Session: Longitudinal Mediation Models in Prevention Science (Society for Prevention Research 25th Annual Meeting)

Statistical mediation is the extent to which the relation between an independent variable X and a dependent variable Y is explained by a third variable, a mediator M, which transmits the influence of X to Y. Many data sets in the prevention sciences include variables that are hypothesized mediators. Given the nature of mediational hypotheses that an independent variable X leads to a mediator M which in turn leads to an outcome Y, mediation researchers have strong interests in longitudinal mediation models as well.

The first poster in this session examines longitudinal mediation from a structural equation modeling perspective using latent change scores. Specifically, the authors examined how time metric influenced interpretation of results in latent change score mediation models with real data from a longitudinal alcohol study.

The second poster in this session investigates the performance of survival-mediator models using different missing data handling methods for different types of missing data assumptions. The authors found that modern missing data techniques must be used in order to achieve unbiased results in the survival-mediator model when missingness is not random.

The third poster in this session considers measurement theory in the context of a two-wave longitudinal mediation model. The authors examined the effects of violating longitudinal measurement invariance in the two-wave mediation model, and found that non-invariance led to biased results.

The fourth poster in this session examines violations of key causal inference assumptions for a three-wave mediation model. The authors found that causal inference methods performed better when certain assumptions were violated, but that traditional regression methods were easier to implement and interpret for longitudinal mediation.

The fifth poster in this session investigates sequential ignorability assumptions of mediation. When X cannot be randomly assigned in a longitudinal mediation design, all of the effects in the mediation model may be influenced by unmeasured confounders.

In conclusion, the posters in this session complement each other by examining different longitudinal mediation models, and different aspects of those models. Longitudinal mediation models are important for examining mechanisms of change over time in prevention research.