviance information criterion (DIC) was used for comparison of BSEM models because it is more appropriate than BIC for BSEM [41]; lower DIC suggests better model fit. Moreover, when we used approximately zero priors for cross-loadings and/or residual correlations, prior-posterior predictive p-value (PPPp) was used to test for the hypothesis that a set of parameters are approximately zero [63, 64]; a PPPp < .05 imply that this hypothesis is rejected. All data and Mplus syntaxes needed for analyses are available at https://osf.io/mwpxt/.

Results

Dimensionality analyses and CFA with MLR

Exploratory graph analysis showed that two common factors were recommended in all samples. Next, we conducted CFA with MLR to compare model fit of possible factor structures. Fit indices of each model are shown in Table 1. In all samples, AIC and BIC were lower for the target-based structure than for the unidimensional EC structure. Moreover, an improper solution (the psi matrix is not positive definite) was found for the competence-based structure. This improper solution emerged because some correlation coefficients among second-order factors (e.g., correlation between emotion identification and emotion expression) exceeded 1.00, implying factors were overextracted (for detailed factor loadings, see S1 Table). These results suggest that target-based structure is superior to the unidimensional structure and the competence-based structure.

thumbnail
Table 1. Fit indices of CFA with a robust maximum likelihood estimation.

https://doi.org/10.1371/journal.pone.0225070.t001

For the hybrid structure, some variances were negative, suggesting an improper solution. Lance and Fan [65] indicated that this improper solution usually happens in a hybrid-structure model. To solve this issue, they recommended Rindskopf [66]’s reparameterization, which fixes the variance of the residual at one and estimates the coefficient. Following their recommendation, we applied Rindskopf [66]’s reparameterization to the hybrid model; it returned proper solutions in all samples. Although the model fit of the hybrid structure was inferior to that of the target-based structure, the patterns of second-order factor loadings were interesting: factor loadings from the target-based structure (intrapersonal and interpersonal EC, average factor loadings = .75) were much stronger than those from the competence-based structure (emotion identification, emotion comprehension, emotion expression, emotion regulation, and emotion utilization; average factor loadings = .21; for detailed factor loadings, see S2 Table).

With regards to the modified target-based structure, where competence factors in the hybrid structure were replaced by residual correlations, AIC and BIC were the lowest among the possible models, in all samples. Moreover, as in the competence-based structure, an improper solution (non-positive-definite psi matrix) was found for the modified competence-based structure in all samples, because some correlation coefficients among second-order factors exceeded 1.00, implying that factors were overextracted. Taken together, these results suggest that the modified target-based structure is best to represent the EC factor structure as assessed with the PEC.

Standardized second-order factor loadings deriving from the modified target-based structure are shown in Table 2. All hypothesized major loadings were substantially large (≥ .36) and statistically significant. Moreover, when looking at residual correlations among first-order factors, correlations between regulation-self and regulation-other were substantially large in all samples (rs = .41 to .55). However, although SRMRs (except for sample C) and RMSEAs showed adequate fit, CFIs were not acceptable in all samples even for the best-fitted modified two-second-order-factor model. Therefore, we explored the source of model misfit using BSEM.

thumbnail
Table 2. Results of the CFA with a robust maximum likelihood estimation of the modified target-based structure model.

https://doi.org/10.1371/journal.pone.0225070.t002

BSEM

We conducted BSEM using the modified target-based structure model. Table 3 presents the fit indices of the results. In all samples, BSEM with no informative priors and BSEM with cross-loadings were rejected by the data (PPp ≤ .001), with a high 95% lower PP limit. Therefore, we added informative priors for the residual variances/covariances. When d was set to 1000, BSEM analyses gave PPp values higher than .05 in sample B (.278), but lower than .05 in samples A, C, and D (PPp ≤ .042). Therefore, the next step decreased d by 100 and repeated the analyses with the new d. This procedure was repeated until sufficient model fit was achieved. When d was set to 200, PPp values were greater than 0.05 in all samples (.206 to .660). Thus, we adopted d = 200 to maintain the identifiability of the model while resolving model fit and separating parameters that had minor deviations from zero from substantively important misspecifications.

