The importance of trait emotional intelligence and feelings in the prediction of perceived and biological stress in adolescents: hierarchical regressions and fsQCA models

Abstract

The purpose of this study is to analyze the combined effects of trait emotional intelligence (EI) and feelings on healthy adolescents’ stress. Identifying the extent to which adolescent stress varies with trait emotional differences and the feelings of adolescents is of considerable interest in the development of intervention programs for fostering youth well-being. To attain this goal, self-reported questionnaires (perceived stress, trait EI, and positive/negative feelings) and biological measures of stress (hair cortisol concentrations, HCC) were collected from 170 adolescents (12–14 years old). Two different methodologies were conducted, which included hierarchical regression models and a fuzzy-set qualitative comparative analysis (fsQCA). The results support trait EI as a protective factor against stress in healthy adolescents and suggest that feelings reinforce this relation. However, the debate continues regarding the possibility of optimal levels of trait EI for effective and adaptive emotional management, particularly in the emotional attention and clarity dimensions and for female adolescents.

Keywords:

• Trait emotional intelligence
• feelings
• perceived stress
• hair cortisol
• adolescents
• fsQCA models

Introduction

Trait emotional intelligence (EI) is “a constellation of behavioral dispositions and self-perceptions concerning one’s ability to recognize, process and utilize emotion-laden information” (Petrides, Frederickson, & Furnham, 2004, p. 278). Dispositions from the personality domain (empathy and assertiveness) and self-perceptions concerning our social and personal intelligence composed this construct. Therefore, trait EI has traditionally been related to psychological well-being in adults (Fernández-Berrocal, Extremera, & Ramos, 2004; Kong & Zhao, 2013) and non-healthy participants (Flynn & Rudolph, 2014; Van Uum et al., 2008). However, research with healthy adolescents remains scarce.

This direct relation between trait EI and various well-being indicators may also be influenced by affective variables such as moods (Panno, Donati, Chiesi, & Primi, 2015), affects (Kong & Zhao, 2013), or feelings; feelings conceptualize the more personal and biographical subjectivity term (Munezero, Montero, Sutinen, & Pajunen, 2014) and is thus used in this study.

The proposed directionality of the relation between trait EI, feelings, and stress (as an indicator of the absence of well-being) is supported by the following theory: according to Lazarus and Folkman’s (1984) transactional stress model, appraisal to assess a situation is based partially on our personal resources (such as trait EI and feelings). This appraisal may promote inefficient coping with regard to the situation, leading to a stress response. This working scheme, the contribution of trait EI and feelings to adolescents’ stress, is the representation assumed in this study.

The origin of the stress may be psychological or physiological (Russell, Koren, Rieder, & Van Uum, 2012). In the first case, the results indicated that adolescents with greater emotional clarity and repair demonstrated lower perceived stress (Extremera, Durán, & Rey, 2007; Flynn & Rudolph, 2014). In the case of physiological stress, most studies have analyzed cortisol responses using samples of saliva, urine, or serum (Mikolajczak, Roy, Luminet, Fillée, & de Timary, 2007; Noppe et al., 2014). However, these types of measures assess acute stress. Hair cortisol concentrations (HCC) capture systemic cortisol exposure over longer periods of time (Russell et al., 2012; Vanaelst et al., 2012); thus, HCC is the measure incorporated in this study.

In adults, it appears that high clarity and high repair were associated with low levels of salivary cortisol (Salovey, Stroud, Woolery, & Epel, 2002). As with adults, salivary cortisol activity in adolescents appears to be modified by trait emotional functioning, measured by self-reported depression, anxiety, and anger levels (Adam, 2006). Further studies with specific trait EI questionnaires are required to more deeply examine the possible different forms that associations may assume in adolescents as opposed to adults.

However, both psychological and physiological stress measures appear to present a controversial relation (Oldehinkel et al., 2011). In some studies, HCC correlated positively and significantly with perceived stress (Adam, 2006; Rietschel et al., 2016). However, other studies did not observe this significant correlation in adults (Gidlow, Randall, Gillman, Silk, & Jones, 2016; Van Uum et al., 2008) or observed the correlation to be negative (Maldonado et al., 2008). Thus, it would be interesting to include both types of stress measurements in the same study.

In summary, the primary objective of this study was to examine trait EI and feelings as predictors of biological and perceived stress in a sample of healthy adolescents. We expected that high emotional clarity and repair would be negatively associated with stress. This association will improve with the inclusion of feelings.

