- Email: [email protected]
The utility of the bifactor model in understanding unique components of anxiety sensitivity in a South Korean sample
Asian Journal of Psychiatry 22 (2016) 116–123
Contents lists available at ScienceDirect
Asian Journal of Psychiatry journal homepage: www.elsevier.com/locate/ajp
The utility of the bifactor model in understanding unique components of anxiety sensitivity in a South Korean sample$ Chad Ebesutani, Mirihae Kim* , Hee-Hoon Park Duksung Women’s University, South Korea
A R T I C L E I N F O
Article history: Received 10 January 2016 Received in revised form 8 April 2016 Accepted 19 June 2016 Keywords: Anxiety sensitivity Bifactor models Cultural differences
A B S T R A C T
The present study was the first to examine the applicability of the bifactor structure underlying the Anxiety Sensitivity Index-3 (ASI-3) in an East Asian (South Korean) sample and to determine which factors in the bifactor model were significantly associated with anxiety, depression, and negative affect. Using a sample of 289 South Korean university students, we compared (a) the original 3-factor AS model, (b) a 3-group bifactor AS model, and (c) a 2-group bifactor AS model (with only the physical and social concern group factors present). Results revealed that the 2-group bifactor AS model fit the ASI-3 data the best. Relatedly, although all ASI-3 items loaded on the general AS factor, the Cognitive Concern group factor was not defined in the bifactor model and may therefore need to be omitted in order to accurately model AS when conducting factor analysis and structural equation modeling (SEM) in cross cultural contexts. SEM results also revealed that the general AS factor was the only factor from the 2-group bifactor model that significantly predicted anxiety, depression, and negative affect. Implications and importance of this new bifactor structure of Anxiety Sensitivity in East Asian samples are discussed. ã 2016 Elsevier B.V. All rights reserved.
1. Introduction Anxiety sensitivity (AS)—the fear of physical and psychological sensations surrounding anxiety (McNally, 2002; Reiss and McNally, 1985)—is viewed as a cognitive predisposition that is relatively stable. It is a concept of growingly importance in the area of anxiety research and has been noted to be unique from the general tendency to experience negative emotional states, such as negative affect (McNally, 2002). According to the theory of AS, anxious individuals experience fear due to beliefs that anxiety-related sensations are associated with physical, social, and/or psychological negative consequences (Taylor et al., 2007). This theory is empirically supported and AS has been found to be associated (both concurrently and prospectively) with an increased risk of anxiety symptoms (e.g., panic attacks) and the development of anxiety disorders such as panic disorder, and posttraumatic stress disorder (Feldner et al., 2008; Hayward et al., 2000; Li and Zinbarg, 2007; Marshall et al., 2010; Schmidt et al., 2006). Research on AS also has broad application and utility given that AS has found to be
$ This research was supported by the 2015 Duksung Women’s University Research Grant. * Corresponding author at: Department of Psychology, Duksung Women’s University, 419 Ssangmun-dong, Dobong-gu, Seoul 132-714, South Korea. E-mail address: [email protected] (M. Kim).
http://dx.doi.org/10.1016/j.ajp.2016.06.005 1876-2018/ã 2016 Elsevier B.V. All rights reserved.
related not only to anxiety (Ebesutani et al., 2014), but also to constructs outside of the anxiety domain, including depression (Taylor et al., 1996) and substance use (Chavarria et al., 2015). 1.1. The Anxiety Sensitivity Index-3 Due to the increasing interest in AS in recent years, several selfreport measures have been developed to assess this construct. One of the most recent measures developed to assess AS is the Anxiety Sensitivity Index-3 (ASI-3; Taylor et al., 2007). The Anxiety Sensitivity Index-3 (ASI-3; Taylor et al., 2007) was developed to measure the following three fears of anxiety symptoms: (a) physical concerns, (b) social concerns, and (c) cognitive concerns. Based on both nonclinical (n = 4494) and clinical (n = 390) samples of both males and females, Taylor et al. (2007) found support for a three-factor hierarchical structure underlying the ASI-3 in their initial development study. In this model, the three specific domains of physical concerns, social concerns, and cognitive concerns are theorized to be nested within a broader AS dimension. In other words, physical concerns, social concerns, and cognitive concerns together comprise and make up the higher AS dimension. Since then, additional support has been found for this three-factor hierarchical structure in (a) mixed-gendered clinical samples (Kemper et al., 2012; Wheaton et al., 2012) with sample sizes ranging from 506 to 514 clinically-diagnosed
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
individuals, as well as in (b) mixed-gendered undergraduate samples (Lim and Kim, 2012; Wheaton et al., 2012) with sample sizes ranging from 315 to 761 nonclinical college individuals. Support for other structures of AS have also been found, such as Zvolensky et al. (2003) finding empirical support for the social concerns and cognitive concerns items combining to form a single factor, named the ‘Social-Cognitive Concerns’ factor. More recently, studies also found that a bifactor model fits the ASI data well. For example, two recent mixed-gendered university-based studies conducted by Ebesutani et al. (2014) (n = 954 undergraduate students) and Osman et al. (2010) (n = 462 undergraduate students) found significantly better fit for a bifactor model in US samples. Even more recently, a bifactor model of ASI demonstrated utility in modeling AS in the context of researching smoking in undergraduate, clinical, and community samples (Allan et al., 2015a, 2015b; Chavarria et al., 2015). 1.2. Bifactor models Bifactor models were first introduced over seventy years ago by researchers studying cognitive ability (Holzinger and Swineford, 1937). In bifactor models, a “general” factor is posited (as responsible for item variance across all items), as well as “group” factors (responsible for item variation in more narrow subdomains). All dimensions are posited as not correlated, and the reliability of group factors reflect additional shared variance not accounted for by the general factor (Reise et al., 2010). With bifactor structures, “the general and group factors are on equal conceptual footing and compete for explaining item variance.” (Reise et al., 2010). In (the more well-known) hierarchical models, on the other hand, higherorder factors account for the shared variance across lower-order factors (Brown, 2006), and thus second-order factors (e.g., “general AS”) exert their effects through first-order factors. Because of this, one advantage of the bifactor model is that it allows for easier interpretation and understanding of how specific subscale domains predict/relate to other criteria independently from the general factor (cf. Chen et al., 2006). The bifactor model has since been shown to fit the structure of psychological constructs well in adults (e.g., Reise et al., 2007), children and adolescents (e.g., Ebesutani et al., 2011). When put to the test recently in the context of anxiety sensitivity, the bifactor model fit the ASI-3 data well in two USbased studies (Osman et al., 2010; Ebesutani et al., 2014). No studies to date however have examined the degree to which the bifactor structure adequately represents Anxiety Sensitivity among East Asian samples, despite certain contextual differences existing across cultures that suggest the need for further study of this construct in diverse cultural settings.
117
taboo topics (Lee et al., 2009). For example, issues of mental health and related cognitive concerns are not openly addressed in society (such as in school systems or in employment settings). For these reasons, worries of mental health and cognitive consequences of AS (i.e., Cognitive Concerns) are not particularly salient areas of anxiety in Korean society. There will therefore likely be very little unique variance (in item responses) accounted for by a Cognitive Concerns related factor beyond that accounted for by the general AS dimension. 1.4. The present study In the present study, we therefore examined for the first time (a) the applicability and fit of the bifactor model in an East Asian (Korean) sample, (b) the salience of the Cognitive Concerns dimension relative to the general AS factor, and (c) the ability of the AS dimensions to significantly predict anxiety, depression, and negative affect. We hypothesized that a modified bifactor model (with the Cognitive Concerns domain omitted) would be the most supported model given the less emphasis of mental health and related (cognitive) concerns in the Korean society. We also hypothesized that the general AS factor would significantly predict anxiety, depression, and negative affect given its strong, salient nature as found in previous studies (Ebesutani et al., 2014). We did not make specific predictions about the specific group AS factors, given that some studies have found that they did not predict external criteria (e.g., Ebesutani et al., 2014), while others have found that they did predict external criteria, such as mood and anxiety symptoms (e.g., Allan et al., 2015b). 2. Methods 2.1. Participants The present sample included 289 undergraduate psychology students who completed an assessment battery at a large, urban South Korean university. All students who filled out the ASI-3 were included in the present study. This sample comprised 180 females (62.3%) and 109 males (37.7%). Ages ranged from 18 to 27 years old (Mage = 21.19; SD = 2.17). All participants were Korean and fluent in Korean. 2.2. Procedure Following IRB-approved consent procedures, undergraduate students in Introductory Psychology courses over the age of 18 years were asked to complete measures anonymously online for course credit.
1.3. Bifactor model for Asian samples: weaker relevance of the cognitive concerns domain
2.3. Measure
In Asian societies, the social and physical concerns dimensions will likely account for unique variability above and beyond that accounted for by the general AS dimension. For example, Asian societies have been found to have more concerns related to “saving face” than Western societies (Yabuuchi, 2004), as well as a strong emphasis on social relationships (Yum, 1988)—hence the relevance of the social concerns dimension. Physical symptoms have also been found to be strongly related to psychological distress among Asians (Kim et al., 1999)—hence, the relevance of the physical concerns dimension. However, the cognitive domain may be of weaker relevance in the area of anxiety sensitivity, particularly among Korean populations. In South Korea, mental health and related concerns (akin to AS Cognitive Concerns) are still relatively
2.3.1. Anxiety Sensitivity Index-3—Korean version (K-ASI-3, Lim and Kim, 2012) The Anxiety Sensitivity Index-3 (Taylor et al., 2007) assesses the degree to which respondents fear negative consequences related to anxiety symptom (i.e., physical concerns, cognitive concerns, and social concerns). The Korean version of the ASI-3 was translated by Lim and Kim (2012). They noted that they checked the linguistic equivalence between the original English and translated Korean versions of the ASI-3 via standard forwardbackward translation procedures. Items are rated on a 5-point Likert-type scale from 0 = very little to 4 = very much. Internal consistency for these three scales in the present study was adequate, ranging from 0.77 to 0.88.
