Optimizing Program Effects: New Methodological Developments in Testing How a Program Achieves its Intended Impact
Intervention programs are designed to change mediating (intermediate) variables theorized to be causally related to the outcome variable. In this chapter published in Oxford Handbook of Quantitative Methods, we aimed to supplement the existing resources on mediation analysis with a description of recent advances.
Theories in many substantive disciplines specify the mediating mechanisms by which an antecedent variable is related to an outcome variable. In both intervention and observational research, mediation analyses are central to testing these theories because they describe how or why an effect occurs. Over the last 30 years, methods to investigate mediating processes have become more refined.
The purpose of this chapter is to outline these new developments in four major areas:
(1) significance testing and confidence interval estimation of the mediated effect,
(2) mediation analysis in groups,
(3) assumptions of and approaches to causal inference for assessing mediation, and
(4) longitudinal mediation models.
The best methods to test mediation relations are described, along with methods to assess mediation relations when they may differ across groups. Methods for addressing causal inference and models for assessing temporal precedence in mediation models are used to illustrate some remaining unresolved issues in mediation analysis, and several promising approaches to solving these problems are presented.
MacKinnon, D.P., Kisbu-Sakarya, Y., & Gottschall, A. (2013). Developments in mediation analysis. In T.D. Little (Ed.), Oxford Handbook of Quantitative Methods. New York: Oxford University Press.