thumbnail
Table 3. Fit indices of Bayesian structural equation modeling of the modified target-based structure model.

https://doi.org/10.1371/journal.pone.0225070.t003

Potential scale reductions were lower than 1.10 in all samples, and chains indicated clear mixing in trace plots, suggesting good convergence [46]. Following Depaoli and van de Schoot [67], we also checked whether convergence remained after doubling the number of iterations (300,000); potential scale reductions remained lower than 1.10 and deviations of parameters were ≤ |0.02| in all samples, suggesting good convergence. Moreover, PPp values were greater than 0.05 and the 95% PP limit did not include zero in all samples, suggesting good model fit. DIC showed that the model with cross-loadings and residual correlations was superior to the one with only cross-loadings and the one without cross-loadings or residual correlations, in all samples.

Standardized second-order factor loadings and factor correlations of this model (d = 200) are shown in Table 4 (for standardized first-order factor loadings and residual correlations, see S3 Table). All hypothesized major second-order factor loadings were substantively large (≥ .34) and the credible interval did not include zero, except for the loading of utilization-self on intrapersonal EC (factor loadings = .14–.30). As in the results of CFA, intrapersonal and interpersonal EC were significantly correlated with each other (rs = .67–.80). Residual correlations between regulation-self and regulation-other were substantially large in all samples (rs = .39–.55).

thumbnail
Table 4. Results of Bayesian structural equation modeling of the modified target-based structure model (d = 200).

https://doi.org/10.1371/journal.pone.0225070.t004

Next, we looked at newly estimated parameters in BSEM with cross-loadings and residual correlations, to explore what makes the model fit of CFA worse. The results are summarized in Table 5. They suggested that most cross-loadings and residual correlations were substantively small. Indeed, PPPp was more than .05 in all samples, suggesting that the hypothesis that a set of parameters are approximately zero was not rejected (Table 3). Thus, the BSEM analysis suggests that minor cross-loadings and residual correlations contributed to the CFA model misfit.

thumbnail
Table 5. Frequency distribution of the strength of cross-loadings and residual correlations in the model with cross-loadings (prior variances = 0.1) and residual correlations (d = 200).

https://doi.org/10.1371/journal.pone.0225070.t005

Discussion

This study aims to clarify the higher-order structure of the PEC with four different samples (French-speaking Belgian, Dutch-speaking Belgian, Spanish, and Japanese). Dimensionality analyses and CFA with MLR revealed that the modified target-based structure (distinction based on the intrapersonal and interpersonal factors with residual correlations among types of competence) fits best among the possible factor structure models, in all samples. This finding emphasizes the importance of distinguishing between intrapersonal and interpersonal domains in EC. Moreover, the results of BSEM showed that model misfit within the modified target-based structure was caused by minor cross-loadings and residual correlations. Given that the strict constraints of exact-zero cross-loadings and residual correlations are unnecessary in the CFA model [38, 46], these results offer further evidence of the validity of the modified target-based structure.

The importance of distinguishing between intrapersonal and interpersonal domains is consistent with theory in EC-related research areas and other fields in psychology. For example, in the related field of emotion regulation, researchers recently developed a theoretical model assuming that perceiving, understanding, and regulating others’ emotions are related but distinct psychological processes from perceiving, understanding, and regulating one’s own emotions [6870]. More broadly, Leary, Raimi [71] indicated the importance of distinguishing intrapersonal from interpersonal motives in a wide range of psychological phenomena, such as cognitive dissonance, biases in decision-making, and self-conscious emotions. The distinction between intrapersonal and interpersonal domains is increasingly considered as essential to properly understand psychological phenomena.

In the domain of EC, the intrapersonal versus interpersonal higher-order dimensions do more than just provide a parsimonious summary of a complex construct. They are also useful to accurately predict external variables. In fact, previous studies found that intrapersonal and interpersonal EC were differently related to external criteria—for example, intrapersonal EC was more strongly related to objective indices of health [30], depression [33] and regulation of one’s own emotions [36, 72], whereas interpersonal EC was more strongly related to behaviors aimed at regulating others’ negative emotions [7, 73]. These results suggest that intrapersonal versus interpersonal dimensions can afford more nuanced exploration of relationships between EC and external variables and increase its predictive power.