118
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
2.3.2. Positive and Negative Affect Scale—Korean version (K-PANAS; Lee et al., 2003) The Positive and Negative Affect Scale is a 20-item self-report instrument that measures negative affect (NA) with 10 of its 20 items. All items are rated on a 5-point Likert-type scale, with 1 equating to “very slightly or not at all” and 5 equating to “extremely.” We used the NA scale as our criterion measure of negative affect in the present study. Internal consistency of the NA scale in the present study was 0.76. The sample mean for the NA scale was 13.77 (S.D. = 5.00). 2.3.3. Korean version of the 21-item Depression, Anxiety, and Stress Scale (K-DASS-21; Psychology Foundation of Australia, 2013) The Korean version of the 21-item Depression, Anxiety, and Stress Scale (K-DASS-21) assesses depression, anxiety, and stress (Lovibond and Lovibond, 1995). All items are based on a 0–3 Likerttype scale with higher scores indicating more severe problems. To ensure quality of the Korean version, all items were backtranslated into English to ensure that the translated Korean DASS-21 items correctly assess the areas targeted by the original measure. The psychometric properties of the DASS-21 have been supported in previous studies (Antony et al., 1998). We used the depression and anxiety subscales as our criterion measures of depression and anxiety, respectively. Internal consistency of the DASS anxiety scale (a = 0.73) and depression scale (a = 0.85) in the present study were adequate. The sample means for the DASS21 Depression and Anxiety scale were 4.39 (S.D. = 3.92) and 3.95 (S. D. = 3.23), respectively.
2.4. Data-analytic plan 2.4.1. Missing data Among the 289 included participants, 261 participants (90.3%) had no missing items, 23 participants (8%) had only one missing item, two participants (0.7%) had two missing items, and the remaining three participants (1.0%) had three missing items. We conducted multiple-missing-data analysis (also known as Little’s MCAR test; Little and Rubin, 1987) to examine whether data were MCAR. Results showed that the missing data were missing completely at random, x2(17) = 17, ns. We imputed missing data using the recommended full information maximum likelihood (FIML) procedures. 2.4.2. Confirmatory factor analysis We used Mplus version 7.11 (Muthén and Muthén, 2010) to conduct confirmatory factor analysis (CFA). Due to the K-ASI-3 data being categorical (ordinal) data (Brown, 2006), we based analyses on polychoric correlations (Holgado-Tello et al., 2010; Olsson, 1979). The following fit indices were used to evaluate model fit. Comparative Fit Index (CFI) values greater than 0.90 and 0.95 were used as benchmarks for acceptable and good fit, respectively (Hu and Bentler, 1999). Root Mean Square Error of Approximation (RMSEA) values less than 0.08 and 0.05 were used as benchmarks for acceptable and good fit, respectively (Browne and Cudeck, 1993). The competing models examined in the present study relative to the bifactor model were the following. First, we examined (a) the 3-factor (physical, social, & cognitive) second-order model comprising the three original factors of physical concerns, social concerns, and cognitive concerns (see Fig. 1). We also examined the (b) 1-factor (AS) model (whereby all items load on a single factor
Fig. 1. The 3-Factor Correlated-Traits and 3-Factor Bifactor Model.
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
without any lower-order factors). Lastly, the Fear of SocialCognitive Concerns factor was first proposed by Zvolensky et al. (2003) based on their factor analysis conducted on the Anxiety Sensitivity Index-Revised form; in this model, the social concerns and cognitive concerns items form a single factor, leaving only two factors: a physical concerns factor and a social-cognitive concerns factor (Zvolensky et al., 2003). We therefore also examined (c) this 2-factor (physical & socio-cognitive) ASI model, which was also recently examined by Taylor et al. (2007) using the ASI-3. We did not examine a higher-order 3-factor model given that such a model would yield the same fit as the 3-factor (correlated-traits) above (Brown, 2006). Models were compared using the x2 difference test. Given the use of the WLSMV estimator, the “difftest” function available within Mplus had to be employed to obtain the appropriate x2 difference test statistics; this is because degrees of freedom are estimated and x2 values are not distributed as standard x2 when using limited information estimators such as WLSMV are used. This “difftest” procedure and the method employed by Mplus to estimate degress of freedom are described in the Mplus’ Technical Appendices (at www.statmodel.com/ download/webnotes/webnote10.pdf; Asparouhov and Muthén 2006) and in the Mplus User’s Guide (Muthén and Muthén 2010). 2.4.3. Omega hierarchical Coefficient omega hierarchical is an index that quantifies the proportion of variance accounted for by each factor in a bifactor model and the amount of reliability associated with each scale or
119
subscale after accounting for the other factors (see Reise et al., 2010). Based on the most supported bifactor model, we computed Omega HierarchicalGeneral (the amount of total score variance that can be associated with variation on a single latent common to all the items on a scale), and Omega HierarchicalSpecific (which provides an index of the reliability of the subscale score after accounting for the general factor). 2.4.4. Explained common variance (ECV) We also computed Explained Common Variance (ECV), which represents the percent of common variance attributable to the general factor in a bifactor model (see Reise et al., 2013). The ECV has recently been recommended as an index of “degree of unidimensionality” when dealing with potentially multidimensional data. When the ECV for the general factor in a bifactor model is large (ECV > 0.60), the estimates of the factor loadings for a unidimensional model are close to the general factor loadings in the bifactor model (Brouwer et al., 2013). 2.4.5. Percentage of uncontaminated correlations (PUC) We also computed the PUC index, which represents the degree to which multidimensional data can be modeled in a unidimensional structure without being affected by large parameter bias (see Bonifay et al., 2015; Reise et al., 2013). The larger the PUC value, the less affect multidimensional data are affected when modeled as unidimensional.
Table 1 Factor loadings (and standard errors) for the 3-factor and 2-factor bifactor confirmatory factor analysis model of the Anxiety Sensitivity Index-3. 3-factor Bifactor Model Scale/ Item
G
PC
Modified (2-factor) Bifactor Model CC
SC
G
PC
Physical Concerns ASI 3 0.50 (0.055) 0.20 (0.079) ASI 4 0.56 (0.049) 0.34 (0.062)
0.50 (0.055) 0.19 (0.080) 0.57 (0.048) 0.34 (0.061)
ASI 7
0.58 (0.050) 0.52 (0.061)
0.59 (0.049)
0.51 (0.061)
ASI 8
0.52 (0.062)
0.53 (0.062)
0.74 (0.061)
ASI 12
0.59 (0.050) 0.50 (0.062)
0.60 (0.049) 0.49 (0.062)
ASI 15
0.55 (0.062) 0.36 (0.068)
0.56 (0.063) 0.35 (0.068)
0.74 (0.058)
Cognitive Concerns ASI 2 0.68 (0.045)
0.11 (0.158)
0.68 (0.039)
ASI 5
0.72 (0.043)
0.12 (0.173)
0.71 (0.037)
ASI 10
0.84 (0.067)
1.65 (0.2627)
0.76 (0.059)
ASI 14
0.78 (0.047)
0.01 (0.045)
0.78 (0.044)
ASI 16
0.75 (0.041)
0.11 (0.161)
0.75 (0.036)
ASI 18
0.77 (0.043)
0.12 (0.176)
0.77 (0.036)
Social Concerns ASI 1 0.26 (0.062) ASI 6 0.51 (0.047) ASI 9 0.55 (0.051) ASI 11 0.38 (0.056) ASI 13 0.47 (0.052)
0.53 0.72 0.67 0.42 0.45
ASI 17
0.12 (0.061) 0.43 (0.050)
0.43 (0.050)
(0.052) (0.038) (0.045) (0.056) (0.056)
0.26 0.51 0.55 0.38 0.47
(0.062) (0.047) (0.051) (0.056) (0.052)
SC
Abbreviated item content (in English)
It scares me when my heart beats rapidly When my stomach is upset I worry that I might be seriously ill When my chest feels tight I get scared I won't be able to breathe properly When I feel pain in my chest I worry that I'm going to have a heart attack When heart skipping a beat I worry something wrong with me When my throat feels tight I worry I could choke to death
When I cannot keep my mind on a task, I worry that I might be going crazy It scares me when I am unable to keep my mind on a task When I feel spacey or spaced out I worry that I may be mentally ill When my thoughts speed up I worry I might be going crazy When I have trouble thinking clearly I worry something wrong with me When my mind goes blank I worry there is something terribly wrong with me
0.53 0.72 0.67 0.41 0.44
It is important for me not to appear nervous When I tremble I fear what people might think of me I worry that other people will notice my anxiety It scares me when I blush in front of other people When I sweat in social situations I fear people will think negatively of me 0.11 (0.060) I think that it would be horrific for me to faint in public
Note: ASI = Anxiety Sensitivity Index; “G” = the general ASI factor. All loadings in bold were significant.
