Academic Seminar : "Mediation Analysis is a Mess."

Academic Seminar : "Mediation Analysis is a Mess."

Rik Pieters , Tilburg School of Economics and Management (TISEM) of Tilburg University

Date : December 9 (Tuesday)

Time : 10.00 a.m

Place: SOM 1073

Abstract :

Mediation analysis is a popular tool for making causal inferences in
management, marketing, sociology, psychology, and epidemiology, up to
the point that it is hard to submit a paper to the main journals in
these disciplines without one. Yet, it is rarely recognized that
causal inferences from mediation analysis rest on conditions that the
data generating mechanisms need to satisfy. This study specifies six
conditions for making causal inferences from mediation analysis. It
offers a survey of recent mediation analyses in the marketing and
social psychology literature, which shows that the conditions are
rarely discussed or even mentioned, and also that reporting of
mediation results prevents knowledge accumulation in the disciplines.
Then, it presents the results of Monte Carlo experiments which reveal
the grave errors that can result from failures to satisfy the
conditions. Thus, mediated effects may be larger or smaller than
reported, significant mediated effects may be insignificant, and
positive effects may be negative, and the other way around. Finally,
it presents recommendations for ameliorating current mediation
analysis and reporting practice.