A French Translation of the Childhood Anxiety Sensitivity Index (CASI): Factor Structure, Reliability and Validity of This Scale in a Nonclinical Sample of Children

Reiss and McNally (1985) define the anxiety sensitivity (AS) as the fear of anxietyrelated bodily sensations due to beliefs that these sensations will lead to catastrophic outcomes such as physical illness, social embarrassment, loss of control and mental incapacitation. Since the 1980s, this concept has attracted the attention of many researchers because of the central role that AS plays in the etiology and maintenance of anxiety pathologies (e.g., Anderson & Hope, 2009; Leen-Feldner, Feldner, Reardon, Babson, & Dixon, 2008), depression (e.g., Weems, Hammond-Laurence, Silverman, & Ferguson, 1997), and chronic pain (e.g., Tsao, Meldrum, Kim, & Zeltzer, 2007) in child populations. Given the high prevalence of RESEARCH ARTICLE

anxiety disorders among youth -some 10% (Silverman & Treffers, 2001) -it is important to have valid instruments to measure AS, whether for clinical or preventive reasons.
However, this concept has mainly been studied in adults and relatively little in children. After evidence of the importance of AS in adults had accumulated, some researchers investigating anxiety disorders in childhood turned their attention to the study of AS in children (Silverman, Fleisig, Rabian, & Peterson, 1991). Several authors have questioned whether AS can even exist in young children because of their level of cognitive development (e.g., Chorpita, Albano, & Barlow, 1996), while others have argued that the association between bodily sensations and harmful consequences might be rare in children. This does not mean that children have cognitive difficulties interpreting physical symptoms as catastrophic (Weems, Hammond-Laurence, Silverman, & Ginsburg, 1998). Muris, Mayer, Freher, Duncan, and van den Hout (2010) demonstrated in their study that, between the ages of 4 and 13, a substantial proportion of children were able to perceive their physical symptoms as signals of anxiety. Moreover, even if children are not able to make internal attributions, they can learn by observation or by conditioning that anxious symptoms can lead to unpleasant consequences (Weems et al., 1998). Silverman et al. (1991) developed the Childhood Anxiety Sensitivity Index (CASI), a self-report scale to assess AS in children and adolescents aged 6 to 17 years. The CASI consists of 18 items, 16 of which are identical to items in the adult Anxiety Sensitivity Index (ASI; Reiss, Peterson, Gursky, & McNally, 1986). Some items were modified to enhance children's comprehension. Silverman et al. (1991) obtained good internal consistency for their inventory (α = 0.87) and good test-retest reliability in both nonclinical (r = .76) and clinical samples (r = .79). Other studies have reported similar internal consistencies (Chorpita & Daleiden, 2000;Weems et al., 1997). The CASI has been translated into several languages (see Table 1) and these versions show good reliability (α ranging from .77 to .89) and validity indices. An additional measure of child AS has since been developed and used to assess this construct in children: a revised version of the Childhood Anxiety Sensitivity Index (CASI-R; Muris, 2002). The CASI-R consists of 31 items -8 items from the original CASI and 23 items from the Revised Anxiety Sensitivity Index (ASI-R; Taylor & Cox, 1998; revised AS index for adults) -and was developed to improve the scale's factor structure. Two studies have investigated the psychometric qualities of this revised version: Muris (2002) for the Dutch version and Stassart, Hansez, Delvaux, Depauw, and Etienne (2013) for the French version. These authors showed that the scale had satisfactory reliability and validity. Muris (2002) suggested that the CASI-R could not be used with children younger than [12 years] because they might find it difficult to understand some of the items, such as those that mention heart attack and stroke. For those children, he suggested the use of the CASI, which is shorter. In fact, in the literature, the CASI is the most frequently used scale to measure AS in children.