(0.052) (0.038) (0.046) (0.056) (0.056)
120
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
Fig. 2. Bifactor model of the Korean Version of the Anxiety Sensitivity Index-2 positing a general “anxiety sensitivity” factor and two group factors related to “social concerns” and “physical concerns”.
2.4.6. Unique contribution of group factors We then used structural equation modeling (SEM) to examine the utility of the K-ASI-3 factors with respect to predicting external criteria of anxiety, depression, and negative affect. 3. Results 3.1. Confirmatory bifactor analysis The (3-factor) bifactor model (Fig. 1) fit the K-ASI-3 data well. However, all items from the Cognitive Concerns subdomain loaded significantly only on the general AS factor and not on the Cognitive Concerns factor (see Table 1 for all standardized factor loadings1 ). This led to a modified (2-factor) bifactor model (Fig. 2), whereby
1 ASI-3 #10 (i.e., “When I feel spacey or spaced out I worry that I may be mentally ill”) had an unusually large and negative loading on the Cognitive Concerns domain (i.e., 1.65). Closer inspection revealed the presence of an undefined R-square and a high and negative residual variance (i.e., 2.43) associated with this item. Negative residual variances can occur for items with strong floor effects, which may have been the case for ASI-3 #10 (given its item content).
the items that once comprised the Cognitive Concerns subdomain were assigned only to the general AS factor. This modified (2-factor) bifactor model was associated with good model fit. The x2 difference test between the original (3-factor) bifactor model and modified (2-factor) bifactor model was not significant, x2diff(6) = 20.30, ns. The (3-factor) bifactor model therefore did not fit significantly better than the simplified (2-factor) bifactor model. These results provide support for the 2-factor bifactor model due to it being more parsimonious than the (3-factor) bifactor model and not associated with significantly worse model fit. 3.1.1. Model comparisons Given the support for the (2-factor) bifactor model, we compared it to alternate models (see Table 2). The (2-factor) bifactor model fit significantly better than the competing 1-factor (AS) model [x2diff (12) = 296.60, p < 0.001], 2-factor correlated traits model [x2diff (11) = 185.25, p < 0.001], and the 3-factor correlated traits model [x2iff (9) = 49.93, p < 0.001]. These results therefore provide additional support for the 2-factor bifactor model.
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
121
Table 2 Fit statistics for the confirmatory factor analytic models (N = 288). Model
x2
df
RMSEA
CFI
x2diff
Bifactor model Modified (2-factor) Bifactor model (no cognitive subdomain) 1 Factor (Unidimensional) model 2 Factor (Correlated Traits) model 3 Factor (Correlated Traits) model
264.18 279.88 740.41 564.74 329.81
117 123 135 134 132
0.066 0.066 0.125 0.105 0.072
0.957 0.954 0.823 0.874 0.942
20.30, ns 296.60, p < 0.001 185.25, p < 0.001 49.93, p < 0.001
Note: RMSEA = root mean square error of approximation; CFI = comparative fit index; Correlated traits models yield the same fit indices as second-order (hierarchical) models with one higher-order factor and the same number of lower-order factors (Brown, 2006).
Fig. 3. Structural Equation Model Predicting Anxiety, Depression, and Negative Affect.
3.1.2. Omega hierarchical (general and specific) indices Omega hierarchical indices based on the 2-factor bifactor CFA model were as follows: Omega HierarchicalGeneral = 0.82; Omega HierarchicalSpecific Physical Concerns = 0.33; Omega HierarchicalSpecific Social Concerns = 0.46. These results indicate that the general AS factor yielded the most reliable scores and explained the largest proportion of variance after accounting for
the other factors. These results thus support the presence of a strong general AS factor consistent with a bifactor model. 3.1.3. Explained common variance (ECV) and percentage of uncontaminated correlations (PUC) The ECV value was 0.68 and the PUV value was 0.80. The indices support the notion that the two-factor bifactor model—although
122
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
multidimensional in nature—may be modeled and treated as unidimensional without producing a high degree of parameter estimation bias. 3.2. Unique contribution of group factors2 All three measurement models were acceptable (Anxiety Model: RMSEA = 0.051, CFI = 0.953; Depression Model: RMSEA = 0.048, CFI = 0.965; Negative Affect Model: RMSEA = 0.071, CFI = 0.881). All factor loadings were also significant in all models. As seen in Fig. 3, various SEM models were used to test the degree to which the factors belonging to the bifactor K-ASI structure (i.e., General AS, Social Concerns, and Physical Concerns) were significantly associated with external criteria (i.e., Anxiety, Depression, and Negative Affect). Results revealed that only the General AS factor significantly predicted Anxiety, Depression and Negative Affect. 4. Discussion The current study was the first to examine the applicability of the bifactor structure underlying the ASI-3 in an East Asian (Korean) sample. The present study found that the (modified 2factor socio-physical) bifactor model fit the present Korean sample well. This bifactor model also fit significantly better than all competing models tested so far in other studies, including the one-, two-, and three-factor (correlated traits) model. Interestingly, the previously assigned Cognitive Concerns items loaded significantly only on the General AS dimension, but not on the “Cognitive Concerns” dimension. This suggests that the “cognitive concerns” items may not account for unique item variability beyond the general AS dimension. The construct of ‘AS Cognitive Concerns’ itself is therefore not well differentiated apart from the AS General Factor (and therefore interpretation of any subscale scores derived from the originally posited ‘AS Cognitive Concerns’ domain would be not advisable). These findings suggest that anxiety sensitivity does not get expressed among Koreans in the same way as in the US, particularly regarding cognitive concerns (of mental health), such as the fear of ‘going crazy’ or ‘mentally ill.’ More research however is needed to verify whether this is indeed the case before discarding the Cognitive Concerns scale from the K-ASI-3. Additional studies would also be needed to examine whether the salience of this Cognitive Concerns dimension strengths over time, particularly as cognitive-related mental health concerns becomes of increased focus in the future in Korea. It is important to note that other studies have also found that the expression of psychopathology varies as a function of culture and ethnicity. For example, Okazaki (1997) found that AsianAmericans’ behaviors were associated with greater avoidance behaviors in social situations compared to White Americans. Pina and Silverman (2004) also more recently found that Hispanic/ Latino youth experience more somatic symptoms when experiencing anxiety than European Americans. The present findings further support this notion that the experience and structure of psychopathology can differ across cultures. Therefore, attention should be given to such possible differences when constructs (such as anxiety sensitivity) are identified and applied in new, crosscultural settings.
2 At a p < 0.05 significance level, with 4 latent variables, 18 observed variables, and a desired statistical power level of 0.80, a minimum sample size of 88 is recommended to detect a medium and large effect, and a minimum sample size of 387 is recommended to detect a small effect (Westland, 2012).
Implications of this new bifactor structure of Anxiety Sensitivity in East Asian (Korean) samples includes the following. First, this was the first study to extend the applicability of the emerging bifactor model of Anxiety Sensitivity to East Asian (Korean) samples. Second, although recent studies are showing that bifactor models well-represent psychological constructs, this study showed that the exact same bifactor structure cannot be assumed to be applicable across different cultures. For example, although the bifactor structure was the most supported model in our Korean sample, the structure of the bifactor model was slightly different than that supported in the US (Ebesutani et al., 2014). Specifically, Ebesutani et al. (2014) found superior fit for a bifactor model with a general AS factor and three content domains (e.g., x2 = 246.49; RMSEA = 0.048; CFI = 0.997), whereas in the current study, superior fit was found for a bifactor model with a general AS factor and two content domains (e.g., x2 = 279.88; RMSEA = 0.066; CFI = 0.954). Consistent with the cultural considerations, the Cognitive Concerns domain did not emerge as a salience dimension above and beyond the general AS factor. Caution should thus be used when directly applying the same bifactor models to Asian samples. Third, and relatedly, when modeling Anxiety Sensitivity in factor analysis and structural equation modeling (for both research and clinical purposes) in East Asian contexts, it may be necessary to omit the Cognitive Concerns specific domain in order to reduce error with respect to measuring Anxiety Sensitivity in Korean individuals. Lastly, this study suggests that the general AS factor is best able to predict external variables such as anxiety, depression and negative affect and thus should be given more weight over the other specific AS domains. This point however is often misunderstood. This does not mean that the other (specific) domains (of Social Concerns and Physical Concerns) and the structure of the bifactor model do not matter; they do matter. They matter a lot. That is, in order to properly model and estimate the general AS factor, AS should be modeled not only in a bifactor structure, but also with the appropriate bifactor structure—in this case, by omitting the Cognitive Concerns specific domain, yet retaining its items to identify and measure the general AS construct among South Korean individuals. Although the present study helped to identify a new bifactor structure and informed ASI scoring related to East Asian (Korean) individuals, there were limitations worth noting. First, only Korean university students were included in the study. We therefore do not know how well the present results generalize to clinical Korean samples or other East Asian cultures. The measures used in the present study were also all self-report instruments, reducing measurement precision relative to a multi-informant, multi-modal assessment approach. Nonetheless, it is hoped that the present study leads to additional research that further clarifies the structure of Anxiety Sensitivity, including informing how to best measure, score, and interpret this important construct to identify and monitor anxiety sensitivity across diverse populations. References Allan, N.P., Albanese, B.J., Short, N.A., Raines, A.M., Schmidt, N.B., 2015a. Support for the general and specific bifactor model factors of anxiety sensitivity. Personal. Individ. Differ. 74, 78–83. Allan, N.P., Macatee, R.J., Norr, A.M., Raines, A.M., Schmidt, N.B., 2015b. Relations between common and specific factors of anxiety sensitivity and distress tolerance and fear, distress, and alcohol and substance use disorders. J. Anxiety Disord. 33, 81–89. Antony, M.M., Bieling, P.J., Cox, B.J., Enns, M.W., Swinson, R.P., 1998. Psychometric properties of the 42-item and 21-item versions of the depression anxiety stress scales in clinical groups and a community sample. Psychol. Assess. 10, 176–181. Asparouhov, T., & Muthén, B., 2006. Robust chi square difference testing with mean and variance adjusted test statistics. Mplus Web Notes: No. 10. May 26, 2006. Retrieved September 26, 2010, from http://www.statmodel.com/download/ webnotes/webnote10.pdf.
