Introduction
The high rates of comorbidity among emotional disorders (i.e. mood and anxiety disorders) has led to a greater emphasis on examining factors that may be common to both domains. Transdiagnostic factors are shared between multiple forms of psychopathology, contributing to specific disorders as well as comorbidity between disorders. Increased empirical understanding of transdiagnostic factors is both theoretically and clinically important as new therapeutic interventions are developed to address comorbidity and these common variables (Drost et al., Reference Drost, van der Does, van Hemert, Penninx and Spinhoven2014). Although worry has typically been associated with anxiety, recent theories suggest that it may be related to emotional disorders in general (Drost et al., Reference Drost, van der Does, van Hemert, Penninx and Spinhoven2014) and may have relatively equal associations with anxiety, depression and stress (Olatunji et al., Reference Olatunji, Broman-Fulks, Bergman, Green and Zlomke2010). Psychometrically addressing this issue involves simultaneously estimating the relations of Worry to both general emotional distress (GED) and specific distress factors that are partialled on the GED. Whereas the Depression Anxiety Stress Scale (DASS) has been traditionally analysed using a three-factor structure for specific distress, the DASS bifactor model (Gomez, Reference Gomez2013) permits such an evaluation.
The present study sought to replicate the bifactor model of the DASS within a US adult sample utilizing confirmatory factor analysis (CFA). Second, the study evaluated the bifactor scale reliabilities to determine the portion of reliable variance accounted for by the general and domain-specific factors. Finally, using structural equation modelling (SEM) the present study examined the differential relations of these DASS factors to Worry.
Method
Participants
Participants were 456 English speaking US adults (mean age 35.9 years, SD age = 12.7) of 18 years of age or older. The sample consisted of 279 females (61.2%), 177 males (38.8%), with 79.2% identifying as Caucasian, 6.4% African-American, 5.9% Hispanic, 5.0% Asian, and 3.5% other. In addition, 37.3% of participants reported being diagnosed with or treated for anxiety, worry, or panic.
Procedure
Ethical approval for procedures was obtained from the Texas Tech University (TTU) Human Research Protection Program (HRPP; no. 505082). The authors assert that all procedures contributing to this work complied with ethical standards of the TTU HRPP and the Helsinki Declaration of 1975, and its most recent version. Participants were recruited using Mechanical Turk which guided interested participants to the online survey website where, after providing consent, they completed a battery of distress measures and were monetarily compensated for their time (US$.30).
Materials
Depression Anxiety Stress Scale (DASS)
The DASS is a 42-item public-domain self-report questionnaire measuring depression, anxiety and stress. In the present study, subscale descriptives were: Depression mean = 10.84, SD = 10.85, α = 0.96; Anxiety mean = 8.29, SD = 8.43, α = 0.93; Stress mean = 14.45, SD = 10.37, α = 0.95.
Penn State Worry Questionnaire (PSWQ)
The PSWQ is a 16-item public-domain self-report measure of severity and controllability of worry. In the present study, mean = 50.65, SD = 16.27, α = 0.95.
Results
All analyses were conducted within MPlus version 7.11 (Muthén and Muthén, Reference Muthén and Muthén2012) using weighted least squares means and variances method (WLSMV) (please refer to online full report for more details of statistical analyses and results).
Bifactor replication
The bifactor model of the DASS (Gomez, Reference Gomez2013) was replicated utilizing CFA. Due to the ordinal nature of the DASS response scales, items were modelled as categorical indicators. Global fit indices suggested good model fit, χ2 (777) = 1642.67, p = < .000, RMSEA = 0.05 [0.046, 0.053], CFI = 0.98 (Hu and Bentler, Reference Hu and Bentler1999). Furthermore, only items 9 (λanxiety = 0.10, ‘Anxious’) and 30 (λanxiety = 0.03, ‘Fear of being thrown’) demonstrated non-significant loadings on their respective unique factor after accounting for the GED factor. On average, items from the Depression, Anxiety and Stress scales made roughly comparable contributions to the GED factor.
Furthermore, while the traditional oblique three-factor model (M = 3) with Depression, Anxiety and Stress demonstrated adequate fit χ2 (819) = 3828.48, p = < .000, RMSEA = 0.09, CFI = 0.92, the DIFFTEST suggested that the bifactor model demonstrated significantly better fit (χ2diff (42) = 944.87, p < 0.01). Overall, this provided initial support for the generalizability and replicability of the DASS bifactor model to an unselected adult sample.
Factor reliability
Coefficient omega was calculated to provide a model-based estimate of internal consistency reliability that is more appropriate for congeneric tests than coefficient alpha (Rodriguez et al., Reference Rodriguez, Reise and Haviland2015). Omega was found to be high for all scales with estimates of 0.99, 0.98, 0.96 and 0.97 for the GED, Depression, Anxiety and Stress factors, respectively. Furthermore, omega hierarchical was high (0.90), suggesting that the GED factor accounted for a large portion of the reliable variance in total scores. In contrast, omega hierarchical subscale was low to moderate for the residualized factors Depression (0.28), Anxiety (0.15) and Stress (0.19). Overall, whereas the GED factor accounted for a majority of the variance in the item scores, the Depression, Anxiety and Stress factors still accounted for a modest proportion of the residual variance in items.