This study has also implications for emotional education. Emotional education refers to an intervention program aimed at improving EC [74]. Recent research showed that relatively short intervention programs can improve trait-level EC [3, 4]. For effective emotional education, implementers should successfully grasp participants’ current level of EC and respond to it. To achieve this goal, the intrapersonal versus interpersonal EC dimensions will be useful to analyze the characteristics of participants’ EC profiles and design tailored intervention to foster it. Recent research has strongly called for theory-based EC intervention program that is designed according to a theoretical model of EC [74, 75]. The current results suggest that intrapersonal versus interpersonal dimensions can contribute to this line of research by better differentiating among individuals with different EC profiles and providing a useful framework for designing better emotional education content.

The present study revealed that competence-based factors should be replaced by residual correlations. Nevertheless, among competences, residual correlations between regulation-self and regulation-other were significant and large after controlling for intrapersonal and interpersonal factors in all samples. This may reflect the fact that individual differences in regulation of one’s own emotions are positively associated with those in regulation of another person’s emotions. For example, Niven, Totterdell [76] revealed that individual differences in intrinsic affect-improving (the extent to which an individual typically engages in up-regulation of their own emotions) and extrinsic affect-improving (the extent to which that individual typically engages in up-regulation of another person’s emotions) were differentiated but positively associated with each other. Such a positive relationship may be represented as significant residual correlation between regulation-self and regulation-other in the modified target-based structure.

We also found that utilization-self did not significantly load on intrapersonal EC in BSEM results, unexpectedly. Several previous studies have also found that facilitating thought using emotions—which is a competence related to utilizing one’s own emotions—does not reliably emerge in the factor analysis and is not conceptually distinct from the other competences [77]. For example, factor loadings of the facilitating thought using emotions branch were negligible and not statistically significant beyond the general factor [78]. As discussed in Mayer, Caruso [79], this may be because people utilize their emotions by their emotion comprehension competence (or another competence) rather than any competence distinctly related to facilitating thought. More research is needed to confirm the position of utilization-self in EC.

Alongside its strengths, several limitations of this study have to be acknowledged. First, our results are based on self-report measures of EC. Although self-reports are the most widely used method to measure traits and although they have shown evidence of both theoretical and empirical validity [8, 27, 44], traits—including trait-level EC—can also be assessed through observer ratings [80]. Future research should investigate whether the current results can be generalized to alternative methods. Second, given that construct validation is an ongoing process [81], future research should gather further construct validity evidence such as convergent, discriminant, and predictive validity of the PEC.

Despite these limitations, these findings show the importance of distinguishing between intrapersonal and interpersonal domains in EC. This insight sheds new light on the factor structure of the PEC and opens exciting perspectives for future research.

Supporting information

S1 Table. Results of the CFA with a robust maximum likelihood estimation of the competence-based structure model.

https://doi.org/10.1371/journal.pone.0225070.s001

(PDF)

S2 Table. Results of the CFA with a robust maximum likelihood estimation of the hybrid structure model.

https://doi.org/10.1371/journal.pone.0225070.s002

(PDF)

S3 Table. Results of the Bayesian structural equation modeling of the modified two second-order factor model with cross-loadings and residual correlations.

https://doi.org/10.1371/journal.pone.0225070.s003

(PDF)

S1 Text. Details and previous research applications of the hybrid structure model.

https://doi.org/10.1371/journal.pone.0225070.s004

(PDF)

Acknowledgments

We thank Hervé Avalosse, Rebecca Verniest and Sigrid Vancoreland from the R&D Department from the Mutualité Chrétienne-Christelijke Mutualiteit for their help in the data collection.

