Poster #3 - Modeling changes in executive functions during middle childhood: a bifactor modelling approach

• In Event: Poster Session 12

Abstract

Background: Executive functions (EF) have been mostly investigated through latent variable methods (e.g., confirmatory factory analysis-CFA) with several studies pointing to changes in EF structure during childhood from a unitary to a multidimensional factor structure. Specific EF factors in adulthood include inhibition, working memory and cognitive flexibility. How EF differentiate into these factors remains elusive. As done in the field of intelligence, we propose to use bifactor models, in which each task-indicator loads not only into a common factor, capturing their common variance, but also into their corresponding specific factor, capturing factor-specific variance to better characterize the development.
In the context of bifactor models, we consider two competing hypotheses: the differentiation hypothesis postulates that during childhood the variance captured by the common factor decreases with age, leading to an increase of the variance explained by specific factors (i.e. inhibition, working memory, cognitive flexibility); in contrast, the dynamic mutualism theory posits de-differentiation with age – or an increased loading on the common factor from childhood to late adolescence.
Hypotheses: As children’s performance becomes less idiosyncratic with age, higher loadings (and thus explained variance) in both the common factor and specific factors is expected for older as compared with younger children (H1). Furthermore, we hypothesize variability in the relative increase of loadings in specific factors at older age (H2), as previous studies suggest the specific factor of cognitive flexibility differentiates slower than others. Finally, dynamic mutualism theory predicts that the common factor will show greater relative increase in explained variance compared to specific factors (H3).
Methods: Data from 300 children (8-12 years) who completed a battery of nine tasks tapping different EF components (e.g., inhibition, cognitive flexibility and working memory) was recently collected in primary schools (Switzerland). Data analysis plan: The sample will be divided between younger and older children through median-split procedure. Next, measurement invariance (MI) will be performed between both groups to test our hypotheses in a stepwise procedure. The starting point is a bifactor model with a common factor and specific factors (i.e., inhibition, cognitive flexibility and working memory), MI will allow to test if:
1) The latent factors of the bifactor model show a similar organization across age groups (i.e., both groups fit a bifactor model with 3 specific factors or both fit a bifactor model with 2 specific factors) ensuring configural invariance, a prerequisite to test H1-H3.
2) The factor loadings from common and specific latent factors are different between the two age groups (also known as metric invariance testing), allowing us to assess H1, H2 and H3.
3) If, however, metric invariance is found in step 2), whether differences exist between both groups in the intercepts of the task-indicators that form the model will be tested (also known as scalar invariance testing). This will allow us to characterize possible performance differences across age groups.

Note that if configural invariance is not supported, Bayesian CFA model comparison will be performed to determine the most likely structure of EF for each group separately.

Authors

• Marc Yanguez Escalera, university of geneva
  Presenting Author

• Benoit Bediou

• Julien Chanal

• Daphne Bavelier