Relation with Worry
The relations of the latent constructs of Worry to GED, Depression, Anxiety and Stress were estimated utilizing SEM and results from the DASS bifactor model. In the initial model, GED was modelled as an exogenous latent variable predicting Worry. This initial model demonstrated good fit χ2 (1552) = 2641.01, p = < .000, RMSEA = 0.04 [0.037, 0.042], CFI = 0.96. Adding paths from Worry to both Depression (χ2diff (1) = ‒9.396, p < 0.01) and Anxiety (χ2diff (1) = ‒5.194, p = 0.02) but not to Stress (χ2diff (1) = 0.80, p = 0.37) significantly improved fit. However, constraining the partial paths from Depression and Anxiety to Worry to be equal significantly decreased model fit (χ2diff (1) = 5.61, p = 0.02).
Thus the final model included GED (pr = 0.81, p < 0.001), Depression (pr = ‒0.13, p = 0.002) and Anxiety (pr = ‒0.19, p = 0.002), but not Stress, as significant incremental predictors of Worry. The final model demonstrated good fit, χ2 (1550) = 2622.72, p = < .000, RMSEA = 0.04 [0.036, 0.042], CFI = 0.96 and is shown in Fig. 1. Similarly, bivariate analyses indicated GED (r = 0.84, p < 0.001), Depression (r = ‒0.14, p = 0.002) and Anxiety (r = ‒0.20, p = 0.002) but not Stress (r = 0.01, p = 0.91) significantly predicted Worry. This suggests the relation of GED with Worry was substantially stronger than that of either Depression or Anxiety.
Figure 1. The structural model of relations between Emotional Distress, Depression, Anxiety, Stress and Worry. r = standardized bivariate regression coefficients; pr = standardized partial regression coefficients (i.e. partial correlations); D = DASS Depression item; A = DASS Anxiety item; S = DASS Stress item; W = PSWQ item. **p < 0.01, ***p < 0.001.
Whereas researchers may wish to examine worry as a predictor of distress, this would have led to multiple, bivariate relations that may overestimate the variance accounted for by each distress construct, both general and unique. By utilizing Worry as the dependent variable, the paths were estimated simultaneously and therefore each represents the partial relation to Worry after accounting for the other types of distress. For example, the final path from Anxiety to Worry represents the unique variance in Worry accounted for by Anxiety after controlling for GED and Depression; thus providing a more conservative estimate of the relation that is disorder specific. Furthermore, while the data provides a means of examining the inter-individual relations between these constructs at a single time point, directionality cannot be inferred from cross-sectional data. Thus the direction of the estimation becomes less important.
Discussion
Results supported and replicated the bifactor model for the DASS, promoting its generalizability in a US community population. Reliability estimates for the GED factor and the three domain-specific factors (i.e. Depression, Anxiety and Stress) were all strong (>0.96). Furthermore, the omega hierarchical and omega hierarchical subscale estimates indicated that the GED factor accounted for a substantial portion of the reliable variance in total scores whereas the Depression, Anxiety and Stress factors only accounted for modest portions of the reliable variance. This result is not surprising given that these three specific factors are estimated using residualized scores based on fewer items than the GED factor (Rodriguez et al., Reference Rodriguez, Reise and Haviland2015). Additionally, the item loadings for the Bifactor and M = 3 Depression factors demonstrated a strong correlation (r = .94), suggesting that these two factors are assessing similar configurations of item loadings. In contrast, the M = 3 Anxiety and Stress item loadings are more strongly correlated with their item loadings on the Bifactor GED factor than the specific factors (see Table 1 of full online report.) These results are consistent with the lower proportion of reliable variance in the bifactor Anxiety and Stress specific scales (see above). Although this suggests that the GED factor and its relations may be reliably interpreted, it also suggests exercising caution when interpreting the three domain-specific scores.
This study also examined the relations between worry and the common (i.e. GED) and unique (i.e. Depression, Anxiety and Stress) factors of the bifactor DASS model. General Emotional Distress was a large, positive incremental predictor of Worry while Depression and Anxiety were small, negative incremental predictors. That is, worry is primarily correlated with non-specific emotional distress. These results further support the notion that worry is a transdiagnostic variable and not necessarily domain specific. The reason for the negative partial relation of worry with the two specific scales is unclear and may reflect noise or error on these residualized factors; however, the bivariate relations disqualify suppressive effects as an explanation. Interestingly, this result seems to further support the transdiagnostic role of worry.
Although our results support the replicability of the bifactor model proposed by Gomez (Reference Gomez2013), there may be alternative models with comparable fit. Another potential limitation is the modest omega hierarchical subscale coefficients of the Depression, Anxiety and Stress factors. These estimates limit the confidence in a specific conclusion about the relationship of these subscales with other variables. Of course, due to the cross-sectional nature of the data, directionality of the relations cannot be determined. Finally, although participants included in this study had self-reported symptoms of depression, anxiety and worry, future studies should examine the generalizability of the bifactor model of the DASS in clinical populations. Replication within a sample of individuals with diagnosed emotional disorders, including comorbid anxiety and depression, may be particularly enlightening.
Conclusions
The results of this study provide initial support for the validity of the bifactor model of the DASS within a US community sample and provide estimates of the reliability of the resulting general and domain-specific factors. Furthermore, this is the first study to examine relations between these factors and a theoretically related construct, worry. Although low omega hierarchical subscale estimates prevented subscale specific inferences from being made, overall findings further support the idea that worry is a transdiagnostic factor shared across specific dimensions of distress and perhaps different emotional disorders and may partly explain their comorbidity.
Conflicts of interest: All authors declared no conflicts of interest.
Financial support: This research received no specific grant or external funding from any funding agency, commercial or not-for-profit sectors.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S1352465817000777
Comments
No Comments have been published for this article.