References

  1. 1. Saarni C. Emotional competence: How emotions and relationships become integrated. In: Thompson RA, editor. Nebraska Symposium on Motivation, 1988: Socioemotional development. Lincoln, NE: University of Nebraska Press; 1990. p. 115–82.
  2. 2. Hodzic S, Scharfen J, Ripoll P, Holling H, Zenasni F. How efficient are emotional intelligence trainings: A meta-analysis. Emot Rev. 2018; 10: 138–48.
  3. 3. Kotsou I, Nelis D, Gregoire J, Mikolajczak M. Emotional plasticity: Conditions and effects of improving emotional competence in adulthood. J Appl Psychol. 2011; 96: 827–39. pmid:21443316
  4. 4. Nelis D, Kotsou I, Quoidbach J, Hansenne M, Weytens F, Dupuis P, et al. Increasing emotional competence improves psychological and physical well-being, social relationships, and employability. Emotion. 2011; 11: 354–66. pmid:21500904
  5. 5. Szczygiel D, Mikolajczak M. Is it enough to be an extrovert to be liked? Emotional competence moderates the relationship between extraversion and peer-rated likeability. Front Psychol. 2018; 9: 804. pmid:29875728
  6. 6. Nozaki Y, Koyasu M. Can we apply an emotional competence measure to an eastern population? Psychometric properties of the Profile of Emotional Competence in a Japanese population. Assessment. 2016; 23: 112–23. pmid:25670840
  7. 7. Nozaki Y. Emotional competence and extrinsic emotion regulation directed toward an ostracized person. Emotion. 2015; 15: 763–74. pmid:25938611
  8. 8. Keefer KV. Self-report assessments of emotional competencies: A critical look at methods and meanings. J Psychoeduc Assess. 2015; 33: 3–23.
  9. 9. Mayer JD, Salovey P. What is emotional intelligence? In: Salovey P, Sluyter D, editors. Emotional development and emotional intelligence: Educational implications. New York, NY: Basic Books; 1997. p. 3–31.
  10. 10. Petrides KV, Furnham A. Trait emotional intelligence: Psychometric investigation with reference to established trait taxonomies. Eur J Personality. 2001; 15: 425–48.
  11. 11. Bar-On R. The emotional intelligence inventory (EQ-i): Technical manual. Toronto, ON, Canada: Multi-Health Systems.; 1997.
  12. 12. Mikolajczak M, Petrides KV, Coumans N, Luminet O. The moderating effect of trait emotional intelligence on mood deterioration following laboratory-induced stress. Int J Clin Health Psychol. 2009; 9: 455–77.
  13. 13. Lumley MA, Gustavson BJ, Partridge RT, Labouvie-Vief G. Assessing alexithymia and related emotional ability constructs using multiple methods: interrelationships among measures. Emotion. 2005; 5: 329–42. Epub 2005/09/29. pmid:16187868
  14. 14. Petrides KV, Pita R, Kokkinaki F. The location of trait emotional intelligence in personality factor space. Br J Psychol. 2007; 98: 273–89. pmid:17456273
  15. 15. Harms PD, Credé M. Emotional intelligence and transformational and transactional leadership: A meta-analysis. J Leadersh Organ Stud. 2010; 17: 5–17.
  16. 16. Martins A, Ramalho N, Morin E. A comprehensive meta-analysis of the relationship between emotional intelligence and health. Pers Indiv Differ. 2010; 49: 554–64.
  17. 17. Miao C, Humphrey RH, Qian S. Are the emotionally intelligent good citizens or counterproductive? A meta-analysis of emotional intelligence and its relationships with organizational citizenship behavior and counterproductive work behavior. Pers Indiv Differ. 2017; 116: 144–56.
  18. 18. Sánchez-Álvarez N, Extremera N, Fernández-Berrocal P. The relation between emotional intelligence and subjective well-being: A meta-analytic investigation. J Posit Psychol. 2016; 11: 276–85.
  19. 19. Schutte NS, Malouff JM, Thorsteinsson EB, Bhullar N, Rooke SE. A meta-analytic investigation of the relationship between emotional intelligence and health. Pers Indiv Differ. 2007; 42: 921–33.