Method

Participants

The sample comprised 170 pupils, aged 12–14 years (M = 12.79 years; SD =0.73). The sample comprised 88 female adolescents, accounting for 52.1% of the total. This age range was selected because of the recent entry of these pupils into high school and, therefore, the presence of new stressors related to pressures and expectations within the school environment (de Anda et al., 2000). All of the pupils attended two public or two private high schools in Valencia (Spain), located across a range of working to upper-middle class areas, which were selected using a convenience sampling method. No variability across the schools was observed in students’ scores. The distribution of the parents’ educational levels was as follows: university degree (30%), high school (32%), primary studies (32%), and no education (5%).

Measures

Trait EI was evaluated using the TMMS-24 “Trait Emotional Meta-Mood Scale”, which was adapted and validated in Spain by Fernández-Berrocal et al. (2004) and based on the Trait Meta-Mood Scale by Salovey, Mayer, Goldman, Turvey, and Palfai (1995). The scale assesses meta-knowledge of the three elements composing EI, which include the following: (1) emotional attention (eight items), which is the extent to which people tend to observe and think about their feelings and moods; (2) emotional clarity (eight items), which evaluates the understanding of one’s emotional states; and (3) emotional repair (eight items), which involves the individual’s beliefs regarding the ability to regulate his or her feelings. All 24 items are scored on a scale ranging from 1 (completely disagree) to 5 (completely agree). The TMMS subscales have been demonstrated to have adequate psychometric properties (attention, α = .90; clarity, α = .90; repair, α = .86; Fernández-Berrocal et al., 2004). In this study, the reliability was α = .85 for attention, α = .84 for clarity, and α = .78 for repair.

SPANE, the Scale of Positive and Negative Experiences (Diener et al., 2010), assesses positive and negative feelings. It is a brief 12-item scale, with 6 items devoted to positive experiences (SPANE-P) and 6 items designed to assess negative experiences (SPANE-N) occurring within the past 4 weeks. Each SPANE item is scored on a scale ranging from 1 to 5, where 1 represents “very rarely or never” and 5 represents “very often or always”. The two scores can be combined by subtracting the negative score from the positive score, obtaining the hedonic balanced score (SPANE-B). The scale has been demonstrated to have good validity and reliability (SPANE-B α = .89) (Diener et al., 2010), which were also observed in this study (SPANE-B α = .89).

The PSS-4 Perceived Stress Scale (Cohen & Williamson, 1988; Spanish version of Herrero & Meneses, 2006) is a brief version of the Perceived Stress Scale (PSS) (Cohen, Kamarck, & Mermelstein, 1983). This self-reported questionnaire comprises four items that evaluate the degree to which individuals believe their lives have been unpredictable, uncontrollable and overloaded in the previous month. It has a four-point response scale, for example: “I never/sometimes/often/always felt that I was unable to control the important things in my life”. A higher score indicates a greater presence of perceived stress. Previous research has demonstrated this scale to be reliable and valid (Lee, 2012) with acceptable alpha values in Spanish children (α = .68), (Herrero & Meneses, 2006). In this study, the reliability was α = .68.

To determine HCC, a lock of hair from each participant proximal 3-cm hair segment was weighed (25–40 mg, approximately 150 hair strands) and cut from the back of the head, as close to the scalp as possible using small surgical scissors. This area has been identified as having the most stable HCC. The locks of hair were taped to a paper form with scalp end markings and stored in individual sets at room temperature until analysis. The hair was extracted in 1 ml methanol shaken slightly for 36 h at 37.7 °C. The extract was transferred to a glass tube and the methanol was evaporated under a constant stream of nitrogen and reconstituted in 250 μl phosphate-buffered saline pH 8.0. Cortisol quantification in the hair was analyzed using a direct enzyme-linked immunosorbent (ELISA) test using the Salivary Cortisol ELISA SLV-2930 kit (DRG International Inc., Springfield, NJ, USA), following the manufacturer’s directions. Cortisol was evaluated by a solid-phase, competitive chemiluminescent enzyme immunoassay (Immulite, Siemens, Munich, Germany), displaying within-run and between-run imprecision lower than 10%, recovery rates between 92% and 120%, and a limit of detection of 0.2 nmol/l (Tecles et al., 2014). No intra- and inter-day fluctuations of cortisol were calculated as they are mitigated when using hair cortisol (Groschl, 2008; Wosu et al., 2015). Participants taking medications that may affect the hypothalamic–pituitary–adrenal (HPA) axis or with any illness in which an abnormal production of cortisol may be present (primarily patients with Cushing’s and Addison syndrome) were excluded from hair collection.

