Cognitive Emotion Regulation

From a theoretical point of view, researchers emphasize the affinity between the models of cognitive emotional regulation (CER) and cognitive coping [6]. Coping refers to the efforts of an individual to master demands (conditions of harm, threat, or challenge) that are appraised (or perceived) as exceeding or taxing his or her resources [6]. The traditional coping model [7] contains both cognitive (e.g., figuring out what to do) and behavioral (e.g., sharing one’s feelings with someone else) coping strategies. Measurement instruments designed according to this model provide data that reflect the combined effectiveness of these two strategies but not that of each separate strategy. This limitation is addressed in the CER model, in which cognitive factors are conceptualized and measured independently of behavioral factors [6]. Cognitive emotion regulation refers to the cognitive means of handling the intake of emotionally arousing stimulations [6,8]. Given the operative nature of CER, its model can be easily analyzed in a research setting [6,9]. In response to the absence of an adequate assessment directly examining CER strategies, Garnefski and colleagues developed the Cognitive Emotion Regulation Questionnaire (CERQ) [6]. From an application perspective, the outcomes obtained using CREQ by this group indicate that cognitive emotional regulation processes moderately affect the relationship between risks in people experiencing negative life events (e.g., myocardial infarction, chronic disease, or cancer) [10–12] and certain mental disorder symptoms (e.g., depressive symptoms) [13].

Cognitive Emotion Regulation Questionnaire

CERQ was developed to assess Dutch secondary school students' CER strategies in relation to stressful life events [6]. To date, this measurement remains the only available scale that adopts a conceptually pure quantification of CER strategies. The questionnaire was previously developed using adolescent and adult samples; however, it had been validated recently using a sample of children aged 9 to 11 [8]. Identical to CERQ, the version of the CERQ for children (CERQ-k) evaluates nine conceptually separate CER strategies, Self-blame pertains to projecting blame onto oneself. Acceptance signifies accepting what one has experienced. Rumination implies reflecting on feelings and thoughts associated with a negative event. Positive refocusing involves embracing positive thoughts rather than fixating on a negative and stressful event. Refocus on planning indicates thinking about the best next step to take after experiencing a negative event. Positive reappraisal means considering the possible positive results or interpretations of a negative event. Putting into perspective entails comparing a negative event to other possible outcomes. Catastrophizing refers to making a negative event out to be worse than it actually was. Finally, Other-blame suggests blaming others for what one has experienced. In a broad sense, empirical studies using children and adult samples suggest that these nine dimensions are sufficiently intercorrelated that they can be grouped into adaptive and less adaptive CER strategies [6,9,14,15]. Adaptive strategies include Positive reappraisal, Refocus on planning, Positive refocusing, Acceptance, and Putting into perspective, whereas less adaptive strategies include Self-blame, Catastrophizing, Rumination, and Other-blame.

