The Factorial Structure of the SCARED-R in a Portuguese Community Sample

Aim: To evaluate the three-, four-, fiveand nine-factor structures of the Screen for Child Anxiety Related Emotional Disorders revised version (SCARED-R) in a Portuguese sample. We further aimed at assessing the gender and age patterns of anxiety symptoms. Method: The Portuguese version of the SCARED-R was administered to a community sample of 1,314 children, aged 10-13 years. Confirmatory factorial analysis and multivariate analysis of variance (MANOVA) were employed. Results: The five-, fourand three-factor models presented an acceptable fit to the data. An unacceptable fit to the data was obtained for the one-factor model. The refined nine-factor model presented good fit to the data after the removal of items with low factorial weights. Based on theoretical considerations, this nine-factor model was considered the best model for assessing children’s anxiety symptoms. A hierarchical structure with a second-order factor called "General Anxiety" was proposed. Adequate internal consistency and criterion related validity were demonstrated. Effects of gender and age on the anxiety scores were found. Conclusion: The SCARED-R is a reliable tool for screening anxiety symptoms, and can be initially administered to identify children at high risk for specific DSM-IV defined anxiety disorders.

that can persist into adulthood, or act as a risk factor for the development of other psychiatric disorders (Costello et al., 2003).
There is evidence suggesting that a subclinical manifestation of anxiety disorders is relatively prevalent among children who never had psychiatric disorders (Muris, Merckelbach, Schmidt, & Mayer, 1999).These findings highlight the importance of identifying clinically and non-clinically anxious children at an early stage so that a suitable intervention can be provided (Essau, Muris, & Ederer, 2002;Muris, Merckelbach, Schmidt, et al., 1999).This, however, requires valid and reliable instruments to screen and diagnose children who suffer from anxiety symptoms (Isolan et al., 2011;Su, Wang, Fan, Su, & Gao, 2008).
Structured and semi-structured interviews are used to evaluate those symptoms, but they are time-consuming and require trained interviewers (Muris, Merckelbach, Schmidt, et al., 1999).Inversely, self-report questionnaires are easy to administer, require a minimum amount of time and capture information about anxiety symptoms that can be used to decide if interviews should be subsequently employed for a thorough diagnosis (Silverman & Ollendick, 2005).Thus, numerous self-report questionnaires for the assessment of anxiety symptoms in children and adolescents have been developed and analysed in the literature.
The Screen for Child Anxiety Related Emotional Disorders -SCARED (Birmaher et al., 1997) measures childhood anxiety symptoms that can be specifically linked to the DSM-IV taxonomy, namely symptoms of generalized anxiety disorder, separation anxiety disorder, panic disorder, social phobia and school phobia.Due to problems in the discrimination between patients with social phobia and other anxiety disorders, three items were added to the original 38-item version of SCARED, resulting in a 41-item self-report questionnaire (Birmaher et al., 1999).The literature provides support for the psychometric properties of the SCARED (e.g., Muris, Merckelbach, Mayer, van Brakel, et al., 1998;Muris, Merckelbach, Ollendick, King, & Bogie, 2002;Muris, Merckelbach, Schmidt, et al., 1999).
As the SCARED (Birmaher et al., 1997;Birmaher et al., 1999) and its five factors do not measure all types of anxiety separately as established in the DSM-IV, Muris, Merckelbach, Schmidt, and colleagues (1999) revised the SCARED in three ways.All the items from the original scale were included in the revised version.First, school phobia items from the original scale were assigned to the separation anxiety disorder subscale which means that the number of factors of the former was reduced to four.Then, 15 new items were added to assess specific phobias (i.e., animal phobia, environmental-situation phobia, and blood-injection-injury phobia).These items formed three new factors.Finally, two other factors were composed of 13 new items that were added to assess symptoms of obsessive-compulsive disorder and traumatic stress disorder.As a result, this revised version of the SCARED was renamed SCARED Revised version or SCARED-R.The SCARED-R is composed by 66 items, and has nine subscales/factors that cover the entire anxiety disorders' spectrum that may occur in children, as defined by the DSM-IV (Muris, Merckelbach, Schmidt, et al., 1999).
The SCARED-R is a valuable self-report instrument, as it covers a broad range of the anxiety disorder spectrum (Muris & Steerneman, 2001).It also allows the measurement of specific phobias that are quite prevalent in children (LeBeau et al., 2010).
The main aim of the current study was to evaluate the three-, four-, five-and nine-factor structures of the Portuguese version of SCARED-R.This study further aimed at investigating gender and age patterns of anxiety symptoms.