Procedure

The above-mentioned questionnaires were administered collectively during a regular class lasting approximately 1 h. The effect of the question order in the questionnaires was controlled. Simultaneously, hair cortisol samples were collected individually. Before collection and during the interview to obtain parents’ consent, parents were asked about their child’s medications that may affect the HPA axis or any illness with an abnormal production of cortisol (three participants were excluded because of these criteria). The necessary consents from the county government (Conselleria d'Educació, Investigació, Cultura i Esport), schools, parents, and the ethical university commission from Valencia University (code H1385330676977) were obtained before recruiting the adolescents.

Data analysis

First, descriptive analysis and correlations among variables were calculated. Next, the effects of age, gender, trait EI, feelings regarding perceived stress, and levels of hair cortisol were analyzed using two different methodologies. Two hierarchical regression models and a fuzzy-set qualitative comparative analysis (fsQCA), which allows for conjunctions of all logically possible combinations of conditions (Sereikhuoch & Woodside, 2012), were run. The first strategy is based on linear models and is focused on the individual contribution of each variable, whereas the second strategy, qualitative comparative analysis (QCA), allows for an in-depth analysis of how causal conditions contribute to an outcome. QCA assumes that the influence of a particular attribute on a specific outcome depends on a combination of attributes rather than on individual levels of the attributes (Boquera, Martínez-Rico, Pérez-Campos, & Prado-Gascó, 2016; Calabuig, Prado-Gascó, Crespo-Hervás, Núñez-Pomar, & Añó, 2016; Eng & Woodside, 2012; Ragin, 2008). Conversely, latent profile analyzes search for similar cases, as does cluster analysis, whereas QCA searches for combinations, patterns, or recipes that can explain a particular output. QCA also considers equifinality, that is, different ways to arrive at a particular result.

