Portuguese validation of the Cognitive Emotion Regulation Questionnaire short version in youth: Validity, reliability and invariance across gender and age

ABSTRACT The Cognitive Emotion Regulation Questionnaire (CERQ) is a multidimensional measure widely used to access nine cognitive emotion regulation strategies. In this study, we examined the psychometric properties of the CERQ-Short Portuguese version. A sample of 1052 adolescents (aged 10–25 years old, 60.53% females) completed the CERQ-Short form, the Positive and Negative Affect Schedule, the Generalized Anxiety Disorder Scale, and the Patient Health Questionnaire. Confirmatory factor analyses showed that the nine latent dimensions of CERQ-Short provided an acceptable fit to the data. Measurement invariance (for gender and three age-groups), reliability and construct validity were adequate. The adaptative strategies were positively associated with higher positive affect, and maladaptive strategies with higher negative affect, symptoms of generalized anxiety and depression. These results suggest that CERQ-Short is a valid and reliable measure for Portuguese-speaking samples. Moreover, CERQ-Short’s length makes it a cost-effective tool for both clinicians and researchers.

young adults. As emotion-eliciting events are better decoded, there is an increase in regulatory strategies (Sanchis-Sanchis et al., 2020). Throughout childhood and adolescence, individuals become better at regulating their emotions independently, contrary to infants and younger children who rely more on their significant adults to regulate emotions.
The Cognitive Emotion Regulation Questionnaire (CERQ) measures selfregulatory cognitive emotion regulation strategies (ERSs) in response to stressful life events (Garnefski et al., 2001). The CERQ evaluates cognitive processes after the experience of negative or stressful situations to comprehend how these processes affect the course of emotional development (Garnefski et al., 2001). The CERQ has 36 items, with four items per each of the nine ERSs dimensions: Acceptance, Positive Refocusing, Refocus on Planning, Positive Reappraisal, Putting into Perspective, Catastrophizing, Rumination, Self-blame, and Other-blame (Garnefski et al., 2001). The CERQ authors also provide an 18-item short version, with two items per dimension, allowing faster screening and its inclusion in larger assessment protocols (Garnefski & Kraaij, 2006).
The CERQ is a commonly employed emotion regulation measure. Originally in Dutch, the CERQ-Short has been validated in other languages (Cakmak & Cevik, 2010;Ireland et al., 2017;Orgilés et al., 2019), with studies reporting adequate psychometric properties for this short version. While the CERQ-36 has been validated in Portuguese (Martins et al., 2016;Moreira et al., 2020), CERQ-Short has not. The goal of this study is to validate the CERQ-Short version in a young Portuguese population.
Research has shown gender and age differences in CERQ dimensions. Although measurement invariance testing is scarce, some evidence shows adequate invariance for both age (Moreira et al., 2020) and gender (Chamizo-Nieto et al., 2020). To the best of our knowledge, measurement invariance was only tested in the CERQ36. Consequently, we will also investigate whether invariance holds across age and gender in the CERQ-Short.
Although the adaptability of ERSs is context-related, extensive research indicates ERSs can be maladaptive given certain mental health disorders (Schäfer et al., 2017). Thus, the first five ERSs are considered adaptative strategies associated with positive affect (Ireland et al., 2017), emotional intelligence and gratitude (Chamizo-Nieto et al., 2020), whereas the latter four are considered maladaptive and associated with high levels of depression and anxiety (Chamizo-Nieto et al., 2020;Garnefski & Kraaij, 2018;Martins et al., 2016).
Based on previous findings, we hypothesize that the nine-factor structure will have the best fit to our sample and subsamples (gender, age). Also, we hypothesize that the CERQ-Short adaptative strategies (Acceptance, Refocus on Planning, Positive Refocusing, Positive Reappraisal, Putting into Perspective) will negatively correlate with generalized anxiety symptoms (GAS), depressive symptoms, and negative affect while positively correlating with positive affect. In contrast, we expect the maladaptive strategies (Rumination, Catastrophizing, Self-blame and Other-blame) to correlate positively with GAS, depressive symptoms, and negative affect, while correlating negatively with positive affect.

Measures
ERSs were measured with the CERQ-Short by selecting the items with the highest factor loadings in each dimension (Garnefski & Kraaij, 2006). The Portuguese version, translated and validated by Martins et al. (2016), was used. The CERQ-Short has 18 items, with two items for each dimension, and uses a 5-point scale (from 1 = never to 5 = always). Previous Portuguese studies with the CERQ36 have reported adequate psychometric properties, with most Cronbach's alpha coefficients above .70, excepting for Acceptance, Refocus on Planning, Self-blame (62 < α < .70) (Martins et al., 2016;Moreira et al., 2020).
Positive and negative effects were assessed using the Portuguese version of the Positive and Negative Affect Schedule Short-Form (PANAS-SF; Galinha et al., 2014). PANAS-SF includes ten items, five for positive and five for negative affect. All items used a 5-point scale (from 1 = Very slightly or not at all to 5 = Very much so). The two-factor model of the measure showed a good fit to our sample (CFI = .97, TLI = .96, SRMR = .04, RMSEA = .05 with 90% CI[.04, .06]). Good internal consistency values were obtained for both the positive (α = .77, 95% CI [.75, .79]) and negative subscales (α = .79, 95% CI[.77, .81]).