  20. 20. Peña-Sarrionandia A, Mikolajczak M, Gross JJ. Integrating emotion regulation and emotional intelligence traditions: A meta-analysis. Front Psychol. 2015; 6: 160. pmid:25759676
  21. 21. Sarrionandia A, Mikolajczak M. A meta-analysis of the possible behavioural and biological variables linking trait emotional intelligence to health. Health Psychol Rev. 2019: Advance online publication. pmid:31284846
  22. 22. Richardson M, Abraham C, Bond R. Psychological correlates of university students' academic performance: A systematic review and meta-analysis. Psychol Bull. 2012; 138: 353–87. pmid:22352812
  23. 23. Miao C, Humphrey RH, Qian S. A meta-analysis of emotional intelligence and work attitudes. J Occup Organ Psychol. 2017; 90: 177–202.
  24. 24. Miao C, Humphrey RH, Qian S. A meta-analysis of emotional intelligence effects on job satisfaction mediated by job resources, and a test of moderators. Pers Indiv Differ. 2017; 116: 281–8.
  25. 25. Malouff JM, Schutte NS, Thorsteinsson EB. Trait emotional intelligence and romantic relationship satisfaction: A meta-analysis. Am J Fam Ther. 2014; 42: 53–66.
  26. 26. Brackett M A., Rivers S E., Salovey P. Emotional intelligence: Implications for personal, social, academic, and workplace success. Soc Personal Psychol Compass. 2011; 5: 88–103.
  27. 27. Petrides KV, Mikolajczak M, Mavroveli S, Sanchez-Ruiz MJ, Furnham A, Perez-Gonzalez JC. Developments in trait emotional intelligence research. Emot Rev. 2016; 8: 335–41.
  28. 28. Brasseur S, Gregoire J, Bourdu R, Mikolajczak M. The Profile of Emotional Competence (PEC): Development and validation of a self-reported measure that fits dimensions of emotional competence theory. PloS one. 2013; 8: e62635. pmid:23671616
  29. 29. Parker JD, Michael Bagby R, Taylor GJ, Endler NS, Schmitz P. Factorial validity of the 20-item Toronto Alexithymia Scale. Eur J Personality. 1993; 7: 221–32.
  30. 30. Mikolajczak M, Avalosse H, Vancorenland S, Verniest R, Callens M, van Broeck N, et al. A nationally representative study of emotional competence and health. Emotion. 2015; 15: 653–67. pmid:25893449
  31. 31. Min MC, Takai J. The effect of emotional competence on relational quality: Comparing Japan and Myanmar. Curr Psychol. 2018: Advance online publication.
  32. 32. Min MC, Islam MN, Wang L, Takai J. Cross-cultural comparison of university students’ emotional competence in Asia. Curr Psychol. 2018: Advance online publication.
  33. 33. Batselé E, Stefaniak N, Fantini-Hauwel C. Resting heart rate variability moderates the relationship between trait emotional competencies and depression. Pers Indiv Differ. 2019; 138: 69–74.
  34. 34. Constant E, Christophe V, Bodenmann G, Nandrino J-L. Attachment orientation and relational intimacy: The mediating role of emotional competences. Curr Psychol. 2018: Advance online publication.
  35. 35. Kotsou I, Leys C, Fossionc P. Acceptance alone is a better predictor of psychopathology and well-being than emotional competence, emotion regulation and mindfulness. J Affect Disorders. 2018; 226: 142–5. pmid:28972931
  36. 36. Nozaki Y. Cross-cultural comparison of the association between trait emotional intelligence and emotion regulation in European-American and Japanese populations. Pers Indiv Differ. 2018; 130: 150–5.
  37. 37. Fantini-Hauwel C, Mikolajczak M. Factor structure, evolution, and predictive power of emotional competencies on physical and emotional health in the elderly. J Aging Health. 2014; 26: 993–1014. pmid:24920650
  38. 38. Schmitt TA, Sass DA, Chappelle W, Thompson W. Selecting the "Best" factor structure and moving measurement validation forward: An illustration. J Pers Assess. 2018; 100: 345–62. pmid:29630411