In general, one of the advantages of QCA from linear models is the possibility of addressing multiple conjunctural causations straightforwardly. Addressing these causations in a straightforward manner is particularly important when theory suggests that there can be more than one manner in which to achieve a particular output. Another advantage is the possibility of identifying the combinations of multiple causes. In regression analysis, there is a limit to the number of interaction effects that can be included in one analysis but not in QCA. In addition, the results of the fsQCA analysis are more detailed and allow a more horizontal complexity than does regression analysis (Seawright, 2005; Vis, 2012). QCA also offers a more systematic way to analyze complex causality and logical relations between causal conditions and an outcome than linear models (Legewie, 2013). Finally, QCA analysis may be performed with small samples (Eng & Woodside, 2012; Ragin, 2008). In general, as is demonstrated in the literature, the two methodologies are complementary (Boquera, Martínez-Rico, Pérez-Campos, & Prado-Gascó, 2016; Calabuig, Prado-Gascó, Crespo-Hervás, Núñez-Pomar, & Añó, 2016; Seawright, 2005; Vis, 2012). To perform fsQCA, raw data responses are transformed into fuzzy-set responses. First, all missed data are deleted, and all constructs (variables) are calculated by multiplying their item scores. Before performing the analysis, the analysis values must be recalibrated between 0 and 1. The recalibration is quite important because it may affect the final result, indicating more or fewer observations or participants that achieved a particular output. When we consider only two values, we proceed with 0 (not having the characteristic, fully outside the set) and 1 (having the characteristic, fully in the set). However, to perform the recalibration with more than two values, we must consider the following three thresholds: the first one (0) considers that an observation with this value is fully outside the set (low agreement); the second one (0.5) considers a median point, neither inside nor outside the set (intermediate level of agreement); and the last value (1) considers the observation to be fully in the set (high level of agreement). This process was the direct method of calibration proposed by the author of the methodology, Ragin (Ragin, 2008), and it is the most used on literature (Barton & Beynon, 2015; Felício, Rodrigues, & Samagaio, 2016; Hsieh, Chang, Huang, & Chen, 2016; McNamara, 2015; Primc & Cater, 2015; Rey-Martí, Ribeiro-Soriano, & Palacios-Marqués, 2016; Schneider & Wagemann, 2012; Stockemer, 2013; Woodside, 2013). With continuous variables or with factors from a survey (formed by different items), we must introduce these three values to proceed to an automatic recalibration of values between 0 and 1. In these cases, the literature suggests that with continuous variables or with factors, the three thresholds must be percentiles 10, 50, and 90 (Woodside, 2013). Thus, values of each variable (except gender, 0 = women and 1 = men, and age, 0 [12], 0.49 [13], and 1 [14]) were recalibrated, considering the following three thresholds: percentile 10 (low agreement or fully outside the set), percentile 50 (intermediate level of agreement, neither inside nor outside the set) and percentile 90 (high agreement or fully in the set). Finally, necessary and sufficient conditions tests evaluate the effects of different variables on a particular outcome (PSS and cortisol levels) and on the absence of the output (PSS and cortisol levels). A sufficient condition expresses a combination of conditions that can produce a particular outcome although that particular outcome can be achieved by other combinations of conditions. Conversely, a condition is necessary when it must always be present for the occurrence of a particular outcome. According to Sereikhuoch and Woodside (2012), to calculate sufficient conditions, the fsQCA analysis involves two stages. First, a truth table algorithm transforms the fuzzy-set membership scores into a truth table that lists all logically possible combinations of causal conditions and each configuration’s empirical outcome. Second, fsQCA analysis generates the following three possible solutions: complex, parsimonious, and intermediate solutions. The complex solution is the most restrictive, and the parsimonious solution is the least restrictive. Previous studies (Ragin, 2008) suggest including the intermediate solution (the solution that is presented here). When considering a sufficient analysis, solution coverage considers variance explained (number of observations that can be explained by a particular combination of conditions), whereas solution consistency expresses a model’s possible reliability. In addition, when we consider each condition, raw coverage indicates how many cases or observations can be explained by the conditions (variance explained). Conversely, the unique coverage expresses the number of observations (variance) that can be explained by a particular combination of conditions but not by other combination of conditions. To choose the most important condition, we must consider the raw coverage. Regarding necessary analysis and similar to sufficient analysis, the consistency indicates the adequacy of the condition to predict a particular outcome (≥0.90), whereas coverage considers variance explained by a condition (Ragin, 2008). SPSS 23 (SPSS, Chicago, IL) was used to perform regression analysis, and fsQCA 2.0 software was used to perform fsQCA (Department of Sociology, Tucson, Arizona).

Results

Table 1 shows descriptive statistics and calibration values for the variables under study.

Table 1. Descriptive statistics and calibration values.

			TMMS24
	Cortisol	SPANE-B	Attention	Clarity	Repair	PSS
Mean	3.26	10.04	23.22	24.36	27.17	8.14
SD	1.87	7.99	6.55	6.40	6.09	2.15
Minimum	0.07	−13	4	8	11	4
Maximum	9.54	24	39	38	40	14
Calibration values
Percentile 10	0.92	185,457.6	90.6	148.2	332.8	2
Percentile 50	3.02	9,164,250	2304	6912	13824	16
Percentile 90	5.85	101,718,750	60,000	7344	96,000	54

SPANE-B: hedonic balance; TMMS-24: Trait Emotional Meta-Mood Scale; PSS: Perceived Stress Scale.

The correlations between study variables are presented in Table 2. Age did not yield significant correlations with the variables in the study, with the exception of emotional clarity. Perceived stress (PSS) demonstrated significant and negative correlations with emotional clarity, repair, and hedonic balance (SPANE-B). Hair cortisol levels, however, did not yield significant correlations with any variable in the study. Neither stress measure (perceived and hair cortisol levels) presented significant correlations with the other. Finally, all trait EI components were correlated; however, only clarity and repair correlated positively and significantly with hedonic balance.

Table 2. Correlation between age and the variables under study.

	1	2	3	4	5	6	7
1. Age	1
2. PSS	−.06	1
3. Cortisol	.02	.11	1
4. Attention	.14	−.01	.15	1
5. Clarity	.19*	−.36**	.00	.41**	1
6. Repair	.06	−.25**	.02	.25**	.37**	1
7. Hedonic balance	.13	−.53**	−.09	.11	.34**	.47**	1

PSS: Perceived Stress Scale.

* p ≤ .05.