The factor structure of AS scales has been subject to dispute in the past. Some studies reported a one-dimensional structure that could be perceived as a fundamental fear (McNally, 1999;Reiss, 1991;Taylor, Koch, & Crockett, 1991). Other studies proposed a multidimensional construction with four factors (e.g., Telch, Shermis, & Lucas, 1989;Wardle, Ahmad, & Hayward, 1990): (1) fear of cognitive symptoms (dizziness and derealization), (2) fear of cardiopulmonary sensations (palpitations), (3) fear of gastrointestinal sensations (nausea), and (4) fear of loss of emotional control and behavioral symptoms (shaking). Another model has also been imposed: a hierarchical model with several lower-order factors loading on one higherorder factor (anxiety sensitivity). This model reconciles the two earlier ones: those authors who identified a one-dimensional structure of the AS focused more on the upper level of  3,4,6,7,8,9,10,11,14,15,16,18 Non-Autonomic Scale  3,4,6,7,8,9,10,11,13,14,16,18 Psychological Concerns  3,4,6,7,8,9,10,11,13,14,16,18 Psychological Concerns the hierarchical model while the ones who claimed a multidimensional structure paid more attention to the lower levels. Currently, the hierarchical construct with several lower-order factors loading on one higherorder factor (anxiety sensitivity) is generally accepted by most researchers and appears stable across both clinical and nonclinical populations (Zinbarg, Mohlman, & Hong, 1999). However, although different authors have agreed that AS has a hierarchical structure in children, they do not agree on the number and nature of factors. In addition, the number of items composing the CASI differs for different authors: some consider that the 13-item model is the best-fitting one, while others prefer the 18-item model or the 12-item model. Table 1 summarizes the different studies that have conducted a factor analysis of the CASI. These differences in the factor structure of the CASI can be explained by different causes such as the version used (18-, 13-, or 12-item versions), the origin of the sample (different nationalities), different methods of analysis, or a problem with the translation of the scale. Despite the differences, some similarities appear. All the authors distinguish between the factor representing the fear of physiological symptoms and factors representing mental, social and control concerns. Moreover, some items are always associated with the same factor: items 6, 9, 10, 11, and 14 with the physical concerns factor; items 2, 12, and 15 with the mental concerns factor; and items 1, 5, and 17 with the social concerns factor.
This study aims to define a factor structure for AS as assessed by the French version of the CASI. In order to contribute additional information to the debate about the number and nature of first-order factors, a first exploratory factor analysis was studied in a Belgian sample. We hypothesized that a three-or four-factor solution would be most valid. Then we tested two other models: (1) the pre-existing factor solution of  with the 18-item CASI) because this initial solution is still widely used for the original version of the CASI and was confirmed by Muris et al. (2001); and (2) the factor solution of Silverman et al. (2003, with the 13-item CASI) because these authors found that the 13-item version offers a better factor structure for AS from an empirical and theoretical point of view. Moreover, these two solutions were also chosen above the others presented in Table 1 because the literature focuses more on hierarchical models with three or four first-order factors than two first-order factors.