  39. 39. Howard JL, Gagné M, Morin AJS, Forest J. Using bifactor exploratory structural equation modeling to test for a continuum structure of motivation. J Manage. 2018; 44: 2638–64.
  40. 40. Wu C-H, Chen LH. Examining dual meanings of items in 2 × 2 Achievement Goal Questionnaires through MTMM modeling and MDS approach. Educ Psychol Meas. 2009; 70: 305–22.
  41. 41. Asparouhov T, Muthén B, Morin AJS. Bayesian structural equation modeling with cross-loadings and residual covariances: Comments on Stromeyer et al. J Manage. 2015; 41: 1561–77.
  42. 42. Marsh HW, Bailey M. Confirmatory factor analyses of Multitrait-Multimethod data: A comparison of alternative models. Appl Psychol Meas. 1991; 15: 47–70.
  43. 43. Hopwood CJ, Donnellan MB. How should the internal structure of personality inventories be evaluated? Pers Soc Psychol Rev. 2010; 14: 332–46. pmid:20435808
  44. 44. Costa PT Jr., McCrae RR. Revised NEO Personality Inventory (NEO-PI-R) and NEO Five-Factor Inventory (NEO- FFI) professional manual. Odessa, FL: Psychological Assessment Resources; 1992.
  45. 45. Marsh HW, Muthén B, Asparouhov T, Lüdtke O, Robitzsch A, Morin AJS, et al. Exploratory structural equation modeling, integrating CFA and EFA: Application to students' evaluations of university teaching. Struct Equ Model. 2009; 16: 439–76.
  46. 46. Muthén B, Asparouhov T. Bayesian structural equation modeling: A more flexible representation of substantive theory. Psychol Methods. 2012; 17: 313–35. pmid:22962886
  47. 47. de Beer LT, Bianchi R. Confirmatory factor analysis of the Maslach Burnout Inventory: A Bayesian structural equation modeling approach. Eur J Psychol Assess. 2019; 35: 217–24.
  48. 48. Dombrowski SC, Golay P, McGill RJ, Canivez GL. Investigating the theoretical structure of the DAS-II core battery at school age using Bayesian structural equation modeling. Psychol Schools. 2018; 55: 190–207.
  49. 49. Fong TCT, Ho RTH. Factor analyses of the Hospital Anxiety and Depression Scale: A Bayesian structural equation modeling approach. Qual Life Res. 2013; 22: 2857–63. pmid:23670233
  50. 50. Fong TCT, Ho RTH. Dimensionality of the 9-item Utrecht Work Engagement Scale revisited: A Bayesian structural equation modeling approach. J Occup Health. 2015; 57: 353–8. pmid:25958976
  51. 51. Golay P, Reverte I, Rossier J, Favez N, Lecerf T. Further insights on the French WISC-IV factor structure through Bayesian structural equation modeling. Psychol Assess. 2013; 25: 496–508. pmid:23148651
  52. 52. Reis D. Further insights into the German version of the Multidimensional Assessment of Interoceptive Awareness (MAIA): Exploratory and Bayesian structural equation modeling approaches. Eur J Psychol Assess. 2019; 35: 317–25.
  53. 53. Elasy TA, Gaddy G. Measuring subjective outcomes: Rethinking reliability and validity. J Gen Intern Med. 1998; 13: 757–61. pmid:9824522
  54. 54. Golino HF, Epskamp S. Exploratory graph analysis: A new approach for estimating the number of dimensions in psychological research. PloS one. 2017; 12: e0174035. pmid:28594839
  55. 55. Golino HF, Demetriou A. Estimating the dimensionality of intelligence like data using Exploratory Graph Analysis. Intelligence. 2017; 62: 54–70.
  56. 56. Golino HF. EGA: Exploratory Graph Analysis: Estimating the number of dimensions in psychological data. R package version 0.4 ed2019.
  57. 57. R Core Team. R 3.5.0: A language and environment for statistical computing. R Foundation for Statistical Computing; 2018.
  58. 58. Muthén LK, Muthén BO. Mplus user's guide. Eighth Edition. Los Angeles, CA: Muthén & Muthén; 1998–2017.