** p ≤ .01.

Predictive analyses of study variables on cortisol levels and perceived stress were performed by two hierarchical regressions (Table 3). In the last step, the cortisol level was added in the prediction of PSS, and PSS was added in the prediction of cortisol levels. In the prediction of perceived stress, five steps were established in the model (R² = .30). First, age and gender were entered (ΔR² = .004, p = .40 and ΔR² = .00, p = .89, respectively), followed by the entry of trait EI components (ΔR² = .18, p ≤ .001), hedonic balance (ΔR² = .15, p ≤ .001) and, finally, cortisol level (ΔR² = .002, p = .47). Thus, the final model indicates that emotional clarity (β= −.27; p ≤ .001) and hedonic balance (β= −.45; p ≤ .001) may predict levels of perceived stress in a negative and significant manner.

Table 3. Hierarchical regression analyzes for prediction of PSS and HCC.

	Criterion variables
	Perceived stress		Cortisol levels
Predictors	ΔR^2	β	ΔR^2	β
Step 1	.004		.000
Age		−.07		.01
Step 2	.000		.16***
Age		−.07		.03
Gender		.01		−.39***
Step 3	.18***		.02
Age		−.01		.03
Gender		.03		−.38***
TMMS24
Attention		.17*		.17*
Clarity		−.38***		−.11
Repair		−.17*		.01
Step 4	.15***		.011
Age		.02		.04
Gender		−.01		−.39***
Attention		.11		.15
Clarity		−.27***		−.08
Repair		.03		.06
SPANE
Hedonic balance		−.46***		−.12
Step 5		.002	.003
Age		.02		.04
Gender		.01		−.39***
Attention		.10		.14
Clarity		−.27***		−.06
Repair		.03		.06
Hedonic balance		−.45***		−.09
Cortisol level/PSS		.05		.06
Total
R^2 adj	.30***		.16***

* p < .05.

*** p < .001.

Regarding the prediction of HCC, five steps were established for the model (R² = .16), (Table 3). First, age and gender were included in the equation (ΔR² = .00, p = .89; ΔR² = .16, p ≤ .001, respectively). In the third step, all of the trait EI components, that is, attention, clarity, and repair, were added to the equation as independent variables. In the fourth step, hedonic balance was included (ΔR² = .01, p = .16), and in the final step, perceived stress was entered (ΔR² = .003, p = .47). In the final model, only gender (β = −.39; p ≤ .001) appeared to be a significant predictor of hair cortisol levels. That is, the male gender was related to higher levels of HCC.

Fuzzy-set qualitative comparative analysis

The prediction of cortisol levels and perceived stress by means of fsQCA models was also tested. We began by testing whether any of the causal conditions can be considered a necessary condition for a high stress level and for a low stress level (considering cortisol and perceived stress). Then, we analyzed sufficient conditions for perceived stress (Table 4) and for cortisol (Table 5).

Table 4. Findings from fsQCA intermediate solution for gender, age, hedonic balance, attention, clarity, and repair for the high level of perceived stress.

Frequency cutoff: 1; consistency cutoff: 0.800523; all variables are present with the exception of hedonic balance	Raw coverage	Unique coverage	Consistency
∼Repair * Attention *∼Hedonic balance * Younger * Women	0.178714	0.064576	0.822127
Repair *∼Clarity *∼Hedonic balance * Women	0.154265	0.040659	0.758328
∼Attention * Hedonic balance * Older * Women	0.125963	0.039862	0.697620
∼Repair * Clarity *∼Attention * Younger * Women	0.119851	0.019134	0.789842
Repair * Attention *∼Hedonic balance * Younger * Men	0.101648	0.101116	0.825243

∼: Low level; solution coverage: 0.417087; solution consistency: 0.747754.

Table 5. Findings from fsQCA intermediate solution for gender, age, perceived stress, hedonic balance, attention, clarity, and repair for the high level of cortisol.

Frequency cutoff: 1; consistency cutoff: 0.886170; all variables are present with the exception of hedonic balance	Raw coverage	Unique coverage	Consistency
Attention * Women	0.426874	0.101818	0.812528
Clarity * Women	0.360223	0.057859	0.782855
Perceived stress * Hedonic balance * Women	0.206368	0.016752	0.839536
Repair * Perceived stress * Women	0.181062	0.003921	0.833242

∼: Low level; solution coverage: 0.530236; solution consistency: 0.771345.

