The structure of early adolescent temperament: EFA and EGA approaches for the ABCD study

• In Event: Poster Session 09
  In Poster Session: PS 09 Topic - Methods, History, Theory

Abstract

Early adolescent temperament is longitudinally related to outcomes spanning many domains in adulthood, including personality (e.g., Hirone et al., 2018) and psychopathology (e.g., Rabinowitz et al., 2019). The Early Adolescent Temperament Questionnaire-Revised is a 65-item measure designed to capture three broad dimensions of temperament – emotionality, reactivity, and self-regulation – over twelve narrow subscales (Capaldi & Rothbart, 1992; Hoffman, et al., 2019; Latham, et al., 2019). Despite this, there has not been methodological consensus regarding the factor structure of early adolescent temperament. Previous factor analyses of the scales have found many different numbers of factors, ranging from as few as three (Capaldi & Rothbart, 1992: negative emotion/somatic arousal, emotion/sensitivity, and high intensity pleasure/sensation seeking; Demirpence & Putnam, 2020: effortful control, surgency, and affiliativeness) to as many as nine (Muris & Meesters, 2009: frustration/inhibitory control, affiliativeness, high intensity pleasure, pleasure sensitivity, shyness, perceptual sensitivity, activity level, fearfulness and activation control). Moreover, network models for the relationships between EAT-Q items have not been considered, leaving open the possibility that direct causal relationships between items, which are not accounted for by latent variable models, may be present.
The proposed study is an exploratory factor analysis (EFA: Sellbom & Tellegan, 2019) and an exploratory graph analysis (EGA; Golino & Epskamp, 2017), of EAT-Q items arising from the Adolescent Brain Cognitive Development Study (ABCD study; Jernigan, et al., 2018), an ongoing multi-site longitudinal study. The EAT-Q was administered in the baseline survey to N = 11,878 subjects at the ages of 9-10 years old. Within the EFA, the number of factors will be determined by scree plot followed by a factor analysis with oblique rotation to allow for the factors to correlate. The EGA will begin with lasso regularization for network estimation, followed by the walktrap algorithm for community estimation, allowing clusters of items to be determined without a priori specification. The methods will be compared via their fit indices, which will inform the appropriateness of each model for the ABCD study data, and by the resulting number of factors and clusters, which will yield structures of temperament for early adolescence. Additionally, the centrality measures of the network model will be assessed which will further edify the structure of early adolescent temperament by indicating which EAT-Q items are most central to temperament. These analyses will be completed by November 15th.
Methodologically, running both EFA and EGA contributes to a larger debate regarding the viability of EGA as a method apart from EFA. Previous work on this subject has shown that EGA is more accurate under particular conditions (Golino & Demetriou, 2017; Golino & Epskamp, 2017) and the structure of the model lends itself to better treatment options (Bak, et al., 2016; Beard, 2016; McNally, 2016; Haslbeck & Fried, 2018; Lutz, 2018). Substantively, using both methods allow for better understanding of early adolescent temperament, which, as mentioned above, has numerous consequences for future outcomes. Thus, we propose this EFA/EGA study of ABCD study EAT-Q data in order to further knowledge of early adolescent temperament and the use of EGA.

Authors

• Lindley R. Slipetz, Wake Forest University
  Presenting Author

• Veronica Tess Cole, Wake Forest University
  Non-Presenting Author