The Use of Multilevel Random Coefficient Models for the Analysis of Mediational Change Over Time

• In Event: Promising Scholarship in Education: Dissertation Fellows and Their Research

Abstract

Mediation analysis permits researchers to address questions concerning mechanisms underlying associations between variables by positing the relationship between initial (x) and outcome (y) variables is at least partly due to an intervening variable, the mediator (m). Longitudinal mediation models explicitly model temporal precedence among x, m, and y variables, but such models currently employed in the literature do not address a potentially novel research question – Does the mediated effect change over time? Finding a differential pattern of mediation over time has the potential to explain conflicting results often found in substantive literature and can be informative for designing effective interventions.
This study develops a multilevel modeling method to evaluate mediational change over time (the M-COT model). This model combines multilevel growth modeling and multilevel mediation approaches into a single comprehensive analysis. The M-COT model produces an estimate of the rate of change in meditational processes not directly available from a series of single-level mediation analyses at each of the different timepoints.
This study uses simulation techniques to determine the necessary conditions for successful application of the M-COT model and compares post hoc estimates of timepoint-specific effects to those of single-level mediation models estimated timepoint by timepoint. Sample size, the nature of the change in the mediated effect over time (e.g. constant, linear, etc.), and attrition levels were varied. Model misspecification and selection was also examined. Results indicate the multilevel technique performs well across conditions and M-COT estimates of timepoint-specific mediated effects have smaller standard errors, and thus more power, than single-level estimates.

Author

• Erin Hilary Arruda, University of California, Los Angeles