Missing data approaches for longitudinal neuroimaging research: Examples from the Adolescent Brain and Cognitive Development (ABCD) Study®

Abstract

This paper addresses the challenges of managing missing values within expansive longitudinal neuroimaging datasets, using the specific example of data derived from the Adolescent Brain and Cognitive Development (ABCD) ${\text{Study}}^{\text{®}}$. The conventional listwise deletion method, while widely used, is not recommended due to the risk that substantial bias can potentially be introduced with this method. Unfortunately, recommended alternative practices can be challenging to implement with large datasets. In this paper, we advocate for the adoption of more sophisticated statistical methodologies, including multiple imputation, propensity score weighting, and full information maximum likelihood (FIML). Through practical examples and code using ABCD ${\text{Study}}^{\text{®}}$ data, we illustrate some of the benefits and challenges of these methods, with a review of how these advanced methodologies bolster the robustness of analyses and contribute to the integrity of research findings in the field of developmental cognitive neuroscience.