  59. 59. Li CH. Confirmatory factor analysis with ordinal data: Comparing robust maximum likelihood and diagonally weighted least squares. Behav Res Methods. 2016; 48: 936–49. pmid:26174714
  60. 60. Elliot AJ, Murayama K. On the measurement of achievement goals: Critique, illustration, and application. J Educ Psychol. 2008; 100: 613–28.
  61. 61. Brown TA. Confirmatory factor analysis for applied research. New York, NY: Guilford Press; 2006.
  62. 62. Hu LT, Bentler PM. Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Struct Equ Model. 1999; 6: 1–55.
  63. 63. Asparouhov T, Muthén B. Prior-Posterior Predictive P-values (Mplus Web Notes: No. 22). 2017.
  64. 64. Hoijtink H, van de Schoot R. Testing small variance priors using prior-posterior predictive p values. Psychol Methods. 2018; 23: 561–9. pmid:28368177
  65. 65. Lance CE, Fan Y. Convergence, admissibility, and fit of alternative confirmatory factor analysis models for MTMM data. Educ Psychol Meas. 2015; 76: 487–507. pmid:29795875
  66. 66. Rindskopf D. Parameterizing inequality constraints on unique variances in linear structural models. Psychometrika. 1983; 48: 73–83.
  67. 67. Depaoli S, van de Schoot R. Improving transparency and replication in Bayesian statistics: The WAMBS-Checklist. Psychol Methods. 2017; 22: 240–61. pmid:26690773
  68. 68. Reeck C, Ames DR, Ochsner KN. The social regulation of emotion: An integrative, cross-disciplinary model. Trends Cogn Sci. 2016; 20: 47–63. pmid:26564248
  69. 69. Zaki J, Williams WC. Interpersonal emotion regulation. Emotion. 2013; 13: 803–10. pmid:24098929
  70. 70. Nozaki Y, Mikolajczak M. Extrinsic emotion regulation. Emotion. 2019: Advance online publication.
  71. 71. Leary MR, Raimi KT, Jongman-Sereno KP, Diebels KJ. Distinguishing intrapsychic from interpersonal motives in psychological theory and research. Perspect Psychol Sci. 2015; 10: 497–517. pmid:26177950
  72. 72. Pekaar KA, Bakker AB, Born MP, van der Linden D. The consequences of self- and other-focused emotional intelligence: Not all sunshine and roses. J Occup Health Psychol. 2019; 24: 450–66. pmid:30284844
  73. 73. Nozaki Y, Koyasu M. The relationship between trait emotional intelligence and interaction with ostracized others' retaliation. PloS one. 2013; 8: e77579. pmid:24194890
  74. 74. Pérez-González JC, Qualter P. Emotional intelligence and emotional education in the school years. In: Dacre Pool L, Qualter P, editors. An introduction to emotional intelligence. England: John Wiley & Sons; 2018. p. 81–104.
  75. 75. Kotsou I, Mikolajczak M, Heeren A, Grégoire J, Leys C. Improving emotional intelligence: A systematic review of existing work and future challenges. Emot Rev. 2019; 11: 151–65.
  76. 76. Niven K, Totterdell P, Stride CB, Holman D. Emotion regulation of others and self (EROS): The development and validation of a new individual difference measure. Curr Psychol. 2011; 30: 53–73.
  77. 77. Joseph DL, Newman DA. Emotional intelligence: An integrative meta-analysis and cascading model. J Appl Psychol. 2010; 95: 54–78. pmid:20085406
  78. 78. Palmer BR, Gignac G, Manocha R, Stough C. A psychometric evaluation of the Mayer–Salovey–Caruso Emotional Intelligence Test Version 2.0. Intelligence. 2005; 33: 285–305.
  79. 79. Mayer JD, Caruso DR, Salovey P. The ability model of emotional intelligence: Principles and updates. Emot Rev. 2016; 8: 290–300.
  80. 80. Elfenbein HA, Barsade SG, Eisenkraft N. The social perception of emotional abilities: Expanding what we know about observer ratings of emotional intelligence. Emotion. 2015; 15: 17–34. pmid:25664949
  81. 81. Flake JK, Pek J, Hehman E. Construct validation in social and personality research: Current practice and recommendations. Soc Psychol Pers Sci. 2017; 8: 370–8.