Development of CERQ

The psychometric property examination of CERQ-k in Dutch children indicated that the values of Cronbach’s α values ranged from 0.67 to 0.79 for adaptive CER strategies and 0.67 to 0.79 (except for Acceptance, 0.62) for less adaptive CER strategies [8]. However, this previous study did not perform confirmatory factor analysis (CFA) to verify the hypotheses in the model and test–retest reliability to assess temporal stability. Other versions of CERQ have been developed in several researches, such as the French version (CERQ-F) using a sample of French young adults [14], the Chinese version (CERQ-C) using a sample of Chinese university students [16], the Spanish version (CERQ-S) using a Spanish sample [9], the Persian version (CERQ-P) using a sample of university students in Tabriz [17], and the Turkish version (CERQ-T) using a Turkish sample [15]. These versions of CERQ confirmed sufficient Cronbach’s α values, such as those ranging from 0.72 (Self-blame) and 0.83 (Catastrophizing) in CERQ-T to 0.76 (Self-blame and Putting into perspective) and 0.90 (Refocus on planning) in CERQ-C. Temporal stabilities suggested that these versions of CERQ (excluding CERQ-F and CERQ-P) were stable. For example, the test–retest reliability for the nine subscales ranged from 0.50 (Other-blame) to 0.70 (Self-blame) in CERQ-T. Confirmatory factor analysis (CFA) revealed positive support for this underlying first-order nine scale structure in these versions of CERQ (excluding CERQ-P). Meanwhile, a second-order factorial model was confirmed in CERQ-F and CERQ-S. As mentioned earlier, these two higher order factors encompass CER styles that have been categorized as adaptive and less adaptive CER strategies. Validity was also established in several studies, including a comparison of CERQ with other measures related to affective functioning, such as the Children’s Depression Inventory (CDI) [8] and the Positive and Negative Affect Scale (PANAS) [9]. Positive correlations were observed between less adaptive CER strategies and the self-reporting of depressive symptoms, anxiety, and negative affect; adaptive CER strategies were negatively correlated with depression symptoms and anxiety but positively correlated with positive affect [9]. Overall, these results indicated that adaptive strategies were predictive of improved psychological functioning, whereas less adaptive strategies were predictive of depressive symptoms. Criterion-related validity of CERQ-k was established with a comparison of the CERQ-k to the Children’s Depression Inventory, Fear Survey Schedule for Children, and the Non-productive Thoughts Questionnaire for Children [8]. With regard to CERQ-k, Self-blame and Catastrophizing predict depressive symptoms, worry, and fearfulness, whereas Positive refocusing and Refocus on planning predict the absence of depressive symptoms [8]. Due to a good reliability and validity of the Chinese version Children’s Depression Inventory (CDI-C) relative to other scales [18,19], we decided to use CDI-C to test the validity of the Chinese Child Version of the Cognitive Emotion Regulation Questionnaire (CERQ-Ck). One of the objectives of the present study is to revise the CERQ-Ck. According to the previous studies, the hypotheses were as follows.

1. As in the original form, CERQ-Ck will demonstrate a nine-factor structure.
2. CERQ-Ck will exhibit good internal consistency and test–retest reliability values.
3. To confirm construct validity, adaptive CER strategies will correlate negatively with depression symptoms, whereas less adaptive CER strategies will correlate positively with depression symptoms.

Studies Using CERQ

Two studies have established that males and females adopt different coping strategies. Women are more likely to report the use of Rumination, Catastrophizing, and Positive refocusing than men [20]. In addition, females score higher on Rumination, Catastrophizing, and Putting into perspective, whereas males score higher on Other-blame [9]. These findings illustrate that women generally tend to focus more on their emotional experience, acknowledging and discussing emotions more openly and ruminating more on sadness than men [20]. To the best of our knowledge, no study has yet tested whether the underlying measurement structure of all versions of CERQ is gender invariant. Psychological constructs, such as CER, constitute latent variables that cannot be measured directly. Consequently, psychological measures function as indicators of the latent construct. To compare a latent construct meaningfully across different groups, each observed indicator must relate to the latent variable in the same manner across all groups. The gender invariance of CERQ-k is important because differences in CERQ-k scores should only reflect differences in CER strategies. Hence, CERQ-k differences should not relate to differences in response tendencies between gender groups. For example, if one group avoids the extreme answer categories of the rating scale whereas the other group prefers these extreme response categories, then comparing the CER strategies of these groups can lead to rash and inappropriate conclusions. Consequently, one of the objectives of the present study is to examine the measurement invariance of CERQ-Ck across genders. The hypotheses was that structural invariance will exist across gender groups.

Two cross-cultural studies that tested CER strategies have demonstrated that cultural variations play an important role in examining the relationship between CER strategies and psychopathological symptoms [9]. Such cultural influences can explain why cognitive strategies that attribute the causes of a stressful event to others (Other-blame) in collectivist cultures (e.g., Chinese) are associated with high levels of anxiety [16], although they are not evident in individualist cultures (e.g., Spanish and Dutch). Similarly, cognitive strategies (e.g., Positive reappraisal) that contribute to a reappraisal of a stressful event in terms that favor personal growth are only associated with low levels of anxiety and depression in individualist cultural models [9]. In collectivist models, however, highly impersonal emotional regulation strategies (e.g., Positive refocusing) are preferably used [9,16]. Therefore, the present research aims to further investigate the relationship between strategies and depression symptoms in Chinese children.

