Multisource Longitudinal Network and Latent Variable Model Analyses of ADHD Symptoms in Children

ABSTRACT

Objective: Multisource longitudinal network analysis was used to determine if between-child and within-child variance of attention-deficit/hyperactivity disorder (ADHD) symptoms provided unique findings of ADHD relative to latent variable model (LVM) analyses.

Method: Mothers and fathers of 802 Spanish first-grade children (54% boys) provided ratings of ADHD symptoms at two time points six weeks apart (assessment 1: 723 mothers and 603 fathers; assessment 2: 667 mothers and 584 fathers). Network and latent variable models were applied to the ratings.

Results: Inattention, hyperactivity, and mixed hyperactive/impulsive symptom communities occurred for the within- and between-children’s symptom networks with the results being consistent across mothers and fathers, especially for the between-children’s symptom networks. LVM analyses identified three factors with the same symptoms on each factor as in the symptom communities. These models also showed invariance across mothers and fathers as well as assessments.

Conclusions: Longitudinal networks provided several useful insights for ADHD, including centrality symptoms that differed across between- and within-child levels. However, many findings were also largely consistent with the LVM analyses. Future studies should use novel methods (e.g., intensive longitudinal measurement) and analytic tools to determine if more unique theoretical and clinical findings emerge when applying network analysis to longitudinally measured ADHD symptoms.

Acknowledgments

This research was supported by two grants from the Ministry of Economy and Competitiveness of Spanish Government under award numbers PSI2014-52605-R and PSI2017-82550-R (AEI/FEDER, UE). Stephen Becker is supported by award number K23MH108603 from the National Institute of Mental Health (NIMH). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health (NIH) or the Spanish Government. We thank Cristina Trias for assistance with the study.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental Material

Supplemental material for this article can be accessed online at https://doi.org/10.1080/15374416.2020.1756297.

Notes

¹ Proponents of network analysis criticize the LVM for the use of the latent variable as the cause of symptom occurrence without the specification of the causal mechanism (i.e., the use of the latent variable as a cause of the symptoms without measurement of the latent variable independent of symptom measurement). The LVM, however, does not have to give causal properties to the latent variable (Preszler & Burns, 2019). The latent variable can instead represent the common variance in a set of symptoms with additional research required to identify the external correlates of the latent variable with these correlates then providing the explanation of the latent variable.

² The children from the Mallorca and Madrid schools did not differ significantly (ps >.05) on the IN and HI measures at assessments one and two for the two sources. Children not present for the second assessment also did not differ significantly (ps >.05) from children present for the second assessment on the IN and HI measures for the first assessment for the two sources.

³ There was little missing information at the item level. Missing information involved the absence of a mother or father (i.e., only one rating for the child). The WLSMV estimator for the factor analyses uses a pairwise approach to the treatment of missing information (i.e., each correlation in the polychoric correlation matrix is based on the maximum amount of information). Network analysis, however, requires the use of listwise deletion across the two assessments within each source because current network analysis software packages do not allow for simultaneous missing data procedures. This resulted in the same 585 mothers and 480 fathers at each assessment. Given the different procedures for the treatment of the missing rater across the LVM and network analysis, it is important to remember this if the results differ across the LVM and network analysis. However, if the findings are the same across the two procedures, then this difference becomes less important.

Additional information

Funding

This work was supported by the Ministerio de Economía y Competitividad [PSI2014-52605-R,PSI2017-82550-R] and National Institute of Mental Health [K23MH108603].