Method Participants
Data were collected from two samples recruited from several regular primary and secondary schools in the area of Liège (Belgium) by the same independent experimenters in different conditions. A first sample of 153 children, aged 9 to 13 years (64 boys, M = 10.6 years, SD = 1.20; 89 girls, M = 10.4 years, SD = 1.10), responded to the CASI in a study of the exploratory factor analysis of this scale. The second sample, comprising 200 children also aged 9 to 13 years (95 boys, M = 10.6 years, SD = 1.17; 105 girls, M = 10.7 years, SD = 1.27), was used for the confirmatory factor analysis and the evaluation of the psychometric qualities of the French version of the CASI. This second sample responded to all the questionnaires described in this study. The participants and their parents gave informed consent prior to the start of the study. To target a large population, participants were recruited in schools, but the children completed the questionnaires at home in order not to encroach on school time. The questionnaires were administered in a standardized way to children at home without their parents being present. Administration time was 5 minutes for the first sample and 40 minutes for the second sample.

Measures
Childhood Anxiety Sensitivity Index (CASI;  The CASI is an 18-item questionnaire using a 3-point response scale designed to assess AS in children. Scores on the CASI can range from 18 to 54, with higher scores reflecting more AS. The internal consistency of the total CASI score was good (Cronbach's α = .87): the Cronbach's alphas for the separate dimensions of AS ranged from .33 to .82. The analysis of the scale's validity appeared satisfactory . The French translation used in this study was produced by Vanasse, Houde-Charron, and Langlois (2010). These authors found satisfactory validity and good internal consistency, with a Cronbach's alpha of .87 (Vanasse et al., 2010).  Turgeon & Chartrand, 2003) The STAIC consists of two 20-item scales using a 3-point response scale: a State scale, which measures transitory anxiety reactions to particular situations, and a Trait scale, which measures a stable predisposition to react anxiously, regardless of the situation. The detailed instructions, normative data, and various reliability and validity parameters (internal consistency and concurrent validity) for this scale are summarized by Spielberger et al. (1973) for the English version and by Turgeon and Chartrand (2003) for the French version. In this study, Cronbach's alpha was .77 for the trait scale and .79 for the state scale, indicating acceptable internal consistency.

Exploratory Factor Analysis (with Extraction of Principal Components)
The exploratory factor analysis was done with STATISTICA 9 (StatSoft, 2010). The number of factors retained was decided using the following criteria: (1) Kaiser's (1961) criterion of retaining factors with unrotated eigenvalues of approximately 1 or greater, (2) the scree test (Cattell, 1966), and (3) the interpretability of the resulting factor structure (Gorsuch, 1983). Six factors with eigenvalues greater than 1 emerged, but the scree test and the interpretability of the factor structure suggested that a three-or four-factor solution was more valid; see Table 2 (Tabachnick & Fidell, 2001).
The three-factor model accounted for 39.10% of the total variance. The first factor corresponded to physical concerns, the second factor to mental incapacity concerns, and the third factor to social concerns. The significant interfactor correlation between Factors 1 and 2 was .63, for Factors 2 and 3 it was .34, and for Factors 1 and 3 it was .36.
The four-factor model accounted for 45.50% of the total variance. The first factor corresponded to the physical concerns scale, the second factor to mental incapacity concerns, the third factor to social concerns, and the fourth factor to loss of control. The interfactor correlations were all significant: between Factors 1 and 2 (r = .53), 1 and 3 (r = .33), and 1 and 4 (r = .50); between Factors 2 and 3 (r = .18), and 2 and 4 (r = .32); and between Factors 3 and 4 (r = .26).

Confirmatory Factor Analysis
The confirmatory factor analysis (CFA) was done with LISREL 8.80 (Jöreskog & Sörbom, 2006). The estimation of Robust Maximum Likelihood (RML) was applied to the covariance matrices. Goodness-of-fit indices of six models are presented in Table 3. The four-factor solution resulting from our exploratory factor analysis (EFA) was the best solution and provided the best fit to the data (figure 1). This model had the lowest Root Mean Square Error of Approximation (RMSEA), χ 2 /df values and the highest Normed Fit Index (NFI), Non-Normed Fit Index (NNFI) and Comparative Fit Index (CFI) values. The χ 2 difference test could not be used for the comparisons between models resulting from EFA and other models because these models were not nested, and thus the Expected Cross Validation Index (ECVI) and Akaike Information Criterion (AIC) values were used. Given that the ECVI and AIC values depend on the number of parameters to estimate, 1 these values were compared separately to the model with the 18-item CASI and the model with the 13-item CASI.  When looking at fit indices for the 18-item CASI, the four-factor models resulting from EFA had the lowest AIC and ECVI values. The completely standardized factor loadings of the CASI items were all significant ( Table 4) and greater than .30 (salient loading; Gorsuch, 1983), except item 13.