Research Objective

From a developmental point of view, the relationship between the cognitive strategies of children and psychological outcomes is an important research area. As children grow older, their emotion regulation ability increases and shifts from mainly external, behavior-oriented emotion regulation strategies to internal, cognitive-based strategies [8,21]. When young children reach the age of 8 or 9, they have learned to regulate their emotions by means of cognition, thoughts about themselves, and their feelings toward others [22]. Thus, theoretically valid CER measures for children are necessary. Nevertheless, psychological assessment tools for measuring the CER strategies of children remain limited in China. To fill in this gap, the present study revised a Chinese version of CERQ-k and examined its factor structure and psychometric properties using a sample of Chinese children.

Participants

The study protocol was approved by the institutional review board of the College of Psychology of Liaoning Normal University (China). The parents of the students had to sign a parental consent form for their children to participate in the study.

A total sample of 1403 elementary students, which consisted of 667 females (47.5%) and 736 males (52.5%), was tested via cluster sampling. All participants were aged between 9 and 11 (mean age = 10.14, SD = 0.87) and were from four primary schools in Dalian (Liaoning Province, China). To confirm convergent validity, a subsample of 1083 elementary students completed CERQ-Ck and the Chinese version of CDI (CDI-C) [23]. To assess the test–retest reliability of CERQ-Ck, a subsample of 76 participants with different ages and genders (mean age = 9.87, SD = 0.87, 35 females and 41 males) from the total sample was asked to answer CERQ-Ck 1 month later. A total of 21 psychology graduate students assisted the teachers in charge of the class to ensure the smooth progress of the study. The students were provided with a comprehensive description of the research in their classrooms. All the respondents who volunteered to participate in the study were rewarded with a gift (e.g., pencil).

Procedure

The adaptation procedures used in this study were selected based on the guidelines developed by the International Test Commission [24,25]. Considering the advantages and disadvantages of the forward-adaptation and the backward-adaptation methods [26], all 36 CERQ-k items of the original instrument were translated into Chinese by two functionally bilingual translators into Chinese. The result of this work was assembled in a draft that was administered to 15 Chinese speakers of Chinese to obtain feedback. A psychology professor (whose first language was Chinese) and a postgraduate student (whose first language was English) who were proficient in both languages blindly translated the items back into English. Afterward, the final Chinese and the English back-translated versions were compared by three graduate students. Based on the judgments of a psychology professor, three primary school teachers, 12 psychology graduate students and 30 elementary students, CERQ-Ck was developed by modifying 8 items to improve the comprehension of children. For example, the item “I think that worse things can happen” (Putting into perspective) was replaced with “I think that certain things that happen to me are better than others” because the former is difficult for Chinese children to understand.

First, the project leader, a psychology professor, trained the teachers and psychology students (research assistants) regarding their task assignments, including stating the implications of the present study, reading the scale instructions using Mandarin, issuing and recovering questionnaires, and answering the questions of students. Second, all 1403 elementary students completed CERQ-Ck, whereas a subsample of 1083 elementary students also answered CDI-C administration of the teachers for approximately 1 hour. Finally, self-assessments were used to follow up with 76 elementary students after 1 month.

Measures

Statistical Analyses

All analyses were conducted using SPSS version 20.0, except for CFA and multigroup CFA, which were conducted using Mplus version 7.0. The missing values (0.275% in CERQ-Ck, 0.578% in CDI-C) were inserted through the series mean method. An itemmetric analysis for the CERQ-Ck was executed for the total sample. The corrected item-total correlations were calculated to examine item validity. For internal consistency, the Cronbach’s α values of the nine scales and test–retest reliability were computed for the total sample. The correlations between the scales of CERQ-Ck and the total scores of CDI-C were calculated.

