Mediation analysis

Abstract Understanding mechanisms between the urban environmental exposome and cognitive functioning outcomes is a key aim of the SPACE project. Structural equation modelling (SEM) offers a framework in which to consider potential mechanisms identified in DAGs (see IAP, Presentation 1) and evaluate their contribution to total effects between the environment and cognitive functioning. SEM allows multiple regression-based pathways to be computed simultaneously, meaning that variables can flexibly be specified as both outcomes and predictors in one overarching model. Another advantage of SEM is its use of latent modelling, which reduces measurement error (particularly useful in a multivariate context with many variables). This presentation will provde a brief introduction to SEM, latent modelling, and mediation analysis. Using state of the art approaches to assessing mediation with longitudinal data in a SEM framework, and with data from waves 1 and 2 of NICOLA, we will then discuss the application of mediation in SEM to relevant paths identified in the DAGs, including psychosocial and behaviour mediators (e.g. depressive symptoms, social isolation, physical activity). Finally, we will proceed through a worked example (the path from the urban environmental exposome, via depressive symptomatology as a mediator, to cognitive functioning as an outcome), and discuss a specific advantage of SEM mediation analysis: model modification.