Considering the necessary conditions test, none of the variables is a necessary condition for the presence of high or low levels of perceived stress or cortisol because all consistency values were under 0.90 (Ragin, 2008). Later, with regard to sufficient conditions, the combination of conditions that led to high levels of perceived stress (Table 4) and physiologic stress (Table 5) were calculated. Regarding sufficient conditions, all variables were present for the high level of stress with the exception of hedonic balance, which was absent. The frequency cutoff in the truth table was set at 1, and the consistency cutoffs were was set at 0.800523 for perceived stress (Table 4) and 0.886170 for cortisol (Table 5). The intermediate solution for perceived stress (Table 4) indicated five combinations of causal conditions that can produce high levels of perceived stress and four levels in the case of cortisol (Table 5).

The fsQCA model was informative when the consistency was above 0.70 (Sereikhuoch & Woodside, 2012). Considering the results obtained, both solutions appear to be adequate. Regarding the perceived stress solution (Table 4), the five combinations can explain 42% of cases with high perceived stress levels (solution coverage: 0.417087; solution consistency: 0.747754), whereas in the cortisol solution, four combinations can explain 53% of the cases with high cortisol levels (solution coverage: 0.530236; solution consistency: 0.771345).

In the case of the perceived stress solution (Table 4), the two most important sufficiency conditions combinations (based on raw coverage) were a low level of repair, a high level of attention, a low level of hedonic balance, a younger age and the female gender (raw coverage: 0.18; consistency: 0.82) and a high level of repair, low levels of clarity and hedonic balance, and the female gender (raw coverage: 0.15; consistency: 0.76).

However, considering the cortisol solution (Table 5), the two most important sufficiency conditions combinations were a high level of clarity and the female gender (raw coverage: 0.36; consistency: 0.78) and a high level of attention and the female gender (raw coverage: 0.43; consistency: 0.81).

Discussion

The goal of this work was to examine the combined contribution of trait EI and feelings on healthy adolescents’ stress (measured by self-report and by HCC). We expected that high emotional clarity and repair would be negatively associated with perceived and objective stress. These associations could be improved by the inclusion of feelings.

The first hypothesis was only partially supported by data. Analyzing results from both methodologies, fsQCA models and regression models revealed the primary findings. In the regression models, less emotional clarity and less hedonic balance were related to higher levels of perceived stress. In the fsQCA models, these two variables (clarity and hedonic balance) were intertwined and enriched by the inclusion of attention and repair, respectively, in the two primary sufficient conditions. That is, in the first condition, paying strong attention to one’s emotions without believing oneself able to regulate those emotions combined with a low hedonic balance and the female gender was related to high levels of perceived stress. The second condition indicated that believing oneself able to regulate one’s emotional states combined with a poor understanding of one’s emotions, a low hedonic balance, and the female gender was associated with high perceived stress.

As Davis and Nichols (2016) suggested, an imbalance in constituent components of trait EI, as demonstrated in the results above, may characterize vulnerability; that is, an excessive attention to (negative) emotions coupled with a lack of competency to repair those emotional states resulted in greater physiological stress. These authors suggested that “a balance between EI facets is optimal for adaptation” (p. 8). The combination of different emotional dimensions in diverse and imbalanced grades (high/low scores) perhaps present a more realistic picture of the situation. In this sense, the use of fsQCA models in the study, including combinations of conditions, becomes quite valuable.

If we focus on HCC, no emotional variable emerged in the predictive model, and only two emotional variables (clarity and attention) appeared in the two primary sufficient conditions of the fsQCA models. In the first condition, extreme attention to emotions and the female gender were related to high HCC. The second condition demonstrated that a good understanding of emotional states and the female gender were associated with high HCC.

As demonstrated, in the entire sample of adolescents, high clarity was related to low perceived stress. However, in the case of female adolescents, both high and low levels of clarity were related to high levels of stress (independent of the methodology used to measure the stress). Perhaps the clarity dimension in female adolescents indicates a similar pattern to emotional attention in the sense that high or low scores on emotional attention are not productive for emotional management and moderate levels are preferable (Gasper & Clore, 2000).