A CFA with robust maximum likelihood estimation (MLR) was conducted to test the structure of CERQ-Ck [28]. The model fit was assessed against several frequently cused goodness-of-fit indices: the chi-squared goodness-of-fit statistic [29], the Tucker–Lewis Index (TLI) [30], the comparative fit index (CFI) [30], the root mean square error of approximation (RMSEA) [31], and the standardized root mean square residual (SRMR) [32]. A value of 0.90 or higher for TLI and CFI implies an acceptable fit: RMSEA values less than 0.08 indicate a moderate fit, whereas values higher than 0.08 signify a poor fit [33]. SRMR values below 0.08 are regarded as a good fit [32].

Finally, multigroup CFA was performed to test structural invariance across gender groups. According to the multivariate non-normal distributed data in our study, the present work used MLR as estimator. A series of goodness-of-fit indices was used for model testing and comparison because measurement invariance incurred nested models. The significant results from the Satorra–Bentler scaled χ² difference indicated that the model with smaller χ² exhibited a statistically better fit. To identify a significant decrease of fit between a series of progressive restricted models, ΔCFI (< 0.01) was suggested as the criterion [34]. In addition, we calculated the ΔRMSEA (< 0.015) of highly restrictive invariant models for samples larger than 300, as suggested by Chen [34].

Itemmetric Analysis and Reliability

Means, SD, skewness, kurtosis, and Cronbach’s α were calculated (Table 1). According to Finney and Distefano [35], a univariate kurtosis of less than 7 and a skewness of less than 2 demonstrate sufficient normality. The kurtosis and skewness values of the nine scales ranged from –0.47 to 0.58 and –0.57 to 0.90, respectively, which indicated that the data were of normal distribution. In the total sample, the Cronbach’s α values of the nine scales ranged from 0.66 (Acceptance) to 0.73 (Refocus on planning). All scales were acceptably reliable. The test–retest correlations were statistically significant at 0.53 (Acceptance, p < 0.01) to 0.70 (Catastrophizing, p < 0.01), which indicated acceptable temporal stability. The corrected item-total correlations ranged from 0.39 (Item 6, Table 2) to 0.59 (Item 16, Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Mean, SD, Skewness, Kurtosis, Cronbach’s α, and the test–retest reliability for scales of the CERQ-Ck.

https://doi.org/10.1371/journal.pone.0150206.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Corrected Item-Total Correlation and Factor Loadings for items of the CERQ-Ck.

https://doi.org/10.1371/journal.pone.0150206.t002

Relationship between CERQ-Ck Scales

Table 3 provides the correlation matrix of the nine CERQ-Ck scales. The correlations among the nine scales ranged from –0.23 (Positive reappraisal, Catastrophizing, and Other-blame) to 0.53 (Refocus on planning and Positive reappraisal) in the total sample, with a mean correlation coefficient of 0.22.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Correlations among Scales of the CERQ-Ck and the CDI-C.

https://doi.org/10.1371/journal.pone.0150206.t003

Convergent Validity

Correlations were calculated between the nine scales of CERQ-Ck and the total score of CDI-C. The results are provided in Table 3. As expected, the total scores of CDI-C were correlated significantly with the nine scales of CERQ-Ck. Positive refocusing, Refocus on planning, Positive reappraisal, and Putting into perspective were negatively correlated with depression symptoms. By contrast, the other five scales were positively correlated with depression symptoms. In particular, Acceptance was positively correlated with depression symptoms. Consistent with other children samples [8], the findings indicated a significant correlation between the five scales (Self-blame, Positive refocusing, Refocus on planning, Catastrophizing, and Other-blame) of the original CERQ-Ck and the self-reported depressive symptoms, which supported a high degree of convergent validity for CERQ-Ck.

Factor Structure of CERQ-Ck

According to Mardia and Foster [36], if the value of Mardia’s normalized multivariate kurtosis is greater than 3, then it indicates a nonnormal distribution. In our study, a value of 170.8 for Mardia’s normalized multivariate kurtosis indicated a nonnormal distribution. As recommended by Sass [28], MLR with continuous obvious indicators is more appropriate when data are multivariate nonnormal.

