Measurement Invariance of the Strength and Difficulties Questionnaire in Early Childhood

• In Event: How do we account for missing and unobserved data?: Diminishing the impacts of hidden bias through novel statistical approaches

Abstract

Research indicates noteworthy differences in rates of internalizing (i.e., withdrawn behaviors, worry, and somatic symptoms) and externalizing (i.e., dysregulation, high activity, and inattention) symptoms in children across age, race, sex, and socio-economic status (SES). Boys (Fernandez-Castelao & Kroner-Herwig, 2014), Black children (Caldwell et al., 2016), and children from families with a lower SES (Peverill et al., 2021) have been rated as having more externalizing problems. Girls (Rosenfield & Mouson, 2012) and older children (Solmi et al., 2021) have been identified as having higher internalizing symptoms. Critically, the validity of these findings rests on the assumption that the measures “mean the same thing” across groups and developmental time (i.e., Measurement Invariance; MI). Without assuring MI, results could be biased because they may represent differences in measurement and not true differences in the latent construct across groups.
A widely used tool to measure internalizing and externalizing symptoms is the Strengths and Difficulties Questionnaire (SDQ; Goodman et al., 2000). Only a few papers to date have evaluated MI of the SDQ longitudinally and they generally found MI across age for school-aged children, gender, and parental education (e.g., Croft et al., 2015; Dahlberg et al., 2019; Murray et al., 2021). However, no publications to our knowledge have examined MI across diverse economic and racial groups in very young children. The aim of the present study was to test MI on the internalizing and externalizing scales of SDQ across child age, race, sex, and SES.
Data were drawn from the Family Life Project (FLP; N=1,292)—a study of children in rural low-income contexts (Vernon-Feagans et al., 2013). Parents reported sociodemographic information (Table 1) and child symptoms via the SDQ when children were 35-, 58-, and 90 months old. We leveraged moderated nonlinear factor analysis (MNLFA; Gottfredson et al., 2019) to evaluate MI of the SDQ. This method estimates the “true effect” of demographic variables on the latent factor (i.e., eta) and simultaneously adjusts for Differential Item Functioning (DIF). DIF occurs when there are any mean (i.e., intercept) differences of an item and/or any differences in how it loads onto the latent factor across demographic groups. If items are functioning differently across groups then these items may not be tapping the same construct across groups and interpreting any group-level differences would be like comparing apples to oranges.
Results indicated that there were several items that were demonstrating DIF. The internalizing symptoms model (Figure 1a) indicated that two items demonstrated DIF by child age and race. This included Black children being rated as seeming more worried as they aged compared to White children. The externalizing symptoms model (Figure 1b) had four items that demonstrated DIF across child age, race, sex, and SES. This included an item about children getting along well with others contributing less information to the latent factor for higher SES families. In sum, testing for and adjusting for DIF via the MNLFA approach greatly improves our confidence in the meaning of latent SDQ scores and improves our ability to understand any true group differences.

Authors

• Alyssa R Palmer, University of Minnesota - Twin Cities
  Presenting Author

• Ann S. Masten, University of Minnesota - Twin Cities

• Daniel Berry, University of Minnesota - Twin Cities