In fact, the results regarding attention indicate concordant findings with the above result. Emotional attention cannot be deemed significant for the entire sample although it appears again in the female gender. That is, for female adolescents, high levels of attention were related to high levels of stress, which were again independent of the methodology used to measure those levels. These results regarding the attention dimension are consistent with other results observed in adults (Landa, López-Zafra, Martos, & del Carmen Aguilar-Luzón, 2008; Limonero, Tomás-Sábado, Fernández-Castro, & Gómez-Benito, 2004) and may be the result of the emotional spiral that is created when excessive attention is paid to the emotions and ruminant negative thoughts (Fernández-Berrocal, Ramos, & Extremera, 2001).

All of these results related to emotional clarity and attention in female adolescents may demonstrate the negative aspects of high levels of trait EI, that is, “inverted-U-shaped effects, whereby positive phenomena reach inflection points at which their effects turn negative” (Grant & Schwartz, 2011, p. 61). Davis and Nichols (2016) noted the possibility of the existence of “optimal” levels of EI (both within and across EI constructs). In this sense, these authors stated that further research is required in applied settings to verify the adaptive nature of EI because negative effects (some of which were indirect) were observed to operate across multiple contexts (health, academic, occupational, among others). These negative effects are referred to as “the dark side” of the EI construct by the authors (p. 2).

The inclusion of feelings helps to improve the prediction of perceived stress beyond the contribution of trait EI, partially supporting the second hypothesis. In these combinations of emotional variables, hedonic balance was always negatively related to perceived stress (both in the regression and in the fsQCA models) but not to objective stress (HCC). This relation between variables was similar for the entire sample and for female adolescents. This finding reinforces the hypothesis that feelings influence the relation between trait EI and self-reported stress in adolescents, modulating the subjective experience of the situation. However, the role of feelings in the case of the physiological experience of stress does not appear to be relevant.

A special comment concerns the lack of concordance observed in this study between cortisol as a measure of stress and perceived self-reported stress measures. This finding supports the results of previous studies (Gidlow et al., 2016; Van Uum et al., 2008), which suggest various explanations, such as that the PSS scores and cortisol levels are parameters dependent on a large variety of factors (Van Uum et al., 2008) or that methodological issues involve the assessment periods of time for both measures (Russell et al., 2012). Other authors, such as Gidlow et al. (2016, p. 168), suggested that HCC may provide a “better marker of clinical conditions”. However, in the general population and in association with perceived stress, HCC is perhaps not a useful means of validation. Further studies may help to identify the specific contexts, influencing factors and type of participants to present a complete in-depth picture of stress assessment.

Finally, some limitations in the current study should also be considered. First, some pertinent demographic and clinical factors that could have helped to characterize the sample better, such as race, IQ, substance use, hospitalizations, and early life stress, are missing. Moreover, because the sampling was not probabilistic, was obtained in the Valencian community, and was based on normal adolescents in the age range of 12–14 years, the generalization of the results is necessarily limited by these cohort restrictions. In this sense, it would be interesting to extend this study to other populations in Spain and Spanish-speaking countries. Third, although the majority of the instruments used in this study indicate adequate psychometric properties, another limitation concerns the moderate internal consistency of the PSS-4 (0.68). Finally, it is strongly suggested that potential correlates and confounders related to hair cortisol levels, such as the adolescents’ body mass index or recent hair washing must be considered in future studies, as suggested by various authors (Adam, 2006; Dettenborn, Tietze, Kirschbaum, & Stalder, 2012; Rippe et al., 2016).

Despite these limitations, there are important contributions of this study, which uses combined approaches to assess stress in adolescents and different data analysis methodologies. In general, trait EI has been demonstrated to be a protective factor against stress in healthy adolescents in combination with the reinforcement of feelings. However, the debate continues regarding the possibility of optimal levels of trait EI for effective and adaptive emotional management, particularly in the attention and clarity dimensions and for female adolescents. These findings must be considered in addressing the design of health-promotion programs focused on adolescent stress. The need to decrease the incidence of pediatric stress-related diseases, such as bullying, child obesity, or stressful child transitions, by means of stress management programs is critical (Stavrou et al., 2017). An important component of these stress programs, in addition to study results and the transactional stress model, may be the understanding that adolescents’ self-perception of personal resources can be modified to improve the appraisal of the situation. In this manner, youths will learn to be an active part of their own stress management, developing an awareness that stress is not the univocal result of an external event, but the combination of different variables.c

Disclosure statement

None of the authors has any financial, professional or personal conflicts of interest to declare.

Additional information

Funding

Funding from the Spanish Ministry of Science and Competitiveness [PSI2013 43943R] is gratefully acknowledged.