Method Participants
After obtaining ethical approval by a supervising institution and funding agency (FCT), 1,791 children from the 5 th and 6 th grades (middle school) from all four public schools in Terceira Island -Azores/Portugal, were invited to participate.A number of 1,314 were authorized by their parents to participate in the study, and from these, 76 were posteriorly excluded because they did not complete the questionnaire adequately.The final sample consisted of 1,238 students (600 boys and 638 girls) aged between 10 and 13 years, with a mean age of 11.21 (SD = 0.92).No significant differences were found in the demographics between children who were included in the study and those who were excluded.
Children are asked to indicate how frequently they have each symptom on a 3-point scale: 0 = almost never, 1 = sometimes, 2 = often.The SCARED-R total and subscale scores can be calculated by adding the responses of the 66 items.Higher scores reflect higher levels of anxiety symptoms.

Model
Factor's Name Factor's Items
Children completed a socio-demographic questionnaire and the SCARED-R during class time.Teachers were trained to help the children if they needed and to ensure the independence and confidentiality of their responses.

Data Analysis
Confirmatory factor analysis (CFA) was performed to test the several factor structures proposed.The weighted least squares means and variance adjusted (WLSMV) estimator for ordinal items was used as implemented in Mplus, version 6.12 (Muthén & Muthén, 2007).We used as indices of goodness of fit the χ 2 /df (chi-square by degrees of freedom ratio), CFI (Comparative Fit Index), TLI (Tucker-Lewis index) and RMSEA (Root Mean SCARED-R in a Portuguese Sample values of CFI and TLI were above .9and RMSEA was lower than .05(Kline, 1998;Marôco, 2010).Weighted Root Mean Square Residual (WRMR) was also used to compare the different models' fit.The best fit of the model produced the lowest WRMR (Muthén & Muthén, 2007) and values close or under 1 are suitable (Yu, 2002).To improve the models' fit, items with factorial weights lower than .40were removed (Bollen, 1989;Marôco, 2010).
In order to investigate whether each dimension's manifest variables were strongly related to each other, and whether they were consistent manifestations of their factors, the composite reliability (CR) and the average variance extracted (AVE) were estimated, determining the factors' convergent validity (Fornell & Larcker, 1981;Marôco, 2010).According to Hair, Anderson, Tatham, and Black (1998), values of AVE j ≥ .5 and CR j ≥ .7 indicate adequate convergent validity.Discriminant validity was assessed by comparing the AVE for each factor with the squared correlation (ρ 2 ) between the factors.Thus, if AVE i and AVE j ≥ ρ ij 2 there was evidence of discriminant validity (Fornell & Larcker, 1981;Marôco, 2010).Reliability was analysed using the standardized ordinal α coefficient (Gadermann, Guhn, & Zumbo, 2012).
A single model was chosen for further criterion validity analyses, taking into account theoretical considerations to measure the anxiety symptoms and the models' goodness of fit.After analysing the correlations among the factors, a hierarchical structure with a 2 nd order latent factor called General Anxiety was proposed.A total score of general anxiety can be obtained by summing across relevant items (Muris, Merckelbach, Schmidt, et al., 1999).
The hierarchical structure with a 2 nd order factor as well as the nine 1 st order factors were used to investigate age and gender differences in levels and types of anxiety symptoms through a multivariate analysis of variance test (MANOVA).