Internal Consistency, Means and Standard Deviations
The CASI's reliability was assessed using Cronbach's alpha (α) (see Table 5). A series of t-tests revealed significant gender differences on the CASI as a whole (t (198) = 4.02, p < .001) and for the factors other than social concerns: the physical concerns factor (t (198) = 4.26, p < .001), the mental incapacity concerns factor (t (198) = 2.68, p = .008), and the losing control factor (t (198) = 2.75, p = .007). As presented in Table 5, girls had higher means; no effect of age or educational level was observed.

Item Analysis
In Table 6, the items of the CASI are listed and the results of the item analysis are presented. The item discrimination or the item-total correlation is the capacity of the items to differentiate individuals who have high trait levels from individuals who have low trait levels. (Embretson & Reise, 2000). Items with an item-total correlation less than 0.30 were considered questionable (Nunnally & Bernstein, 1994). The results indicated that item 13 fell considerably below this critical value, and items 2 and 7 just below the critical value. The item difficulty index indicated the likelihood of correct response (DeMars, 2010) and items ranging beyond 1.96 were considered difficult. 2 We observed that item 5 was questionable, and item 8 was just above the critical value.

Correlations and the Prediction of Anxiety
The CASI correlated significantly with Trait STAIC score (with the Bonferroni correction for the use of multiple correlations, the criterion of α = .0125 was used). To determine how the CASI might predict anxiety, a hierarchical multiple regression analysis was conducted. To control for age, sex and state anxiety, these variables were entered in Steps 1 and 2; to investigate the incremental value   of the CASI and the interaction terms, these indexes were entered in Steps 3 and 4 ( Table  7). Sex and State STAIC score were significant predictors of Trait STAIC score, but age was not. The CASI explained an additional 22% of the variance, making a significant contribution to the prediction of trait anxiety scores. The interaction terms were not significant.

