- About Us
- What We Do
- Our Findings
- News & Events
- Contact Us
Prady S.; Croudace T.; Pickett K.; Petherick E.; Wright J.
Background: Goodman’s Strengths and Difficulties Questionnaire (SDQ) is a short, multi-dimensional screening instrument for detecting psychiatric disorders in children and adolescents. A 5-subscale structure (4 problem oriented; Emotional symptoms, Peer problems, Hyperactivity/inattention, Conduct problems, and 1 positive; Prosocial behaviour) is typical and has been empirically explored in several studies of primary school age children. We aimed to expand upon its psychometric properties using categorical data factor analysis procedures (multidimensional item response theory in Mplus), applied to data from parent ratings of preschoolers. The structure may be influenced by; (1) treatment of ratings as ordinal, (2) different properties of items comprising the 5 factors, and (3) the contribution of 5 positively worded questions in the otherwise negatively worded problem subscales (accommodated here by an uncorrelated methods factor).
Objectives: To examine the latent structure of parent reported SDQ ratings in children aged 3, from a subsample of Born in Bradford, a community birth cohort study in an economically deprived city in northern England.
Methods: We analysed data compiled on singleton children from 943 questionnaires completed in English and 232 in Urdu using confirmatory factor analysis with a weighted least squares estimator. We tested 7 models: (1) 3 factors comprising Internalising, Externalising and Prosocial subscales; (2) 4 factors comprising Emotional, Peer, Conduct and Hyperactivity subscales; (3) 5 factors comprising Emotional, Peer, Conduct, Hyperactivity and Prosocial subscales; (4) 6 factors comprising model 3 plus a methods factor; (5) a 2nd-order model comprising Emotional, Peer, Conduct, Hyperactivity and Prosocial subscales forming a Total difficulties scale; (6) a bifactor model of the 5 subscales and (7) a bifactor model of the 5 subscales plus methods factor. Model fit was assessed using acceptable thresholds for CFI, TLI and RMSEA statistics.
Results: Models 1-4 had poor fit. For the English cases, marginally acceptable fit was observed with the methods factor (model 5). The 2 bifactor models had good fit with best fit observed for model 7 (bifactor analysis of 5 subscales plus methods factor). Questionnaires in Urdu demonstrated less good fit. Regressing English bifactor coefficients on sex, adjusted for age and parentrated difficulties, there were associations between boys and worse total (p = 0.03), Hyperactivity/ inattention (p = 0.001) and Conduct (p = 0.009) scores.
Conclusion: In this pre-school population, bifactor analysis suggests that the SDQ may usefully characterize a general severity continuum as well as a multi-dimensional screening profile of scores, after accounting for the variance in the underlying methods factor.