Equivalence of the Short Form of the Reynolds Adolescent Depression Scale across Groups

Abstract

The present study examines the equivalence of the short-form version of the Reynolds Adolescent Depression Scale (RADS-SF) for measuring depression in adolescents across gender, age, and ethnic groups. A sample of 8,692 randomly selected New Zealand secondary school students participated in the Youth'07 Health and Wellbeing Survey that included the RADS-SF. The reliability was assessed using Cronbach's alpha and item-total correlations. The validity was assessed using multigroup confirmatory factor analysis, and correlation to other questions in the survey considered likely to be associated with depression. The RADS-SF scores ranged from 10 to 40 (Mdn = 18), with a mean score of 19.14 (SD = 6.19) and Cronbach's alpha of .88. Configural, metric, and scalar equivalence was supported across gender, age, and ethnic groups (New Zealand European, Māori, Pacific, Asian, and Other), with all tested models having good fit to the data. The correlations between the RADS-SF and other variables such as suicidal ideation and well-being were also equivalent across groups. The RADS-SF was found to be equivalent in measuring depression across age, ethnic groups, and gender in a large population of New Zealand adolescents.

Acknowledgments

The Youth'07 study was funded by the Health Research Council of New Zealand, the Department of Labour, the Families Commission, the Accident Compensation Corporation of New Zealand, Sport and Recreation New Zealand, the Alcohol Advisory Council of New Zealand, and the Ministries of Youth Development, Justice and Health. We thank these organisations as well as all the students, staff, schools, project team members and advisory and steering groups who participated and supported the Youth'07 project. A. Szabo and T.L. Milfont contributed equally to the writing of this article, and devised the research questions, conducted the statistical analyses, and drafted the article. E.M. Robinson provided statistical advice and oversight. All authors contributed to the design of the overall Youth'07 survey (except A. Szabo), to the interpretation of analyses, and to the critical revision of the article.

Notes

¹Treating the data as continuous yielded similar results supporting configural, metric and scalar invariance, but some noninvariant items could be identified. These results are available upon request.

Note: Dashes indicate the statistics were not estimated. RADS–SF = short-form version of the Reynolds Adolescent Depression Scale; NZ = New Zealand.

Note: ITC = item-total correlations.

Note: RMSEA = root mean square error of approximation; 90% CI = confidence interval for RMSEA; CFI = comparative fit index; TLI = Tucker-Lewis index.

Note: Dashes indicate the particular fit is not applicable for the tested model. RMSEA = root mean square error of approximation; 90% CI = confidence interval for RMSEA; CFI = comparative fit index; TLI = Tucker-Lewis index; GH = gamma hat; NCI = noncentrality index.

Note: RADS-SF = short-form version of the Reynolds Adolescent Depression Scale; WHO-5 = World Health Organization-Five.

**p < .01, two-tailed.

Note: Dashes indicate the particular fit is not applicable for the tested model. RMSEA = root mean square error of approximation; 90% CI = confidence interval for RMSEA; CFI = comparative fit index; TLI = Tucker-Lewis index; GH = gamma hat; NCI = noncentrality index.

*Measurement weights are constrained to be equal across the particular groups.

²Although this clustering is not the focus of the present research, we ran additional analyses to rule out the possibility of a clustering effect on the psychometric properties of the RADS-SF. According to Muthén and Satorra (1995) and L. Muthén (personal communication, March 6, 2013), when design effects, which are based on intraclass correlations, are lower than 2, clustering does not need to be taken into account. The intraclass correlations of the RADS-SF items were very small, ranging between .02 and .09. This suggests that clustering had no marked effect on the RADS-SF items in our sample.