Exploring Co-development of Executive Functions and Math Achievement Using Cross-lagged Panel Model with Fixed Effects

• In Event: Poster Session 12
  In Poster Session: PS 12 Topic - Education, Schooling

Abstract

Executive functions (EF) play an important role in children’s math skill development, both concurrently and longitudinally (Peng & Kievet, 2019). Previous studies examining the effects of EF on math achievement either consider EF as a unidimensional construct or examine the sub-components separately (Bull et al., 2014; McClelland et al., 2014). Although these studies find strong associations between various EF sub-components and math achievement, few studies explicitly examine the direction of the associations, that is, whether EF and math develop reciprocally (see Miller-Cotto & Byrnes, 2019 for an exception using working memory). Moreover, as much attention has been focused on the development of EF in younger children (i.e., from kindergarten to early elementary school), no clear evidence has been found regarding the bidirectionality of the relationship for children in later years of elementary school.

In the present study, we investigated the long-term bidirectional relationship between working memory, cognitive flexibility and mathematics achievement between kindergarten and fifth grade. We used the nationally representative data from Early Childhood Longitudinal Study-Kindergarten (ECLS-K) which was collected at nine time points between 2011 and 2016.

To examine the bidirectional relationship, we applied a novel modeling approach, the cross-lagged panel model (CLPM) with fixed effects proposed by Allison et al. (2017), which allows for controlling unobserved time-invariant confounders and a clearer focus on within-person variation in panel data. This model effectively solved the critiques raised towards the traditional CLPM which include inadequately separating the between-person and within-person differences and producing biased estimates for cross-lagged effects.

To build the CLPM with fixed effects (fig.1), we estimated autoregressive effects within EF and math achievement, which reflect stability over development as well as cross-lagged effects, which reflect the reciprocal relationship. We also added two latent variables as the fixed effects to remove undesired between-person variation and ensure the estimated cross-lagged effects only reflect the within-person variation of interest.

We found increasing EF leads to better math achievement at the subsequent time point (βs= .041-.103, ps < .01) while controlling for prior math performance. Meanwhile, better math performance at preceding time point also leads to higher EF later (βs= .206-.347, ps < .01) while controlling for prior EF. The reciprocal relationship between EF and math holds from kindergarten to fifth grade (fig.2). The model containing bidirectional paths has a better model fit than nested models, which only contain unidirectional paths. This suggests that EF and math develop in a reciprocal manner rather than a unidirectional way even under a strict control which is controlling for all unobserved time-invariant confounders. However, the predictive power in both directions exhibited a decreasing trend as children grow older. This may suggest that as children age, their mathematics achievement are relying less on EF and more on prior mathematics knowledge. We will discuss this fading using the dual-processing theories of higher cognition.

Overall, the present findings provide strong evidence for the existence of the long-term reciprocal association between EF and math and allow for getting closer to causal estimates regarding the relationship between EF and math.

Authors

• Haobai Zhang, University of Delaware
  Presenting Author

• Dana Miller-Cotto, Kent State University
  Non-Presenting Author

• Nancy C Jordan, University of Delaware
  Non-Presenting Author