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123 Bonifay, W.E., Reise, S.P., Scheines, R., Meijer, R.R., 2015. When are multidimensional data unidimensional enough for structural equation modeling? An evaluation of the DETECT multidimensionality index. Struct. Equ. Model. 22 (4), 504–516. Brouwer, D., Meijer, R.R., Zevalkink, J., 2013. On the factor structure of the Beck Depression Inventory-II: G is the key. Psychol. Assess. 25 (1), 136–145. Brown, T., 2006. Confirmatory Factor Analysis for Applied Research. Guilford Press, New York, NY. Browne, M., Cudeck, R., 1993. Alternative ways of assessing model fit. In: Bollen, K.A., Long, J.S. (Eds.), Testing Structural Equation Models. Sage, Newbury Park, CA, pp. 136–159. Chavarria, J., Allan, N.P., Boffa, J.W., Albanese, B.J., Zvolensky, M.J., Schmidt, N.B., 2015. Decomposing the relationship between anxiety sensitivity and alcohol use. J. Stud. Alcohol Drugs 76, 957–961. Chen, F.F., West, S.G., Sousa, K.H., 2006. A comparison of bifactor and second-order models of quality of life. Multivar. Behav. Res. 41, 189–225. Ebesutani, C., Smith, A., Bernstein, A., Chorpita, B.F., Higa-McMillan, C., Nakamura, B. J., 2011. A bifactor model of negative affectivity: fear and distress components among younger and older youth. Psychol. Assess. 23, 679–691. Ebesutani, C.K., McLeish, A.C., Luberto, C.M., Young, J., Maack, D.J., 2014. A bifactor model of anxiety sensitivity: analysis of the Anxiety Sensitivity Index-3. J. Psychopathol. Behav. Assess. 36 (September (3)), 452–464. Feldner, M., Zvolensky, M., Schmidt, N., Smith, R., 2008. A prospective test of anxiety sensitivity as a moderator of the relation between gender and posttraumatic symptom maintenance among anxiety sensitive young adults. Depress. Anxiety 25, 190–199. Hayward, C., Killen, J., Kraemer, H., Taylor, C., 2000. Predictors of panic attacks in adolescents. J. Am. Acad. Child Adolesc. Psychiatry 29, 207–214. Holgado-Tello, F., Chacón-Moscoso, S., Barbero-García, I., Vila-Abad, E., 2010. Polychoric versus Pearson correlations in exploratory and confirmatory factor analysis of ordinal variables. Qual. Quant. 44 (1), 153–166. Holzinger, K.J., Swineford, R., 1937. The bifactor method. Psychometrika 2, 41–54. Hu, L.T., Bentler, P.M., 1999. Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives. Struct. Equ. Model. 6, 1– 55. Kemper, C.J., Lutz, J., Bahr, T., Ruddel, H., Hock, M., 2012. Construct validity of the Anxiety Sensitivity Index-3 in clinical samples. Assessment 19, 89–100. Kim, K., Li, D., Kim, D., 1999. Depressive symptoms in Koreans, Korean-Chinese and Chinese: a transcultural study. Transcult. Psychiatry 36 (3), 303–316. Lee, H., Kim, E., Lee, M., 2003. A validation study of Korean Positive and Negative Affect Schedule: the PANAS scales. Korean J. Clin. Psychol. 22 (4), 935–946. Lee, S., Juon, H.S., Martinez, G., Hsu, C.E., Robinson, E.S., Bawa, J., Ma, G.X., 2009. Model minority at risk: expressed needs of mental health by Asian American young adults. J. Community Health 34 (2), 144–152. Li, W., Zinbarg, R.E., 2007. Anxiety sensitivity and panic attacks: a 1-year longitudinal study. Behav. Modif. 31, 145–161. Lim, Y.J., Kim, J.H., 2012. Korean Anxiety Sensitivity Index-3: it’s factor structure, reliability, and validity in non-clinical samples. Psychiatry Investig. 9, 45–53. Little, R.J.A., Rubin, D.B., 1987. Statistical Analysis with Missing Data. Wiley, New York. Lovibond, S.H., Lovibond, P.F., 1995. Manual for the Depression Anxiety Stress Scales, 2nd ed. Psychology Foundation of Australia, Sydney, Australia.
123
Marshall, G.N., Miles, J.N., Stewart, S.H., 2010. Anxiety sensitivity and PTSD symptom severity are reciprocally related: evidence from a longitudinal study of physical trauma survivors. J. Abnorm. Psychol. 119, 145–150. McNally, R., 2002. Anxiety sensitivity and panic disorder. Biol. Psychiatry 52, 938– 946. Muthén, L.K., Muthén, B.O., 2010. Mplus User’s Guide, 4th ed. Author, Los Angeles, CA. Okazaki, S., 1997. Sources of ethnic differences between Asian American and white American college students on measures of depression and social anxiety. J. Abnorm. Psychol. 106 (1), 52. Olsson, U., 1979. Maximum likelihood estimation of the polychoric correlation coefficient. Psychometrika 44 (4), 443–460. Osman, A., Gutierrez, P., Smith, K., Fang, Q., Lozano, G., Devine, A., 2010. The Anxiety Sensitivity Index-3: analyses of dimensions, reliability estimates, and correlates in nonclinical samples. J. Personal. Assess. 92, 45–52. Pina, A.A., Silverman, W.K., 2004. Clinical phenomenology, somatic symptoms, and distress in Hispanic/Latino and European American youths with anxiety disorders. J. Clin. Child Adolesc. Psychol. 33 (2), 227–236. Psychology Foundation of Australia, 2013. DASS. Unpublished instrument. Retrieved March 23, 2013, from http://www2.psy.unsw.edu.au/groups/dass. Reise, S., Morizot, J., Hays, R., 2007. The role of the bifactor model in resolving dimensionality issues in health outcomes measures. Qual. Life Res. 16, 19–31. Reise, S., Moore, T., Haviland, M., 2010. Bifactor models and rotations: exploring the extent to which multidimensional data yield univocal scale scores. J. Personal. Assess. 92, 544–559. Reise, S.P., Scheines, R., Widaman, K.F., Haviland, M.G., 2013. Multidimensionality and structural coefficient bias in structural equation modeling a bifactor perspective. Educ. Psychol. Meas. 73 (1), 5–26. Reiss, S., McNally, R., 1985. Expectancy model of fear. In: Reiss, S., Bootzin, R.R. (Eds.), Theoretical Issues in Behavior Therapy. Academic Press, New York, NY, pp. 107– 122. Schmidt, N., Zvolensky, M., Maner, J., 2006. Anxiety sensitivity: prospective prediction of panic attacks and Axis I psychopathology. J. Psychiatr. Res. 40, 691–699. Taylor, S., Koch, W.J., Woody, S., McLean, P., 1996. Anxiety sensitivity and depression: how are they related? J. Abnorm. Psychol. 105 (3), 474. Taylor, S., Zvolensky, M.J., Cox, B.J., Deacon, B., Heimberg, R.G., Ledley, R.G., Cardenas, S.J., 2007. Robust dimensions of anxiety sensitivity: development and initial validation of the Anxiety Sensitivity Index-3. Psychol. Assess. 19, 176–188. Westland, J.C., 2012. Erratum to Lower bounds on sample size in structural equation modeling [Electron. Commerce Res. Appl. 9 (6) (2010) 476–487]. Electron. Commer. Res. Appl. 11 (4), 445. Wheaton, M.G., Deacon, B.J., McGrath, P.B., Berman, N.C., Abramowitz, J.S., 2012. Dimensions of anxiety sensitivity in the anxiety disorders: evaluation of the ASI-3. J. Anxiety Disord. 26, 401–408. Yabuuchi, A., 2004. Face in Chinese, Japanese, and US American cultures. J. Asian Pac. Commun. 14 (2), 261–297. Yum, J.O., 1988. The impact of Confucianism on interpersonal relationships and communication patterns in East Asia. Commun. Monogr. 55 (4), 374–388. Zvolensky, M.J., Arrindell, W.A., Taylor, S., Bouvard, M., Cox, B.J., Stewart, S.H., Eifert, G.H., 2003. Anxiety sensitivity in six countries. Behav. Res. Ther. 7, 841–859.
