@article{2017:demming:conducting,
title = {Conducting Mediation Analysis in Marketing Research},
year = {2017},
note = {The term mediation analysis encapsulates methodological approaches for
determining whether a causal effect of a predictor on an outcome
variable can be explained by another variable. This explanatory
variable is called a mediator. Mediation analysis is therefore
particularly relevant for those interested in grasping the underlying
mechanism of a focal effect. This paper reviews the purpose and aims
of mediation analysis, common approaches of conducting mediation
analysis, and how mediation analysis can enrich research findings. We
especially address the widespread uncertainty of researchers and
practitioners in the marketing field in deciding which method to
implement to explore a proposed mediation mechanism and how to conduct
the analysis. The review contributes to extant literature by
connecting theoretical and practical knowledge in order to help in
applying mediation analysis. Furthermore, the second part of the paper
is organized as a tutorial in order to make the state-of-the-art
methodology applicable to the reader. Here, we illustrate how to
specify, interpret, and report results using PROCESS, a commonly used
macro developed specifically for mediation analysis with SPSS and SAS.
In the first part of the paper we introduce two basic elements of
mediation processes: the indirect and the direct effect. The indirect
effect indicates whether a proposed mediator can explain the
relationship between a predictor variable and an outcome variable.
Therefore, estimating the indirect effect is essential for
establishing mediation. The interpretation of the indirect effect
depends on the model structure – for example the number of proposed
mediators and their interconnection. Based on the indirect effect, we
discuss four commonly used mediation model groups. Conversely, the
direct effect reveals to what extent a proposed mediator can explain
the relationship of a predictor variable and an outcome variable.
Thus, the direct effect determines whether the mediator can fully
explain the relationship (full mediation) or only partially explain it
(partial mediation). We refer to this broad distinction as mediation
types. Determining the mediation type can guide further theory
building, as it may indicate a second mediator that has been omitted
in the analysis thus far. In the methodology section of the paper, we
review three different regression-based approaches for inferring
mediation: the Baron-and-Kenny approach, the Sobel test, and
bootstrapping. The Baron-and-Kenny approach relies on estimating each
relationship of a mediation model separately and then inferring
mediation if all relationships are significant. The Sobel test is a
parametric test of the indirect effect. While common, both approaches
are widely criticized because of their questionable assumptions. In
contrast to these traditional approaches, bootstrapping is a
non-parametric approach that provides an accurate test of the indirect
effect and is based on data resampling. The significance of an
indirect effect is then inferred from a confidence interval of its
bootstrap distribution. As bootstrapping is the state-of-the-art
methodology in mediation analysis, we limit our focus to this approach
for the subsequent tutorial. After the theoretical discussion, the
tutorial illustrates in detail how to specify different model groups
with the easy-to-apply macro PROCESS. In this section we work with an
example data set from the marketing field to illustrate the steps of
the mediation analysis.},
journal = {Marketing ZFP},
pages = {76--98},
author = {Demming, Carsten L. and Jahn, Steffen and Boztug, Yasemin},
volume = {39},
number = {3}
}