The Hierarchical Factor Structure of the Spanish Version of Depression Anxiety and Stress Scale -21 (DASS-21)

Full reference: Ruiz, F. J., García Martín, M. B., Suárez Falcón, J. C., & Odriozola González, P. (2017). The hierarchical factor structure of the Spanish version of Depression Anxiety and Stress Scale -21. International Journal of Psychology & Psychological Therapy, 17(1), 97-105.

Study type: Hierarchical factor structure analysis

Background and objectives

The Depression Anxiety and Stress Scale-21 (DASS-21) represents one of the most widely used self-report instruments for assessing emotional symptoms in clinical and research settings. The DASS-21 was originally designed to maximize discrimination between symptoms of depression and anxiety, recognizing that these disorders present high comorbidity (approximately 50-70% of cases). Despite its widespread use, considerable controversy has existed regarding its factor structure, particularly concerning the nature of the stress dimension and its relationship to the depression and anxiety dimensions.

Historically, the DASS was created with three distinct scales: Depression (measuring low energy, dysphoria, hopelessness, sadness, and apathy), Anxiety (measuring physiological activation and subjective anxiety experience), and Stress (measuring symptoms related to emotional lability, tension, irritability, nervousness, and impatience). Initial studies by Lovibond and Lovibond (1995), upon which the DASS-21 is based, identified a three-factor structure of differentiated factors. However, subsequent studies have yielded mixed results, with some researchers arguing for a two-factor structure (depression and stress loading on the same factor, with anxiety as a separate factor) or a single general factor structure.

Method

Participants

The study utilized four independent samples with a total of 2980 participants. The first sample consisted of 511 Spanish university students (age range 18-68 years, M = 26.74, SD = 10.31) from four Spanish universities. Forty-four percent were studying Psychology, while others were studying Speech Therapy, Law, and Physics. Sixty-one percent were women. Of the overall sample, approximately 3.7% had received psychological or psychiatric treatment, though only 4.3% were in active treatment at the time of the study. The second sample included 762 Colombian university students (age range 18-63 years, M = 21.16, SD = 3.76) from a private university in Bogotá. Forty-six percent of this sample were studying Psychology, while 54% were distributed among other disciplines including Law, Engineering, Philosophy, Communication, Business, Medicine, and Theology. Fifty-six percent were women. Twenty-six percent of this sample had received psychological or psychiatric treatment at some time, though only 2.9% were in active treatment. The third sample was a pool of 813 participants ranging from 18 to 82 years old (M = 34.74, SD = 10.87) recruited through an anonymous survey distributed through social media. Relative educational level was varied, including 34.5% with primary or mid-level education, 42.7% with university or undergraduate education, and 22.3% in postgraduate studies or with postgraduate degrees. Seventy-one percent were women. Only 16.8% of this sample were currently in treatment. The fourth sample consisted of 894 participants (67.4% female) with ages between 18 and 88 years (M = 29.16, SD = 10.13) recruited again through social media. Twenty-one point three percent had primary or mid-level education, 62.5% were university students or had completed university studies, and 16.2% were pursuing or had completed postgraduate degrees. Nine point four percent were currently in treatment, and 5.7% reported psychotropic medication use.

Instrument(s) under study

The Depression, Anxiety, and Stress Scale-21 (DASS-21) was employed, a 4-point Likert-type scale consisting of 21 items describing negative emotional states. Each item is scored on a scale ranging from 0 ("did not apply to me at all") to 3 ("applied to me very much, or most of the time"). The DASS-21 contains three subscales: Depression (7 items), Anxiety (7 items), and Stress (7 items). The Spanish version of the DASS-21 was adapted by Daza et al. (2002), demonstrating good psychometric properties in Spanish university student samples (Fonseca Pedrero et al., 2010).

Data analysis

Participants in Samples 1 and 2 completed the DASS-21 questionnaire in the context of regular classrooms with informed consent. Participants in Samples 3 and 4 responded to an anonymous survey distributed through social media. Prior to conducting factor analyses, all datasets were examined for missing values. Only 13 DASS-21 values were missing across the entire sample, which were imputed using the matching response pattern method (LISREL version 8.71). Responses from 10 Spanish university students were eliminated due to null response patterns.

Confirmatory factor analyses (CFA) were conducted to compare five alternative factor models of the DASS-21. The first model was a single-factor model. The second was a two-factor correlated model with depression and stress loading on the same factor and anxiety as a separate factor. The third model was a two-factor correlated model with anxiety and stress loading on the same factor. The fourth model was a three-factor correlated model (Depression, Anxiety, Stress). The fifth model was the previous model plus a second-order general factor. An unweighted least squares (ULS) estimation methodology was employed because the DASS-21 is measured on an ordinal scale.