Factor Structure
The overall assessment of the structural models of SCARED-R was found to be acceptable for five-, four-and three-factor structures (Table 2).However, the insufficient CFI and TLI obtained in the one and nine-factor structures indicated an unacceptable fit to the data.After the removal of the items that presented factorial weights lower than .40, the refined nine-factor model showed a good fit to the data.However, the one-factor model presented an unacceptable fit to the data.Convergent and discriminant validities were compromised in several factors of the nine-, five-and four-factor models (Table 2).Correlations between the factors were moderate to strong in all models under analysis.The internal consistency was adequate in most models, with alpha coefficients ranging from .70 to .88,excluding the refined five-factor model, where the subscale School phobia presented an alpha coefficient of .47.Finally, the composite reliability was adequate in all models excluding the five-factor model (Table 2).Since the refined nine-factor model presented a good fit to the data, and given the literature pointing the 9-factor SCARED-R as the most complete measure of the anxiety symptoms, further analyses were performed using the refined model.4, 5, 6, 7, 13, 16, 17, 20, 21, 24, 26, 28, 30, 33, 42, 50, 54, 66. b  Poor discriminant validity and strong correlations observed between the factors of this model may be explained by the presence of a 2 nd order factor (Figure 1).In this model, the factorial structure's fit was considered good ( χ 2 /df = 2.7; CFI = 0.91, TLI = 0.91, RMSEA = 0.037; WRMR = 1.65).All the trajectories between the 2 nd order factor and 1 st order factors were statistically significant.The alpha coefficient for the total SCARED-R was .95.The results of MANOVA, based on Pillai's Trace criterion, indicated that gender had a significant effect on the full scale and on the nine subscales' anxiety scores (detailed MANOVA results are presented in Table 3).In all subscales and total scale, girls exhibited higher scores for anxiety symptoms than boys (p < .001).These findings support the criterion-related validity of the SCARED-R for gender, as these results are consistent with evidence produced in previous studies (Isolan et al., 2011;Muris, Merckelbach, Schmidt, et al., 1999).Furthermore, significant age effects were found for separation anxiety disorder (SAD), social phobia (SP), obsessive-compulsive disorder (OCD) and traumatic stress disorder (TSD) subscales (see Table 3).Results suggest that those anxiety symptoms decreased with increasing age.Younger children, with 10 and 11 years old, reported significantly higher separation anxiety (SAD) symptoms than children aged 12 and 13 years (p < .05).Children aged 10 and 11 years also had higher social phobia (SP) and obsessive-compulsive symptoms (OCD) (p < .05).Regarding the symptoms of traumatic stress disorder (TSD), there was a significant difference only between children aged 11 and 13 years, with the youngest presenting higher scores (p < .05).No interaction effect of gender by age was found (see Table 3).