Contents lists available at ScienceDirect
Asian Journal of Psychiatry journal homepage: www.elsevier.com/locate/ajp
The utility of the bifactor model in understanding unique components of anxiety sensitivity in a South Korean sample$ Chad Ebesutani, Mirihae Kim* , Hee-Hoon Park Duksung Women’s University, South Korea
A R T I C L E I N F O
Article history: Received 10 January 2016 Received in revised form 8 April 2016 Accepted 19 June 2016 Keywords: Anxiety sensitivity Bifactor models Cultural differences
A B S T R A C T
The present study was the first to examine the applicability of the bifactor structure underlying the Anxiety Sensitivity Index-3 (ASI-3) in an East Asian (South Korean) sample and to determine which factors in the bifactor model were significantly associated with anxiety, depression, and negative affect. Using a sample of 289 South Korean university students, we compared (a) the original 3-factor AS model, (b) a 3-group bifactor AS model, and (c) a 2-group bifactor AS model (with only the physical and social concern group factors present). Results revealed that the 2-group bifactor AS model fit the ASI-3 data the best. Relatedly, although all ASI-3 items loaded on the general AS factor, the Cognitive Concern group factor was not defined in the bifactor model and may therefore need to be omitted in order to accurately model AS when conducting factor analysis and structural equation modeling (SEM) in cross cultural contexts. SEM results also revealed that the general AS factor was the only factor from the 2-group bifactor model that significantly predicted anxiety, depression, and negative affect. Implications and importance of this new bifactor structure of Anxiety Sensitivity in East Asian samples are discussed. ã 2016 Elsevier B.V. All rights reserved.
1. Introduction Anxiety sensitivity (AS)—the fear of physical and psychological sensations surrounding anxiety (McNally, 2002; Reiss and McNally, 1985)—is viewed as a cognitive predisposition that is relatively stable. It is a concept of growingly importance in the area of anxiety research and has been noted to be unique from the general tendency to experience negative emotional states, such as negative affect (McNally, 2002). According to the theory of AS, anxious individuals experience fear due to beliefs that anxiety-related sensations are associated with physical, social, and/or psychological negative consequences (Taylor et al., 2007). This theory is empirically supported and AS has been found to be associated (both concurrently and prospectively) with an increased risk of anxiety symptoms (e.g., panic attacks) and the development of anxiety disorders such as panic disorder, and posttraumatic stress disorder (Feldner et al., 2008; Hayward et al., 2000; Li and Zinbarg, 2007; Marshall et al., 2010; Schmidt et al., 2006). Research on AS also has broad application and utility given that AS has found to be
$ This research was supported by the 2015 Duksung Women’s University Research Grant. * Corresponding author at: Department of Psychology, Duksung Women’s University, 419 Ssangmun-dong, Dobong-gu, Seoul 132-714, South Korea. E-mail address: [email protected] (M. Kim).
http://dx.doi.org/10.1016/j.ajp.2016.06.005 1876-2018/ã 2016 Elsevier B.V. All rights reserved.
related not only to anxiety (Ebesutani et al., 2014), but also to constructs outside of the anxiety domain, including depression (Taylor et al., 1996) and substance use (Chavarria et al., 2015). 1.1. The Anxiety Sensitivity Index-3 Due to the increasing interest in AS in recent years, several selfreport measures have been developed to assess this construct. One of the most recent measures developed to assess AS is the Anxiety Sensitivity Index-3 (ASI-3; Taylor et al., 2007). The Anxiety Sensitivity Index-3 (ASI-3; Taylor et al., 2007) was developed to measure the following three fears of anxiety symptoms: (a) physical concerns, (b) social concerns, and (c) cognitive concerns. Based on both nonclinical (n = 4494) and clinical (n = 390) samples of both males and females, Taylor et al. (2007) found support for a three-factor hierarchical structure underlying the ASI-3 in their initial development study. In this model, the three specific domains of physical concerns, social concerns, and cognitive concerns are theorized to be nested within a broader AS dimension. In other words, physical concerns, social concerns, and cognitive concerns together comprise and make up the higher AS dimension. Since then, additional support has been found for this three-factor hierarchical structure in (a) mixed-gendered clinical samples (Kemper et al., 2012; Wheaton et al., 2012) with sample sizes ranging from 506 to 514 clinically-diagnosed
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
individuals, as well as in (b) mixed-gendered undergraduate samples (Lim and Kim, 2012; Wheaton et al., 2012) with sample sizes ranging from 315 to 761 nonclinical college individuals. Support for other structures of AS have also been found, such as Zvolensky et al. (2003) finding empirical support for the social concerns and cognitive concerns items combining to form a single factor, named the ‘Social-Cognitive Concerns’ factor. More recently, studies also found that a bifactor model fits the ASI data well. For example, two recent mixed-gendered university-based studies conducted by Ebesutani et al. (2014) (n = 954 undergraduate students) and Osman et al. (2010) (n = 462 undergraduate students) found significantly better fit for a bifactor model in US samples. Even more recently, a bifactor model of ASI demonstrated utility in modeling AS in the context of researching smoking in undergraduate, clinical, and community samples (Allan et al., 2015a, 2015b; Chavarria et al., 2015). 1.2. Bifactor models Bifactor models were first introduced over seventy years ago by researchers studying cognitive ability (Holzinger and Swineford, 1937). In bifactor models, a “general” factor is posited (as responsible for item variance across all items), as well as “group” factors (responsible for item variation in more narrow subdomains). All dimensions are posited as not correlated, and the reliability of group factors reflect additional shared variance not accounted for by the general factor (Reise et al., 2010). With bifactor structures, “the general and group factors are on equal conceptual footing and compete for explaining item variance.” (Reise et al., 2010). In (the more well-known) hierarchical models, on the other hand, higherorder factors account for the shared variance across lower-order factors (Brown, 2006), and thus second-order factors (e.g., “general AS”) exert their effects through first-order factors. Because of this, one advantage of the bifactor model is that it allows for easier interpretation and understanding of how specific subscale domains predict/relate to other criteria independently from the general factor (cf. Chen et al., 2006). The bifactor model has since been shown to fit the structure of psychological constructs well in adults (e.g., Reise et al., 2007), children and adolescents (e.g., Ebesutani et al., 2011). When put to the test recently in the context of anxiety sensitivity, the bifactor model fit the ASI-3 data well in two USbased studies (Osman et al., 2010; Ebesutani et al., 2014). No studies to date however have examined the degree to which the bifactor structure adequately represents Anxiety Sensitivity among East Asian samples, despite certain contextual differences existing across cultures that suggest the need for further study of this construct in diverse cultural settings.
117
taboo topics (Lee et al., 2009). For example, issues of mental health and related cognitive concerns are not openly addressed in society (such as in school systems or in employment settings). For these reasons, worries of mental health and cognitive consequences of AS (i.e., Cognitive Concerns) are not particularly salient areas of anxiety in Korean society. There will therefore likely be very little unique variance (in item responses) accounted for by a Cognitive Concerns related factor beyond that accounted for by the general AS dimension. 1.4. The present study In the present study, we therefore examined for the first time (a) the applicability and fit of the bifactor model in an East Asian (Korean) sample, (b) the salience of the Cognitive Concerns dimension relative to the general AS factor, and (c) the ability of the AS dimensions to significantly predict anxiety, depression, and negative affect. We hypothesized that a modified bifactor model (with the Cognitive Concerns domain omitted) would be the most supported model given the less emphasis of mental health and related (cognitive) concerns in the Korean society. We also hypothesized that the general AS factor would significantly predict anxiety, depression, and negative affect given its strong, salient nature as found in previous studies (Ebesutani et al., 2014). We did not make specific predictions about the specific group AS factors, given that some studies have found that they did not predict external criteria (e.g., Ebesutani et al., 2014), while others have found that they did predict external criteria, such as mood and anxiety symptoms (e.g., Allan et al., 2015b). 2. Methods 2.1. Participants The present sample included 289 undergraduate psychology students who completed an assessment battery at a large, urban South Korean university. All students who filled out the ASI-3 were included in the present study. This sample comprised 180 females (62.3%) and 109 males (37.7%). Ages ranged from 18 to 27 years old (Mage = 21.19; SD = 2.17). All participants were Korean and fluent in Korean. 2.2. Procedure Following IRB-approved consent procedures, undergraduate students in Introductory Psychology courses over the age of 18 years were asked to complete measures anonymously online for course credit.
1.3. Bifactor model for Asian samples: weaker relevance of the cognitive concerns domain
2.3. Measure
In Asian societies, the social and physical concerns dimensions will likely account for unique variability above and beyond that accounted for by the general AS dimension. For example, Asian societies have been found to have more concerns related to “saving face” than Western societies (Yabuuchi, 2004), as well as a strong emphasis on social relationships (Yum, 1988)—hence the relevance of the social concerns dimension. Physical symptoms have also been found to be strongly related to psychological distress among Asians (Kim et al., 1999)—hence, the relevance of the physical concerns dimension. However, the cognitive domain may be of weaker relevance in the area of anxiety sensitivity, particularly among Korean populations. In South Korea, mental health and related concerns (akin to AS Cognitive Concerns) are still relatively
2.3.1. Anxiety Sensitivity Index-3—Korean version (K-ASI-3, Lim and Kim, 2012) The Anxiety Sensitivity Index-3 (Taylor et al., 2007) assesses the degree to which respondents fear negative consequences related to anxiety symptom (i.e., physical concerns, cognitive concerns, and social concerns). The Korean version of the ASI-3 was translated by Lim and Kim (2012). They noted that they checked the linguistic equivalence between the original English and translated Korean versions of the ASI-3 via standard forwardbackward translation procedures. Items are rated on a 5-point Likert-type scale from 0 = very little to 4 = very much. Internal consistency for these three scales in the present study was adequate, ranging from 0.77 to 0.88.
