Latent Variable Statistical Methods for Longitudinal Studies of Multi-Dimensional Health and Education Data: A...

(1) Background: Most studies including health data have relied on reducing all variables to manifest scores, ignoring the latent nature of variables. Moreover, relying only on manifest variables is a limitation of longitudinal studies where identical measures cannot be collected at each time point. (2) Objective: This scoping review aims to identify latent variable statistical methods for longitudinal studies of multi-dimensional health and educational data investigating early health predictors of long-term educational outcomes and developmental trajectories that lead to better or worse than expected outcomes. (3) Eligibility criteria: We included peer-reviewed health and education journal articles, doctoral theses, and book chapters of longitudinal studies of children under 12 years of age that adopted latent variable, multivariate analysis of three or more waves of data. We only included full-text-available, English-written articles, without restriction on date of publication. (4) Sources of evidence: We searched five databases, Scopus, MEDLINE, PsycINFO, ERIC, and Web of Science, and identified 4836 publications for screening. (5) Results: After title, abstract, and full-text screening, nine studies were included in the review, reporting seven statistical methods. These methods were categorised into two groups-variable-oriented modelling and person-oriented modelling. (6) Conclusions: Variable-oriented modelling methods are useful for determining predictors of long-term educational outcomes. Person-oriented modelling methods are effective in detecting trajectories to better or worse than expected outcomes. (7) Registration: Open Science Framework.