Confirmatory Factor Analysis of the Social Responsiveness Scale

Background: The Social Responsiveness Scale (SRS; Constantino et al, 2003) is a questionnaire that measures social impairment and symptoms of autism spectrum disorders (ASD). Several studies have examined the psychometric properties of the SRS and have reported positive results (e.g., Constantino et al, 2007; Constantino & Gruber, 2005; Constantino et al, 2004; Pine et al, 2006). Examinations of the construct validity of the SRS have included several studies that have used principal components analysis to determine the underlying structure of the measure (e.g., Bolte et al., 2008; Constantino et al., 2000). Constantino et al. (2000) reported that a single factor solution best described the data. These results were corroborated in a study by Bolte et al (2008) who reported a single factor solution of ASD symptomology in both normative and clinical samples. However, other investigators have suggested that the SRS may measure overall behavioral impairment rather that ASD-related symptomology (Hus et al., 2013). 

Objectives: The purpose of the current study was to assess the construct validity of the SRS by testing the single factor model of the measure in a large, representative sample of children with ASD. To our knowledge, this is the first independent study of the factor structure of the SRS.  

Methods:  Participants in this study included 395 individuals (82% males) between the ages of 29 and 272 months (mean=102.5, SD=48.77) who participated in the Simons Simplex Collection and Autism Consortium at Boston Children’s Hospital. All individuals were diagnosed with an ASD. CFA was used to test a single factor model of the SRS. Model fit was estimated using the Tucker Lewis Index (TLI), Comparative Fit Index (CFI), and Root Mean Square Error of Approximation (RMSEA). As suggested by Hua and Bentlerb (1999), the following cutoffs were used for good fit:  >.90, >.90 and <.06, respectively.

Results: The CFA indicated an RMSEA of .068, which is slightly higher than the cutoff suggested by Hua et al. The CFI was .58, and the TLI was .57. Both indices were lower than the proposed cutoff of >.90 to suggest good fit. Factor loadings for the one-factor solution ranged from .06 to .71. When a factor loading cutoff of .3 was used, 10 items (15% of the total items) did not load onto the one factor solution. 

Conclusions:  Our results indicate that a one-factor solution may not be optimal to describe the structure of the SRS. All fit indices were below proposed cutoffs for good fit, and 15% of the SRS items did not load onto a one factor model.  These preliminary results suggest that further investigation of the construct validity of the SRS is necessary. Future analyses will include CFA of 2- and 3-factor solutions. We will test a 2 factor solution of social/communication and restricted and repetitive behaviors, and a 3 factor solution of social, communication, and restricted and repetitive behaviors, which replicate ASD symptomology as classified by the DSM-5 and DSM-IV, respectively.