118
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
2.3.2. Positive and Negative Affect Scale—Korean version (K-PANAS; Lee et al., 2003) The Positive and Negative Affect Scale is a 20-item self-report instrument that measures negative affect (NA) with 10 of its 20 items. All items are rated on a 5-point Likert-type scale, with 1 equating to “very slightly or not at all” and 5 equating to “extremely.” We used the NA scale as our criterion measure of negative affect in the present study. Internal consistency of the NA scale in the present study was 0.76. The sample mean for the NA scale was 13.77 (S.D. = 5.00). 2.3.3. Korean version of the 21-item Depression, Anxiety, and Stress Scale (K-DASS-21; Psychology Foundation of Australia, 2013) The Korean version of the 21-item Depression, Anxiety, and Stress Scale (K-DASS-21) assesses depression, anxiety, and stress (Lovibond and Lovibond, 1995). All items are based on a 0–3 Likerttype scale with higher scores indicating more severe problems. To ensure quality of the Korean version, all items were backtranslated into English to ensure that the translated Korean DASS-21 items correctly assess the areas targeted by the original measure. The psychometric properties of the DASS-21 have been supported in previous studies (Antony et al., 1998). We used the depression and anxiety subscales as our criterion measures of depression and anxiety, respectively. Internal consistency of the DASS anxiety scale (a = 0.73) and depression scale (a = 0.85) in the present study were adequate. The sample means for the DASS21 Depression and Anxiety scale were 4.39 (S.D. = 3.92) and 3.95 (S. D. = 3.23), respectively.
2.4. Data-analytic plan 2.4.1. Missing data Among the 289 included participants, 261 participants (90.3%) had no missing items, 23 participants (8%) had only one missing item, two participants (0.7%) had two missing items, and the remaining three participants (1.0%) had three missing items. We conducted multiple-missing-data analysis (also known as Little’s MCAR test; Little and Rubin, 1987) to examine whether data were MCAR. Results showed that the missing data were missing completely at random, x2(17) = 17, ns. We imputed missing data using the recommended full information maximum likelihood (FIML) procedures. 2.4.2. Confirmatory factor analysis We used Mplus version 7.11 (Muthén and Muthén, 2010) to conduct confirmatory factor analysis (CFA). Due to the K-ASI-3 data being categorical (ordinal) data (Brown, 2006), we based analyses on polychoric correlations (Holgado-Tello et al., 2010; Olsson, 1979). The following fit indices were used to evaluate model fit. Comparative Fit Index (CFI) values greater than 0.90 and 0.95 were used as benchmarks for acceptable and good fit, respectively (Hu and Bentler, 1999). Root Mean Square Error of Approximation (RMSEA) values less than 0.08 and 0.05 were used as benchmarks for acceptable and good fit, respectively (Browne and Cudeck, 1993). The competing models examined in the present study relative to the bifactor model were the following. First, we examined (a) the 3-factor (physical, social, & cognitive) second-order model comprising the three original factors of physical concerns, social concerns, and cognitive concerns (see Fig. 1). We also examined the (b) 1-factor (AS) model (whereby all items load on a single factor
Fig. 1. The 3-Factor Correlated-Traits and 3-Factor Bifactor Model.
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
without any lower-order factors). Lastly, the Fear of SocialCognitive Concerns factor was first proposed by Zvolensky et al. (2003) based on their factor analysis conducted on the Anxiety Sensitivity Index-Revised form; in this model, the social concerns and cognitive concerns items form a single factor, leaving only two factors: a physical concerns factor and a social-cognitive concerns factor (Zvolensky et al., 2003). We therefore also examined (c) this 2-factor (physical & socio-cognitive) ASI model, which was also recently examined by Taylor et al. (2007) using the ASI-3. We did not examine a higher-order 3-factor model given that such a model would yield the same fit as the 3-factor (correlated-traits) above (Brown, 2006). Models were compared using the x2 difference test. Given the use of the WLSMV estimator, the “difftest” function available within Mplus had to be employed to obtain the appropriate x2 difference test statistics; this is because degrees of freedom are estimated and x2 values are not distributed as standard x2 when using limited information estimators such as WLSMV are used. This “difftest” procedure and the method employed by Mplus to estimate degress of freedom are described in the Mplus’ Technical Appendices (at www.statmodel.com/ download/webnotes/webnote10.pdf; Asparouhov and Muthén 2006) and in the Mplus User’s Guide (Muthén and Muthén 2010). 2.4.3. Omega hierarchical Coefficient omega hierarchical is an index that quantifies the proportion of variance accounted for by each factor in a bifactor model and the amount of reliability associated with each scale or
119
subscale after accounting for the other factors (see Reise et al., 2010). Based on the most supported bifactor model, we computed Omega HierarchicalGeneral (the amount of total score variance that can be associated with variation on a single latent common to all the items on a scale), and Omega HierarchicalSpecific (which provides an index of the reliability of the subscale score after accounting for the general factor). 2.4.4. Explained common variance (ECV) We also computed Explained Common Variance (ECV), which represents the percent of common variance attributable to the general factor in a bifactor model (see Reise et al., 2013). The ECV has recently been recommended as an index of “degree of unidimensionality” when dealing with potentially multidimensional data. When the ECV for the general factor in a bifactor model is large (ECV > 0.60), the estimates of the factor loadings for a unidimensional model are close to the general factor loadings in the bifactor model (Brouwer et al., 2013). 2.4.5. Percentage of uncontaminated correlations (PUC) We also computed the PUC index, which represents the degree to which multidimensional data can be modeled in a unidimensional structure without being affected by large parameter bias (see Bonifay et al., 2015; Reise et al., 2013). The larger the PUC value, the less affect multidimensional data are affected when modeled as unidimensional.
Table 1 Factor loadings (and standard errors) for the 3-factor and 2-factor bifactor confirmatory factor analysis model of the Anxiety Sensitivity Index-3. 3-factor Bifactor Model Scale/ Item
G
PC
Modified (2-factor) Bifactor Model CC
SC
G
PC
Physical Concerns ASI 3 0.50 (0.055) 0.20 (0.079) ASI 4 0.56 (0.049) 0.34 (0.062)
0.50 (0.055) 0.19 (0.080) 0.57 (0.048) 0.34 (0.061)
ASI 7
0.58 (0.050) 0.52 (0.061)
0.59 (0.049)
0.51 (0.061)
ASI 8
0.52 (0.062)
0.53 (0.062)
0.74 (0.061)
ASI 12
0.59 (0.050) 0.50 (0.062)
0.60 (0.049) 0.49 (0.062)
ASI 15
0.55 (0.062) 0.36 (0.068)
0.56 (0.063) 0.35 (0.068)
0.74 (0.058)
Cognitive Concerns ASI 2 0.68 (0.045)
0.11 (0.158)
0.68 (0.039)
ASI 5
0.72 (0.043)
0.12 (0.173)
0.71 (0.037)
ASI 10
0.84 (0.067)
1.65 (0.2627)
0.76 (0.059)
ASI 14
0.78 (0.047)
0.01 (0.045)
0.78 (0.044)
ASI 16
0.75 (0.041)
0.11 (0.161)
0.75 (0.036)
ASI 18
0.77 (0.043)
0.12 (0.176)
0.77 (0.036)
Social Concerns ASI 1 0.26 (0.062) ASI 6 0.51 (0.047) ASI 9 0.55 (0.051) ASI 11 0.38 (0.056) ASI 13 0.47 (0.052)
0.53 0.72 0.67 0.42 0.45
ASI 17
0.12 (0.061) 0.43 (0.050)
0.43 (0.050)
(0.052) (0.038) (0.045) (0.056) (0.056)
0.26 0.51 0.55 0.38 0.47
(0.062) (0.047) (0.051) (0.056) (0.052)
SC
Abbreviated item content (in English)
It scares me when my heart beats rapidly When my stomach is upset I worry that I might be seriously ill When my chest feels tight I get scared I won't be able to breathe properly When I feel pain in my chest I worry that I'm going to have a heart attack When heart skipping a beat I worry something wrong with me When my throat feels tight I worry I could choke to death
When I cannot keep my mind on a task, I worry that I might be going crazy It scares me when I am unable to keep my mind on a task When I feel spacey or spaced out I worry that I may be mentally ill When my thoughts speed up I worry I might be going crazy When I have trouble thinking clearly I worry something wrong with me When my mind goes blank I worry there is something terribly wrong with me
0.53 0.72 0.67 0.41 0.44
It is important for me not to appear nervous When I tremble I fear what people might think of me I worry that other people will notice my anxiety It scares me when I blush in front of other people When I sweat in social situations I fear people will think negatively of me 0.11 (0.060) I think that it would be horrific for me to faint in public
Note: ASI = Anxiety Sensitivity Index; “G” = the general ASI factor. All loadings in bold were significant.
