David P. MacKinnon talks with
ScienceWatch.com and answers a few questions about
this month's Fast Moving Front in the field of Economics
Article: A comparison of methods to test mediation
and other intervening variable effects
DP;Lockwood, CM;Hoffman, JM;West,
Journal: PSYCHOL METHODS, 7 (1): 83-104 MAR 2002
Addresses: Arizona State Univ, Dept Psychol, Tempe, AZ
Arizona State Univ, Dept Psychol, Tempe, AZ 85287 USA.
Why do you think your paper is highly
The paper investigates an important method in many fields, the extent to
which a variable transmits or mediates the relation between an independent
and a dependent variable. The paper reviews existing tests for mediating
variables from many different fields and describes several new tests.
One reason why the paper is highly cited is that it has a broad perspective
to include tests from many different areas of inquiry. In some cases,
widely used methods did not perform as well as expected in statistical
simulations. Information on the different tests and their statistical
performance had not appeared in the research literature before.
Does it describe a new discovery, methodology, or
synthesis of knowledge?
There are several new tests in the article. A large number of different
methodologies had been used to investigate mediating variables and this
article was the first to compare them theoretically and with a statistical
Would you summarize the significance of your paper in
Mediation analysis is useful in many areas. One modern use of this method
is to determine how treatment and prevention programs achieve their
effects. For example, mediation analysis can be used to assess whether a
program reduced cigarette smoking by increased social support for quitting
Mediation analysis can be used to investigate how a program to reduce heart
attacks achieves its effects, such as whether the program reduced blood
pressure which then reduced subsequent heart attacks. If researchers know
how a treatment or prevention program works then more efficient and less
expensive treatment and prevention programs can be developed.
How did you become involved in this research and were
there any particular problems encountered along the way?
I became involved in this research around 1987 when I was asked to
investigate how a school-based drug prevention program achieved its
effects. In the course of this research, I realized that there were many
different tests of mediation from many different fields. It was unclear
which test or tests were the best to use. As I learned more about these
tests, I learned that they differed widely in their statistical
performance. Importantly, this research received support from the National
Institute on Drug Abuse.
Where do you see your research leading in the
The paper describes a relatively simple mediation model. Future work is now
addressing mediation analysis when variables are categorical, when
longitudinal data are collected, influence of violation of assumptions, and
experimental designs to assess mediation.
Do you foresee any social or political implications for
Mediation analysis has the capability of improving treatment and prevention
programs. Mediation analysis can also improve existing social programs so
that they may have greater effects and even cost less.
David P. MacKinnon, Ph.D.
Foundation Professor of Psychology
Arizona State University
Tempe, AZ, USA
Keywords: mediating variables, statistical simulations,