Discussion
The results of our study suggest that the one-factor model is the weakest to describe the structure of the SCARED-R.In a study by Muris, Merckelbach, Schmidt, and colleagues (1999), the results of the exploratory factor analysis suggested that, in samples of normative children, the SCARED-R possesses a one-factor structure.When the authors carried out the confirmatory factor analysis, they found that the various symptoms cannot be interpreted as reflecting a single, homogeneous dimension of anxiety.The same evidence was found concerning data from a clinical sample of children (Birmaher et al., 1999;Birmaher et al., 1997).According to the authors, these results suggest that anxiety scales emerge as separate factors when data of normative or clinically samples of children are analysed.
The results of the present study confirmed that the structures with three, four, five and nine factors have previous studies (Birmaher et al., 1997;Muris, Merckelbach, Schmidt, et al., 1999;Muris et al., 2000;Vigil-Colet et al., 2009).The three-factor model presented the best indices in terms of goodness of fit.However, this model does not cover all the aspects related to the anxiety disorder.Yet, the nine-factor model gives a thorough perspective on this construct, and the fundamental contribution of our work is the confirmation of this structure, which has never been established.Muris, Merckelbach, Schmidt, and colleagues (1999), and Muris and colleagues (2000) failed to find a satisfactory model for the SCARED-R, when all subscales were included in the analyses.
However, some items included in the models with nine-, five-and four-factor structures, namely Items 3,6,13,17,24,26,30,50 and 54, presented some challenges.They had factor loadings lower than .4and raised some questions about their difficulty level.Some of them were too vague, i.e., their meaning could not be literally understood by children and could lead to ambivalent answers (e.g., in Item 17, many children said they liked to play at school but did not like to study; Item 30 does not specify the reason for the child's concern, whether it is related to school aspects, exams, friend issues, or to family aspects, e.g., separation from parents; or other reasons).Moreover, many items seemed to be ineffective in producing anxiety reactions as they seem to be more related to the children's cultural context than to a disorder.For example, Items 6, 13, 24, 26 and 50 seem to reflect behaviours that do not reveal an intrinsic need but are externally motivated, i.e., they are adopted by the majority of children as a response to educational and parental demands regarding their performance and organization of daily activities.The answers to Item 54 may have been influenced by social desirability.For these reasons, they were removed from the subsequent analyses.
In previous studies, Items 3, 13, 17, 30, and 50 also showed problems, because they did not load adequately on each factor (Isolan et al., 2011;Muris, Schmidt, et al., 2002;Vigil-Colet et al., 2009).Moreover, few studies have considered the analysis of the obsessive-compulsive disorder subscale.In our study, Items 6, 24, 26 and 54 did not clearly reflect the contents of the symptoms of this dimension of anxiety.An interesting result, presented by Muris, Merckelbach, Schmidt, and colleagues (1999), was the low correlation found between each of these items with the total scale, and respective subscale.Again we hypothesize that the contents of these items reflect behaviours that result from parental and cultural needs of the sample rather than internal disorders.Further examination of the items composing this specific subscale may be needed.The removal of the mentioned items improved the goodness-of-fit of those three models.
Almost all models tested showed that SCARED and SCARED-R subscales are reliable in terms of internal consistency as occurred in other studies (e.g., Birmaher et al., 1997;Muris, Merckelbach, Schmidt, et al., 1999;Muris et al., 2000).Only the school phobia subscale of the refined five-factor model showed some weaknesses in terms of reliability, with an alpha coefficient of .47.However, this was expected given that this subscale comprises only two items in the refined model.It should be noted that the model with five factors showed relative shortcomings regarding internal consistency and composite reliability.These results are in line with other studies that report a structure of four factors for SCARED-R, with symptoms of school phobia being incorporated in the subscale of separation anxiety (Muris, Merckelbach, Schmidt, et al., 1999)  factors were not independent and that this relationship may be explained by a 2 nd order factor, which led us to suggest a 2 nd order factor in the anxiety evaluation.
The SCARED-R is a useful tool for screening all the symptoms of the anxiety spectrum disorders (Muris, Merckelbach, Schmidt, et al., 1999).After removing the items that did not load adequately on each factor, the refined nine-factor model included less items than the non-refined nine-factor model and presented better fit to the data.The SCARED-R can be considered a reliable instrument in terms of internal consistency, once it showed high α coefficients for our data (Gadermann et al., 2012).The refined model has a comprehensive overview on the several anxiety problems and also on the general anxiety construct, while it is more parsimonious than the non-refined model.It seems more appropriate for screening purposes, so we chose it to proceed with our analysis.
Criterion validity was assessed by comparing girls' and boys' anxiety symptoms with other studies (e.g., Essau et al., 2002).Gender differences were found for all subscales and total scale scores of the SCARED-R.
Previous studies also showed that girls exhibited higher levels of anxiety symptoms than boys (e.g., Essau et al., 2002;Muris, Merckelbach, Schmidt, et al., 1999;Muris et al., 2000).These results support the validity of the SCARED-R.
Our findings also revealed age differences as younger children reported more symptoms of separation anxiety disorder, social phobia, obsessive-compulsive disorder and traumatic stress disorder than the older ones.
Limitations of the present study should be considered when interpreting our results.The participants of our study were aged between 10 and 13 years and were recruited from a community sample, from a specific Portuguese population.Therefore, it is not possible to generalize our findings to other populations, such as clinical populations or other age groups.Also, validity analysis of the SCARED-R was affected since no other measures were applied.Further research is needed to analyse the validity of SCARED-R, for example, using a clinical diagnostic interview or other relevant instrument.
Findings of our study allowed the confirmation of a nine-factor structure of SCARED-R in Portuguese children.
Since the SCARED-R is a reliable tool, easily administered, and does not involve financial costs, it can be employed at an early stage to identify children at risk for specific DSM-defined anxiety disorders.This screening will allow further evaluation by completing the DSM semi-structured interview to determinate the existence or not of an anxiety disorder and subsequent treatment (Essau et al., 2002;Muris, Merckelbach, Schmidt, et al., 1999).
acceptable fit to the data.All the structures, except the nine-factor model, had been confirmed in several SCARED-R in a Portuguese Sample 10 Psychology, Community & Health 2017, Vol.6(1), 3-15 doi:10.5964/pch.v6i1.183 or distributed across the remaining factors (Vigil-Colet et al., 2009).There are moderate and high correlations between the factors that can be explained by the existence of an overall anxiety factor.The low discriminant validity found for some factors might support this suggestion.Vigil-Colet and colleagues (2009) also found moderate to high correlations between the factors, indicating that the Psychology, Community & Health 2017, Vol.6(1), 3-15 doi:10.5964/pch.v6i1.183

Table 3
Mean and Standard Deviation by Gender and Age for Screen for Child Anxiety Related Emotional Disorders -Revised (SCARED-R).