(0.052) (0.038) (0.046) (0.056) (0.056)
120
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
Fig. 2. Bifactor model of the Korean Version of the Anxiety Sensitivity Index-2 positing a general “anxiety sensitivity” factor and two group factors related to “social concerns” and “physical concerns”.
2.4.6. Unique contribution of group factors We then used structural equation modeling (SEM) to examine the utility of the K-ASI-3 factors with respect to predicting external criteria of anxiety, depression, and negative affect. 3. Results 3.1. Confirmatory bifactor analysis The (3-factor) bifactor model (Fig. 1) fit the K-ASI-3 data well. However, all items from the Cognitive Concerns subdomain loaded significantly only on the general AS factor and not on the Cognitive Concerns factor (see Table 1 for all standardized factor loadings1 ). This led to a modified (2-factor) bifactor model (Fig. 2), whereby
1 ASI-3 #10 (i.e., “When I feel spacey or spaced out I worry that I may be mentally ill”) had an unusually large and negative loading on the Cognitive Concerns domain (i.e., 1.65). Closer inspection revealed the presence of an undefined R-square and a high and negative residual variance (i.e., 2.43) associated with this item. Negative residual variances can occur for items with strong floor effects, which may have been the case for ASI-3 #10 (given its item content).
the items that once comprised the Cognitive Concerns subdomain were assigned only to the general AS factor. This modified (2-factor) bifactor model was associated with good model fit. The x2 difference test between the original (3-factor) bifactor model and modified (2-factor) bifactor model was not significant, x2diff(6) = 20.30, ns. The (3-factor) bifactor model therefore did not fit significantly better than the simplified (2-factor) bifactor model. These results provide support for the 2-factor bifactor model due to it being more parsimonious than the (3-factor) bifactor model and not associated with significantly worse model fit. 3.1.1. Model comparisons Given the support for the (2-factor) bifactor model, we compared it to alternate models (see Table 2). The (2-factor) bifactor model fit significantly better than the competing 1-factor (AS) model [x2diff (12) = 296.60, p < 0.001], 2-factor correlated traits model [x2diff (11) = 185.25, p < 0.001], and the 3-factor correlated traits model [x2iff (9) = 49.93, p < 0.001]. These results therefore provide additional support for the 2-factor bifactor model.
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
121
Table 2 Fit statistics for the confirmatory factor analytic models (N = 288). Model
x2
df
RMSEA
CFI
x2diff
Bifactor model Modified (2-factor) Bifactor model (no cognitive subdomain) 1 Factor (Unidimensional) model 2 Factor (Correlated Traits) model 3 Factor (Correlated Traits) model
264.18 279.88 740.41 564.74 329.81
117 123 135 134 132
0.066 0.066 0.125 0.105 0.072
0.957 0.954 0.823 0.874 0.942
20.30, ns 296.60, p < 0.001 185.25, p < 0.001 49.93, p < 0.001
Note: RMSEA = root mean square error of approximation; CFI = comparative fit index; Correlated traits models yield the same fit indices as second-order (hierarchical) models with one higher-order factor and the same number of lower-order factors (Brown, 2006).
Fig. 3. Structural Equation Model Predicting Anxiety, Depression, and Negative Affect.
3.1.2. Omega hierarchical (general and specific) indices Omega hierarchical indices based on the 2-factor bifactor CFA model were as follows: Omega HierarchicalGeneral = 0.82; Omega HierarchicalSpecific Physical Concerns = 0.33; Omega HierarchicalSpecific Social Concerns = 0.46. These results indicate that the general AS factor yielded the most reliable scores and explained the largest proportion of variance after accounting for
the other factors. These results thus support the presence of a strong general AS factor consistent with a bifactor model. 3.1.3. Explained common variance (ECV) and percentage of uncontaminated correlations (PUC) The ECV value was 0.68 and the PUV value was 0.80. The indices support the notion that the two-factor bifactor model—although
122
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123
multidimensional in nature—may be modeled and treated as unidimensional without producing a high degree of parameter estimation bias. 3.2. Unique contribution of group factors2 All three measurement models were acceptable (Anxiety Model: RMSEA = 0.051, CFI = 0.953; Depression Model: RMSEA = 0.048, CFI = 0.965; Negative Affect Model: RMSEA = 0.071, CFI = 0.881). All factor loadings were also significant in all models. As seen in Fig. 3, various SEM models were used to test the degree to which the factors belonging to the bifactor K-ASI structure (i.e., General AS, Social Concerns, and Physical Concerns) were significantly associated with external criteria (i.e., Anxiety, Depression, and Negative Affect). Results revealed that only the General AS factor significantly predicted Anxiety, Depression and Negative Affect. 4. Discussion The current study was the first to examine the applicability of the bifactor structure underlying the ASI-3 in an East Asian (Korean) sample. The present study found that the (modified 2factor socio-physical) bifactor model fit the present Korean sample well. This bifactor model also fit significantly better than all competing models tested so far in other studies, including the one-, two-, and three-factor (correlated traits) model. Interestingly, the previously assigned Cognitive Concerns items loaded significantly only on the General AS dimension, but not on the “Cognitive Concerns” dimension. This suggests that the “cognitive concerns” items may not account for unique item variability beyond the general AS dimension. The construct of ‘AS Cognitive Concerns’ itself is therefore not well differentiated apart from the AS General Factor (and therefore interpretation of any subscale scores derived from the originally posited ‘AS Cognitive Concerns’ domain would be not advisable). These findings suggest that anxiety sensitivity does not get expressed among Koreans in the same way as in the US, particularly regarding cognitive concerns (of mental health), such as the fear of ‘going crazy’ or ‘mentally ill.’ More research however is needed to verify whether this is indeed the case before discarding the Cognitive Concerns scale from the K-ASI-3. Additional studies would also be needed to examine whether the salience of this Cognitive Concerns dimension strengths over time, particularly as cognitive-related mental health concerns becomes of increased focus in the future in Korea. It is important to note that other studies have also found that the expression of psychopathology varies as a function of culture and ethnicity. For example, Okazaki (1997) found that AsianAmericans’ behaviors were associated with greater avoidance behaviors in social situations compared to White Americans. Pina and Silverman (2004) also more recently found that Hispanic/ Latino youth experience more somatic symptoms when experiencing anxiety than European Americans. The present findings further support this notion that the experience and structure of psychopathology can differ across cultures. Therefore, attention should be given to such possible differences when constructs (such as anxiety sensitivity) are identified and applied in new, crosscultural settings.
2 At a p < 0.05 significance level, with 4 latent variables, 18 observed variables, and a desired statistical power level of 0.80, a minimum sample size of 88 is recommended to detect a medium and large effect, and a minimum sample size of 387 is recommended to detect a small effect (Westland, 2012).
Implications of this new bifactor structure of Anxiety Sensitivity in East Asian (Korean) samples includes the following. First, this was the first study to extend the applicability of the emerging bifactor model of Anxiety Sensitivity to East Asian (Korean) samples. Second, although recent studies are showing that bifactor models well-represent psychological constructs, this study showed that the exact same bifactor structure cannot be assumed to be applicable across different cultures. For example, although the bifactor structure was the most supported model in our Korean sample, the structure of the bifactor model was slightly different than that supported in the US (Ebesutani et al., 2014). Specifically, Ebesutani et al. (2014) found superior fit for a bifactor model with a general AS factor and three content domains (e.g., x2 = 246.49; RMSEA = 0.048; CFI = 0.997), whereas in the current study, superior fit was found for a bifactor model with a general AS factor and two content domains (e.g., x2 = 279.88; RMSEA = 0.066; CFI = 0.954). Consistent with the cultural considerations, the Cognitive Concerns domain did not emerge as a salience dimension above and beyond the general AS factor. Caution should thus be used when directly applying the same bifactor models to Asian samples. Third, and relatedly, when modeling Anxiety Sensitivity in factor analysis and structural equation modeling (for both research and clinical purposes) in East Asian contexts, it may be necessary to omit the Cognitive Concerns specific domain in order to reduce error with respect to measuring Anxiety Sensitivity in Korean individuals. Lastly, this study suggests that the general AS factor is best able to predict external variables such as anxiety, depression and negative affect and thus should be given more weight over the other specific AS domains. This point however is often misunderstood. This does not mean that the other (specific) domains (of Social Concerns and Physical Concerns) and the structure of the bifactor model do not matter; they do matter. They matter a lot. That is, in order to properly model and estimate the general AS factor, AS should be modeled not only in a bifactor structure, but also with the appropriate bifactor structure—in this case, by omitting the Cognitive Concerns specific domain, yet retaining its items to identify and measure the general AS construct among South Korean individuals. Although the present study helped to identify a new bifactor structure and informed ASI scoring related to East Asian (Korean) individuals, there were limitations worth noting. First, only Korean university students were included in the study. We therefore do not know how well the present results generalize to clinical Korean samples or other East Asian cultures. The measures used in the present study were also all self-report instruments, reducing measurement precision relative to a multi-informant, multi-modal assessment approach. Nonetheless, it is hoped that the present study leads to additional research that further clarifies the structure of Anxiety Sensitivity, including informing how to best measure, score, and interpret this important construct to identify and monitor anxiety sensitivity across diverse populations. References Allan, N.P., Albanese, B.J., Short, N.A., Raines, A.M., Schmidt, N.B., 2015a. Support for the general and specific bifactor model factors of anxiety sensitivity. Personal. Individ. Differ. 74, 78–83. Allan, N.P., Macatee, R.J., Norr, A.M., Raines, A.M., Schmidt, N.B., 2015b. Relations between common and specific factors of anxiety sensitivity and distress tolerance and fear, distress, and alcohol and substance use disorders. J. Anxiety Disord. 33, 81–89. Antony, M.M., Bieling, P.J., Cox, B.J., Enns, M.W., Swinson, R.P., 1998. Psychometric properties of the 42-item and 21-item versions of the depression anxiety stress scales in clinical groups and a community sample. Psychol. Assess. 10, 176–181. Asparouhov, T., & Muthén, B., 2006. Robust chi square difference testing with mean and variance adjusted test statistics. Mplus Web Notes: No. 10. May 26, 2006. Retrieved September 26, 2010, from http://www.statmodel.com/download/ webnotes/webnote10.pdf.