In the total sample, the nine first-order factors of CERQ-Ck (Model A) were examined through CFA (Table 4). The indices of Model A demonstrated that the model provided a good fit to the data, with both TLI and CFI above 0.90 and both RMSEA and SRMR considerably below 0.08. The findings supported the underlying nine-factor structure of the original structure proposed by Garnefski et al. [8]. The standardized factor loadings were all significant and ranged from 0.45 (Item 4 to Positive refocusing) to 0.72 (Item 15 to Positive reappraisal and Item 18 to Other-blame), with a mean loading of 0.61, which suggested that the items generally converged meaningfully to the scales.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Primary Goodness-of-Fit and Comparative Indices for the Model of the CERQ-Ck and Tests of structural invariance for multigroup model across gender groups.

https://doi.org/10.1371/journal.pone.0150206.t004

To test whether the distinction between adaptive and less adaptive strategies would fit the data, a second-order factorial model made of two higher order factors (Model B) was proposed. As in the previous CFA, each strategy was defined by its respective items; however, two second-order factors were added. The first factor was defined by the four strategies that were supposed to be adaptive: Positive reappraisal, Refocus on planning, Positive refocusing, and Putting into perspective. The second factor was defined by the five strategies that were supposed to be less adaptive in the present study: Acceptance, Self-blame, Rumination, Catastrophizing, and Other-blame. The fit indices of Model B are presented in Table 4. RMSEA and SRMR suggested that the latent and measurement models, respectively, were acceptable. However, TLI and CFA were not within the acceptable cutoff (0.90). A third-order factorial model (Model C) was also examined. The nine first-order factors that formed the two second-order factors (adaptive and maladaptive CER strategies) formed a third-order factors (CER strategies). The fit indices of Model C, which are similar to those of Model B, are presented in Table 4. Considering the fit indices of three models and comparative fit indices (ΔCFI = -0.055) between Model B with Model A, the overall results suggested that the nine-factor model was better than the higher-order factor models.

Measurement Invariance

We used multigroup CFA with increasingly restricted specifications conducted on the nine-factor model to examine measurement invariance across gender groups. Four common models were classified under this category: configural, metric, scalar, and residual invariance. Configural invariance was tested only for the same specifications for both females and males. This model is the first step to establish measurement invariance because it serves as the comparison standard for subsequent tests. Metric invariance requires the corresponding factor loadings for gender groups to be equivalent. If metric invariance is satisfied, then the obtained ratings can be compared across groups and observed item differences can indicate group differences in the underlying latent construct. Scalar invariance assumes that the intercepts of items are equal across the CERQ-Ck of gender groups. It is the last model necessary to compare scores across groups. All additional tests are optional and may be theoretically meaningful in specific contexts. Residual invariance, which is the most restricted model, demandes equal error variances. Each model is nested in the previous model, and measurement tests become increasingly restrictive. Multigroup CFA that follows this approach is widely accepted as the most powerful and versatile means to test measurement invariance. Consequently, a model is only tested if the previous model has been determined to be equivalent across groups. The results are summarized in Table 4. These models exhibited good fit indices, but the Satorra–Bentler scaled χ² test between the scalar invariance model and the residual invariance model was significant, which indicated that the imposition of constraints (equal error variances across groups) resulted in statistically significant reductions in the fit of the residual invariance model compared with that of the scalar invariance model. However, considering the other comparative fit indices (e.g., ΔCFI < 0.01, ΔRMSEA < 0.015), the overall results suggested that the differences in error variances for both males and females were within the range of statistical tolerance to support equivalence across gender groups. The previously discussed models addressed full measurement invariance because they assessed each given element (i.e., factor loadings, item intercept, residual variance) to be equal in all groups, which indicated that each observed indicator of CERQ-Ck related to the latent variable was the same across gender groups.