The Multidimensional Factor Structure of Preschooler’s Emotional Knowledge and Measurement Equivalence Across Gender, Race, and Ethnicity

• In Event: New Directions in Studying Adult and Children’s Emotion Knowledge

Abstract

Introduction. Emotion knowledge (EK) undergirds children’s ability to categorize and label physiology, expressions, and situations. Across frameworks used to organize EK and related constructs, numerous studies suggest EK is a multidimensional construct (Brackett et al., 2019; Bassett et al., 2012; Castro et al., 2016; Fan et al., 2010). Although fundamental to construct validity (Campbell & Fiske, 1959), no studies of EK in young children have partialed out systematic method variance using multi-trait, multi-method analysis despite estimates suggesting methods may account for over 50% of the total variance (Cote & Buckley, 1987). The first aim of this study is to investigate the factor structure of EK in young children using multiple measures after partialling out method variance. The second aim is to explore the invariance of EK across gender and race. Methods. Data featured 982 preschoolers between 32 and 68 months in 209 classrooms at 65 preschools. Parents identified children as 45% Hispanic/Latine, 49% White, 28% Black/African American, and 20% Multiracial or other background. Children were assessed in English or Spanish using three measures of EK: Affect Knowledge Task-Shortened (AKT-S, 4 sub-scales; Denham et al., 2015), Emotion Matching Task (EMT, 4 sub-scales; Morgan et al., 2010), and an adapted Box Task (A-Box, 1 scale; Russell & Widen, 2002). Assessments varied in methodology with either cartoon depictions of puppets or child or adult photographs embedded within matching, sorting, or labeling tasks. Analyses. Three correlated-trait, correlated-method (CT-CM) confirmatory factor analysis models were run and compared against a baseline model representing the original nine scales without method factors. Models included correlated method factors representing the AKT-S, EMT, and A-Box but differed by way of (1) a univariate EK factor, (2) bivariate emotion perception and emotion understanding factors, and (3) trivariate recognition, understanding, and labeling factors. Results. The trivariate model fit the data best (see Table 1.1). Recognition (⍵ = .92, pseudo-R2 = .38), understanding (⍵ = .83, pseudo-R2 = .39), and labeling (⍵ = .88, pseudo-R2 = .38) factors were reliable when including method variance, but 28%, 45% and 38% of this reliable variance was due to the methods, respectively. Method factors were also reliable, ⍵s = .86–.89 (pseudo-R2s = .39–.44) but between 39 and 90% of the method variance due to the a greater EK factors. The trivariate model demonstrated item slope, intercept, and residual invariance as well as factor variance, covariance, mean invariance between girls and boys and between racial and ethnic groups (Table 1.2). Differences across EK factors were small and not statistically significant between groups (|ds| = 0.02–0.18). Most differences across methods tended to be small (|ds| = 0.02–0.16), but each method presented with its own moderate, statistically significant difference (|ds| = 0.21–0.35). Discussion. Results align with prior research that EK is a multidimensional construct but goes further in suggesting labeling emotions may be distinct from understanding and recognizing emotions, at least in young children after accounting for the methods. Studies not accounting for method variance may lead to inaccurate conclusions about EK due to gender, racial, and ethnic biases in methodology.

Authors

• Navsheen Kaur, Yale University
  Presenting Author

• Craig S. Bailey, Yale School of Medicine