Goodness-of-fit indices were calculated, including the root mean square error of approximation (RMSEA), comparative fit index (CFI), non-normed fit index (NNFI), expected cross-validation index (ECVI), and standardized root mean square residual (SRMR). Gignac (2007) recommendations for interpretation were followed, with RMSEA values below .05 indicating very good fit, CFI and NNFI values above .90 indicating good fit, and SRMR values below .08 indicating acceptable fit.

The Schmid-Leiman transformation was conducted as an alternative to nested factor analysis. This statistical transformation performs a secondary exploratory factor analysis (EFA) using the latent factor correlations obtained from the previous analysis and facilitates interpretation of primary factors (items) relative to higher-order factors by computing direct relations between primary variables and second-order factors. Additionally, the proportion of variance in the general factor accounted for by the general factor was calculated.

Additional confirmatory factor analyses were conducted to test measurement invariance across countries (Spain vs. Colombia) and sample type (undergraduate vs. online). A multiple group baseline method was employed, allowing unstandardized factor loadings to vary across countries and sample types.

Cronbach's alphas were calculated across all samples to explore internal consistency of the DASS-21.

Results

Factor Model Fit

Table 1 presents the goodness-of-fit indices for the five factor models considered. Overall, results were very similar across all three cases. The single-factor model showed acceptable fit, but was inferior to both two-factor correlated models. However, the three-factor correlated model demonstrated the best fit to the data. As expected, the fit of the three-factor correlated model plus a general factor was identical to the model with only three correlated factors.

Explained Variance and Hierarchical Structure

Table 2 presents the variance explained by the general factor in the model with three correlated first-order factors according to the Schmid-Leiman transformation. In all cases, the general factor accounted for more than 70% of the variance, a proportion clearly within the range considered indicative of the presence of a general factor (40%-50%; Gorsuich, 1983). Additionally, all items appeared to represent the general factor because they showed loadings above .30 (Tabachnick & Fidell, 2007).

Internal Consistency

Table 4 presents Cronbach's alpha and descriptive data for each sample. Cronbach's alphas for the total DASS-21 scale were excellent, with values ranging from .92 to .95. Alphas for the subscales were good, with the Depression subscale showing higher values (.86 to .92) than the Anxiety subscale (.80 to .87) and Stress subscale (.80 to .86).

Measurement Invariance

Table 3 presents fit indices for measurement invariance tests of the hierarchical model with three correlated factors. As can be seen, the multiple group baseline model allowed the 21 unstandardized factor loadings to vary across countries and sample types. In this model, no significant differences in fit were found when equality constraints were placed on factor loadings (Model 2), suggesting there was no significant decrement in goodness of fit, indicating that measurements were invariant across countries (Spain vs. Colombia) and sample type (undergraduate vs. online). In both cases, all criteria recommended by Cheung and Rensvold (2002) and Chen (2007) were met, except the chi-square test, which were statistically significant across countries, χ2(21) = 85.1, p <.01, and across sample types, χ2(21) = 66.9, p <.01.

Discussion and conclusions

The results of this study provide strong evidence that the Spanish version of the DASS-21 possesses a consistent hierarchical factor structure of one general factor (Emotional Symptoms) and three correlated first-order factors (Depression, Anxiety, Stress) in Spanish-speaking contexts. The presence of a general factor provides a theoretical basis for using the total DASS-21 score as a general measure of emotional symptoms, while the presence of a second-order factor provides flexibility to both researchers and clinicians, allowing them to choose between separate responses in the three DASS-21 dimensions or summarizing them in an overall measure. Although some limitations have been identified (such as the fact that the functioning of the DASS-21 was tested only in nonclinical samples), the study represents a significant advance in understanding the psychometric properties of the DASS-21 in Spanish-speaking contexts.

Significance and contribution

This study contributes to cross-cultural validation of the DASS-21 by demonstrating that the instrument possesses a consistent hierarchical factor structure across Spanish-speaking contexts using Spanish and Colombian samples. By clarifying that the hierarchical structure provides flexibility for analyzing both specific dimensions of emotional symptoms and general emotional symptom scores, the work extends the theoretical and practical foundation for DASS-21 use in research and clinical assessment in Spanish-speaking populations. Findings regarding measurement invariance across countries and contexts suggest the instrument is transculturally relevant.