Procedure
The ISCTE-University Institute of Lisbon Ethics Committee (ref. 17/2019) approved all procedures. The study was disseminated through social networks and in classes at one university. Researchers collected data from randomly selected classes at five district schools. At the schools, researchers instructed students on how to complete the survey, answered questions and offered to read aloud to minimize differences between reading proficiency levels. Data collection took about 25 min. Contact details were provided. All adult participants provided informed consent, whereas parents/legal guardians provided it for minors. For ethical purposes, the school directors kept the physically signed consent forms in safe deposit boxes. To ensure comprehension of the younger participants, we conducted a pilot study in March 2019, and no changes were needed.

Data analysis
The expanded definition of adolescence may better reflect the development of adolescents in developed countries like Portugal (Sawyer et al., 2018). However, based on this definition, adolescence includes young people at very different stages, who must be disaggregated within age frames (Sawyer et al., 2018). Also, development may be represented better as a series of discrete stages (Leung & Shek, 2020) since it does not always follow a linear trajectory with cumulative changes (Santos et al., 2021). In this study, we disaggregate the 10-14, 15-17 and 18-25 age-groups, according to Arnett's (2013) definitions of early adolescence, late adolescence and emerging adulthood.
The percentage of missing values across the 46 variables varied between 0.1% and 1.24%. In total, 97 out of 1052 cases (9.2%) were incomplete. Graham (2009) describes multiple imputation as the preferred missing data handling technique. We imputed missing data using multivariate imputation by chained equations, using mice (3.2.0) (Van Buuren & Groothuis-Oudshoorn, 2011) and miceadds (3.11-6) (Robitzsch & Grund, 2021) R packages (R Core Team, 2021; version 4.0.5). We used the pmm imputation method (excepting for gender-logreg), set the number of iterations in the mice algorithm to 20, and created 10 different imputed datasets. In the proceeding analysis, each dataset was analysed separately, and results were subsequently pooled using Rubin's rules (Rubin, 1987).
Confirmatory factor analyses (CFAs) and multi-group CFA's were estimated using R package semTools (0.5-4; Jorgensen et al., 2021) with a robust estimator (MLR -maximum likelihood estimation with robust Huber-White standard errors). First, we tested the original Garnefski and Kraaij (2006) nine-factor model. Then, we tested two higher-order factor models: one with the nine dimensions grouped into two latent dimensions of adaptive strategies and maladaptive strategies (Garnefski et al., 2001), and another proposed by Liu et al. (2016) which includes acceptance in the less adaptative strategies. We considered the following indices for establishing model fit criteria: Comparative Fit Index (CFI) ≥ .95; Tucker-Lewis Index (TLI) ≥ .90; Standardized Root Mean Square Residual (SRMR) < .08; and Root Mean Square Error of Approximation (RMSEA) ≤ .08 with a 90% confidence interval (CI) (Hair et al., 2014). Further, measurement invariance by age and gender were evaluated by a series of multiple-groups CFA processes for configural, metric and scalar invariance. As a pre-requisite, the functional equivalence (i.e., the model fit in each group) was confirmed (Milfont & Fischer, 2010). We evaluated the fit of successive models with increasingly stringent constraints, namely ΔRMSEA ≤ .015, ΔCFI ≤ −.010. Also, two ΔSRMR thresholds were used, ΔSRMR ≤ .030 to test factor loading invariance and ΔSRMR ≤ .010 when testing intercepts invariance (Chen, 2007).
The same R package was used to investigate the correlation coefficients since these were based on the latent variables' correlations to account for the measurement error. Cronbach's alpha (α) coefficient with a CI 95%, means, standard deviations, minimum and maximum of the scales are presented.

Confirmatory factor analysis
Three versions of the CERQ-Short were tested. The first-order nine-factor model of the Portuguese CERQ-Short presented a good fit to the data (CFI = .96, TLI = .94, SRMR = .03, RMSEA = .04 with CI 90% [.04, .05]), thus confirming the nine conceptually distinct scales. The standardized factor loadings ranged from .58 to .95 (all p < .001), showing an adequate performance of all items (see Table 1).

Measurement invariance
The hypothesized nine-factor model was tested for each group separately (see Table 2). The fit was good in all subsamples. Therefore, configural, metric and scalar invariance were subsequently tested.