C. Ebesutani et al. / Asian Journal of PsychiatryAJP 22 (2016) 116–123 Bonifay, W.E., Reise, S.P., Scheines, R., Meijer, R.R., 2015. When are multidimensional data unidimensional enough for structural equation modeling? An evaluation of the DETECT multidimensionality index. Struct. Equ. Model. 22 (4), 504–516. Brouwer, D., Meijer, R.R., Zevalkink, J., 2013. On the factor structure of the Beck Depression Inventory-II: G is the key. Psychol. Assess. 25 (1), 136–145. Brown, T., 2006. Confirmatory Factor Analysis for Applied Research. Guilford Press, New York, NY. Browne, M., Cudeck, R., 1993. Alternative ways of assessing model fit. In: Bollen, K.A., Long, J.S. (Eds.), Testing Structural Equation Models. Sage, Newbury Park, CA, pp. 136–159. Chavarria, J., Allan, N.P., Boffa, J.W., Albanese, B.J., Zvolensky, M.J., Schmidt, N.B., 2015. Decomposing the relationship between anxiety sensitivity and alcohol use. J. Stud. Alcohol Drugs 76, 957–961. Chen, F.F., West, S.G., Sousa, K.H., 2006. A comparison of bifactor and second-order models of quality of life. Multivar. Behav. Res. 41, 189–225. Ebesutani, C., Smith, A., Bernstein, A., Chorpita, B.F., Higa-McMillan, C., Nakamura, B. J., 2011. A bifactor model of negative affectivity: fear and distress components among younger and older youth. Psychol. Assess. 23, 679–691. Ebesutani, C.K., McLeish, A.C., Luberto, C.M., Young, J., Maack, D.J., 2014. A bifactor model of anxiety sensitivity: analysis of the Anxiety Sensitivity Index-3. J. Psychopathol. Behav. Assess. 36 (September (3)), 452–464. Feldner, M., Zvolensky, M., Schmidt, N., Smith, R., 2008. A prospective test of anxiety sensitivity as a moderator of the relation between gender and posttraumatic symptom maintenance among anxiety sensitive young adults. Depress. Anxiety 25, 190–199. Hayward, C., Killen, J., Kraemer, H., Taylor, C., 2000. Predictors of panic attacks in adolescents. J. Am. Acad. Child Adolesc. Psychiatry 29, 207–214. Holgado-Tello, F., Chacón-Moscoso, S., Barbero-García, I., Vila-Abad, E., 2010. Polychoric versus Pearson correlations in exploratory and confirmatory factor analysis of ordinal variables. Qual. Quant. 44 (1), 153–166. Holzinger, K.J., Swineford, R., 1937. The bifactor method. Psychometrika 2, 41–54. Hu, L.T., Bentler, P.M., 1999. Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives. Struct. Equ. Model. 6, 1– 55. Kemper, C.J., Lutz, J., Bahr, T., Ruddel, H., Hock, M., 2012. Construct validity of the Anxiety Sensitivity Index-3 in clinical samples. Assessment 19, 89–100. Kim, K., Li, D., Kim, D., 1999. Depressive symptoms in Koreans, Korean-Chinese and Chinese: a transcultural study. Transcult. Psychiatry 36 (3), 303–316. Lee, H., Kim, E., Lee, M., 2003. A validation study of Korean Positive and Negative Affect Schedule: the PANAS scales. Korean J. Clin. Psychol. 22 (4), 935–946. Lee, S., Juon, H.S., Martinez, G., Hsu, C.E., Robinson, E.S., Bawa, J., Ma, G.X., 2009. Model minority at risk: expressed needs of mental health by Asian American young adults. J. Community Health 34 (2), 144–152. Li, W., Zinbarg, R.E., 2007. Anxiety sensitivity and panic attacks: a 1-year longitudinal study. Behav. Modif. 31, 145–161. Lim, Y.J., Kim, J.H., 2012. Korean Anxiety Sensitivity Index-3: it’s factor structure, reliability, and validity in non-clinical samples. Psychiatry Investig. 9, 45–53. Little, R.J.A., Rubin, D.B., 1987. Statistical Analysis with Missing Data. Wiley, New York. Lovibond, S.H., Lovibond, P.F., 1995. Manual for the Depression Anxiety Stress Scales, 2nd ed. Psychology Foundation of Australia, Sydney, Australia.
123
Marshall, G.N., Miles, J.N., Stewart, S.H., 2010. Anxiety sensitivity and PTSD symptom severity are reciprocally related: evidence from a longitudinal study of physical trauma survivors. J. Abnorm. Psychol. 119, 145–150. McNally, R., 2002. Anxiety sensitivity and panic disorder. Biol. Psychiatry 52, 938– 946. Muthén, L.K., Muthén, B.O., 2010. Mplus User’s Guide, 4th ed. Author, Los Angeles, CA. Okazaki, S., 1997. Sources of ethnic differences between Asian American and white American college students on measures of depression and social anxiety. J. Abnorm. Psychol. 106 (1), 52. Olsson, U., 1979. Maximum likelihood estimation of the polychoric correlation coefficient. Psychometrika 44 (4), 443–460. Osman, A., Gutierrez, P., Smith, K., Fang, Q., Lozano, G., Devine, A., 2010. The Anxiety Sensitivity Index-3: analyses of dimensions, reliability estimates, and correlates in nonclinical samples. J. Personal. Assess. 92, 45–52. Pina, A.A., Silverman, W.K., 2004. Clinical phenomenology, somatic symptoms, and distress in Hispanic/Latino and European American youths with anxiety disorders. J. Clin. Child Adolesc. Psychol. 33 (2), 227–236. Psychology Foundation of Australia, 2013. DASS. Unpublished instrument. Retrieved March 23, 2013, from http://www2.psy.unsw.edu.au/groups/dass. Reise, S., Morizot, J., Hays, R., 2007. The role of the bifactor model in resolving dimensionality issues in health outcomes measures. Qual. Life Res. 16, 19–31. Reise, S., Moore, T., Haviland, M., 2010. Bifactor models and rotations: exploring the extent to which multidimensional data yield univocal scale scores. J. Personal. Assess. 92, 544–559. Reise, S.P., Scheines, R., Widaman, K.F., Haviland, M.G., 2013. Multidimensionality and structural coefficient bias in structural equation modeling a bifactor perspective. Educ. Psychol. Meas. 73 (1), 5–26. Reiss, S., McNally, R., 1985. Expectancy model of fear. In: Reiss, S., Bootzin, R.R. (Eds.), Theoretical Issues in Behavior Therapy. Academic Press, New York, NY, pp. 107– 122. Schmidt, N., Zvolensky, M., Maner, J., 2006. Anxiety sensitivity: prospective prediction of panic attacks and Axis I psychopathology. J. Psychiatr. Res. 40, 691–699. Taylor, S., Koch, W.J., Woody, S., McLean, P., 1996. Anxiety sensitivity and depression: how are they related? J. Abnorm. Psychol. 105 (3), 474. Taylor, S., Zvolensky, M.J., Cox, B.J., Deacon, B., Heimberg, R.G., Ledley, R.G., Cardenas, S.J., 2007. Robust dimensions of anxiety sensitivity: development and initial validation of the Anxiety Sensitivity Index-3. Psychol. Assess. 19, 176–188. Westland, J.C., 2012. Erratum to Lower bounds on sample size in structural equation modeling [Electron. Commerce Res. Appl. 9 (6) (2010) 476–487]. Electron. Commer. Res. Appl. 11 (4), 445. Wheaton, M.G., Deacon, B.J., McGrath, P.B., Berman, N.C., Abramowitz, J.S., 2012. Dimensions of anxiety sensitivity in the anxiety disorders: evaluation of the ASI-3. J. Anxiety Disord. 26, 401–408. Yabuuchi, A., 2004. Face in Chinese, Japanese, and US American cultures. J. Asian Pac. Commun. 14 (2), 261–297. Yum, J.O., 1988. The impact of Confucianism on interpersonal relationships and communication patterns in East Asia. Commun. Monogr. 55 (4), 374–388. Zvolensky, M.J., Arrindell, W.A., Taylor, S., Bouvard, M., Cox, B.J., Stewart, S.H., Eifert, G.H., 2003. Anxiety sensitivity in six countries. Behav. Res. Ther. 7, 841–859.