Discussion
This study suggests that AS can be adequately measured using the 18-item French version of the CASI and conceptualized as a hierarchical factor structure with four lowerorder factors -Physical Concerns, Mental Incapacity Concerns, Social Concerns and Losing Control Concerns -loading on a single higher-order factor, Anxiety Sensitivity. The completely standardized factor loadings of the CASI items were all significant, meaning that all items may be relevant in the evaluation of AS. However, item 13 has a standardized factor loading lower than .30. Item analyses revealed that this item presented low discriminability. Silverman et al. (2003) also observed a problem with the    Note. Italics indicate values of r it of less than 0.30 and of item difficulty index above 1.96. classification of item 13; this could indicate a specific conceptual problem involving the item itself and not a problem with its translation into French. As for the item difficulty index, we observed that item 5 ("It is important for me to stay in control of my feelings") was difficult for children. In fact, most children in the sample had to ask questions before they could respond to this item. The interpretation of this item and specifically the concepts of "emotion" and "control" may be problematic for children. This item may need to be reformulated in order to increase their understanding of it. The factor structure obtained was rather similar to those obtained by past studies (Adornetto et al., 2008;Essau et al., 2010;Fullana et al., 2003;Jokić-Begić et al., 2012;Muris et al., 2001;Silverman et al., 2003;van Widenfelt et al., 2002): the same factor labels (physical, mental incapacity, social concerns) applied, as did the same arrangement of items (respectively, items 6, 9, 10 and 14; items 2, 12 and 15; and items 1, 5 and 17). However, the structure observed in this study is actually more similar to the original structure reported by .
As regards reliability, the internal consistency of the CASI was acceptable for the total scale (α = .82), but it was lower for some dimensions of AS. The Cronbach's alphas for the physical, mental incapacity, social and losing control concerns were low at . 77, .55, .69 and .46, respectively (ideally >.75;Nunnally, 1978). Although these values were   (Essau et al., 2010;Jokić-Begić et al., 2012;Muris et al., 2001;van Widenfelt et al., 2002;Walsh et al., 2004) -these results suggested that the use of the global score for AS would be more appropriate with the CASI. Consistent with the literature (Deacon et al., 2002;Muris et al., 2001;van Widenfelt et al., 2002;Walsh et al., 2004), girls scored higher on the CASI than boys. This effect was also observed for the individual factors, except for social concerns. Gender role theory may explain the higher prevalence of AS in girls than in boys (e.g., Palapattu, Kingery, & Ginsburg, 2006). The expression of fear is accepted and even encouraged among girls but is inconsistent with the masculine gender role: boys are expected to face their fears and use adaptive behavior to handle the situation (Bem, 1981). The lack of a gender effect on the social concerns factor was also observed by Stewart, Taylor, and Baker (1997) in an adult population and by Deacon et al. (2002) in a child population. The absence of such an effect may be due to the fact that peer pressure to conform to gender roles is very strong in children (Brody, 1999). Children who are sex-role stereotypical are more likely to be socially accepted, while those who are not tend to be punished by social rejection (Dafflon Novelle, 2006). This observation may explain why girls and boys did not differ on the social concerns factor: the fear of social rejection is as great in girls as in boys.
As observed in the past (Muris et al., 2001;Weems et al., 1998), the CASI correlated significantly with anxiety symptoms, suggesting that high levels of AS were associated with high levels of trait anxiety. Moreover, the CASI explained a unique proportion of the variance in trait anxiety, beyond what is already explained by sex and state anxiety. Several authors have shown that AS can predict the onset of an anxiety disorder (Lau, Calamari, & Waraczynski, 1996;Silverman & Weems, 1999). Theoretically, being sensitive to high anxiety leads to a reduced ability to cope with anxiety disorders (Reiss, 1991;Reiss et al., 1986). Because of the fear of bodily sensations, AS amplifies anxiety responses. In other words, a person with high AS will interpret certain sensations as a sign of imminent danger and, therefore, become more anxious (Reiss, 1991). Accordingly, Schmidt et al. (2007) developed a primary prevention program designed to reduce AS. This program has proven effective in reducing the risk of developing an anxiety disorder in the course of the following year. Thus, AS can be a symptom that predicts anxiety disorders, and its treatment can also be effective in treating such disorders (Balle & Tortella-Feliu, 2010). From a clinical point of view, the use of a questionnaire such as the CASI is therefore useful both for prevention, namely detecting children at risk of developing an anxiety disorder, and for treatment, that is, working on a maintenance factor for anxiety such as AS.
Our study presented some limitations. First, the samples are not representative of the Belgian population as both were recruited entirely from schools in the area of Liège. In subsequent studies, the invariance of the factor structure of this French version should be investigated with samples of children from other areas of Belgium and from other French-speaking areas (France, Switzerland, Quebec). Second, our sample is limited regarding age (9 to 13 years). One cannot assume that the results will generalize across age ranges on the basis of this sample. In future studies, it would be interesting to have CFA data on factor invariance across sex, age, and educational level (primary versus secondary school). Third, the sample was recruited in a nonclinical population. Given that AS may predict the development of anxiety disorders (Anderson & Hope, 2009), future research could determine the French CASI's ability to discriminate between clinical and nonclinical samples. Fourth, the reliability is limited to internal consistency; it will be essential to evaluate test-retest reliability in future. Finally, the results were based exclusively on the children's self-reports and thus could be biased by perceived social desirability.
Overall, this version of the CASI can be seen as having a factor structure with four lowerorder factors loading on a single higherorder factor. It has acceptable reliability for the scale as a whole, and has the potential to be used to identify children with anxiety symptoms. This study provides further support for the claim that AS is a relevant concept in the Belgian French population.
Note: Conflict of interest: none

Appendix
Voici plusieurs phrases que des enfants utilisent pour se décrire. Lis chaque phrase attentivement. Indique à quel point tu as l'impression qu'elle te décrit bien. À la droite de chaque phrase, mets un X en dessous du choix qui te décrit le mieux. Il n'y a pas de bonne ou de mauvaise réponse.