Reliability
Most factors obtained values of acceptable internal consistency varying between .71 (Acceptance) and .76 (Other-blame), though some presented values below .70, with the minimum value being observed for Putting into Perspective and Rumination (α = .61). Inter-correlations were weak to .000 .001 .001 Model1 = nine-factor structure by (Garnefski & Kraaij, 2006); Model 2 = second-order structured by (Garnefski et al., 2001); Model 3 = second-order structured by (Liu et al., 2016); ª Method Robust; CFI = Comparative fit index; TLI = Tucker-Lewis Index; SRMR = Standardized root mean square residual; RMSEA = Root mean square error of approximation; CI = Confidence interval; ΔCFI, ΔSRMR and ΔRMSEA = change in fit indices between contiguous nested models.  strong. The stronger associations are observed between Positive Reappraisal and Refocus on Planning (r = .71, p < .001) and between Rumination and Catastrophizing (r = .74, p < .001), as in the original CERQ-Short development study (Garnefski & Kraaij, 2006). To note that correlations between dimensions considered as adaptative and those considered maladaptive are weak, like in the original version (Garnefski & Kraaij, 2006), confirming they correspond to different constructs. Inter-item correlations scores were between .44 (Putting into Perspective) and .60 (Other-blame), being above the threshold of .30 (Hair et al., 2014).

Discussion
The aim of the present study was to analyse the psychometric properties of the CERQ-Short in a Portuguese sample of 1117 youth participants. The results revealed the best fit for the nine-factor structure, which was consistent with previous studies (Ireland et al., 2017;Moreira et al., 2020), though the second-order model proposed by Garnefski and Kraaij (2006) also showed good fit to the data. The good fit of the Portuguese CERQ-Short found in this study has also been established in other languages (Cakmak & Cevik, 2010;Ireland et al., 2017;Orgilés et al., 2019). The factor loadings were all above the threshold of .40, indicating good fit. The fit obtained in the present study is better than the fit obtained by the two Portuguese studies that analysed CERQ36 (i.e., CFI = .88, SRMR = .06, RMSEA = .05 with CI 90% [.04, .05], (Moreira et al., 2020); CFI = .90, RMSEA = .050; IFI = .90, (Martins et al., 2016)), which is consistent with Ireland et al.'s (2017) findings.
In addition, we assessed the measurement invariance. Functional, configural, metric and scalar invariance were tested since violations of measurement invariance can hinder significant data interpretation. The multigroup analyses showed that the nine-factor structure was adequate for different ages and gender, which strengthens this measure's use for group comparisons. According to the literature, this is the first study that investigated measurement invariance with the CERQ-short version and our results show that the instrument is able to make valid comparison between gender and age. Our results are consistent with those obtained for the CERQ36 version regarding invariance for age (Moreira et al., 2020) and gender (Chamizo-Nieto et al., 2020). This suggests that the Portuguese CERQ-Short version is a psychometrically adequate measure of cognitive ERSs for both gender and young (10-14), middle (15-17) and older adolescents/young adults (18-25 years old).
The CERQ-Short showed adequate internal consistency values for the majority of the nine dimensions. Regarding scale inter-correlation, our findings also provided a pattern similar to other CERQ validation studies, with stronger correlations among adaptative and among maladaptive ERSs, and negative associations between adaptative and maladaptive strategies (Cakmak & Cevik, 2010;Garnefski & Kraaij, 2006;Ireland et al., 2017).
Our third hypothesis was also confirmed. Adaptative ERSs were associated with more positive affect and less symptomatology. In contrast, higher scores in maladaptive ERSs were related to higher negative affect, GAS and depressive symptoms. These relations were expected and provided evidence of construct validity for the CERQ-Short Portuguese version, which is in line with the already reported relation between ERSs and mental health indicators (Schäfer et al., 2017). Similar findings were found in previous studies using CERQ with children, adolescents (Chamizo-Nieto et al., 2020;Garnefski & Kraaij, 2006 and adults (Martins et al., 2016).
Before concluding, some limitations must be mentioned. First, we have not analysed test-retest reliability and other forms of validity (e.g., convergent, discriminant or predictive validity). Second, four of the latent variables had reliability values below .70; Refocus on Planning, Putting into Perspective, Positive Reappraisal, Rumination. A low value may be due to only two items being used on each scale. Cronbach's alpha is positively related to the number of items in the scale, making harder for scales with fewer items to show high values (Hair et al., 2014). Nevertheless, values were all above .60 which is considered the lower limit of acceptability (Hair et al., 2014). Third, we performed no clinical disorder screenings, and based on a meta-analytic review, clinical samples report more maladaptive emotional strategies than non-clinical samples (Aldao et al., 2010). Nevertheless, our sample had mean values in the minimal or mild categories for GAS and depression symptoms, with 69.8% (n = 734) and 68.9% (n = 724) of participants in each scale scoring in the referred categories, suggesting that most of our sample are non-clinical regarding the assessed symptomatology. Fourth, we used a nonrepresentative sampling procedure, although our sample was diverse, including adolescents from rural and urban areas and on different scholastic paths. Finally, not all participants underwent the same procedure (i.e., while the majority responded at school in the presence of a researcher, some 18-year-olds filled out the questionnaire alone). Regardless, participants could ask questions using the provided contact information and participants at schools responded independently. Future studies should investigate whether different data collection procedures (as the ones used) affect the responses.
To conclude, CERQ-Short Portuguese version is a valid and reliable tool to evaluate ERSs. It integrates a wide variety of ERSs in a single questionnaire, can be used in multiple settings and in a broad range of age groups (Ireland et al., 2017). Its brief nature allows for easier integration into assessment protocols. This study validates CERQ-Short use among Portuguese speakers and participation in cross-cultural studies.