The Importance of Testing Mediation and Moderation
Jon A. Krosnick
Departments of Communication,
Political Science,
and Psychology
Stanford University
Outline for Today
• Introduction: Defining and Testing Mediation and Moderation
• Example: News Media Priming
• Example: Public Learning About Policy Issues
Introduction
The Audience Participation Portion of Today’s Show
What is a moderator?
- Definition- Example of moderation
What is a mediator?
- Definition- Example of mediation
Moderator
A variable that changes the impact of one variable on
another.
Predictor Outcome
Moderator
Testing a Moderator Hypothesis
Don’t Presume Linearity
Mediator
The mechanism by which one variable affects another
variable
Testing Mediation
• Step 1: IV DV
• Step 2: IV Mediator
• Step 3: Mediator DV
• Step 4: Effect of IV on DV is significantly reduced by controlling for the mediator:– Sobel (1982) test
(http://www.unc.edu/~preacher/sobel/sobel.htm)– Goodman (1960) test
On the exact variance of products. Journal of the American Statistical Association, 55, 708-713.
First Example:
News Media Priming
News Media Priming
• News media attention to an issue increases the impact of performance assessments in the domain on overall presidential evaluations.
Presidential Performance Handling Unemployment
Overall Presidential Performance
Media Attention to Unemployment
Mediation of Moderation
Unemployment Presidential Performance
Overall Presidential Performance
Media Attention to Unemployment
Accessibility of UnemploymentPerformance Assessment
Unemployment
Story
Unemployment
Attitudes
Unemployment
Study Design
- Pre-exposure questionnaire
- 5 non-political TV news stories
2 stories on either:
Crime Pollution Unemployment
- Post-exposure questionnaire
Measures
• Overall presidential performance• Presidential performance handling:
– Crime– Pollution– Unemployment
• Accessibility (response latency)• Political knowledge• Trust in the news media
Moderation
of Mediation
of Moderation!
Not Mere Accessibility – Persuasion?
Unemployment Presidential Performance
Overall Presidential Performance
Media Attention to Unemployment
Belief that Media Personnel Believe the Issue is Important for the NationTrust
in the Media
Political Knowledge
Belief that the Issue is Important for the Nation
Conclusion
• Understanding moderation and mediation increase confidence in the causality claim
• Understanding mediation changed the political character of the effect (not victims of the architecture of the mind)
• Understanding moderation changed the normative spin of the effect (not naïve people who lack political expertise – people choose to be influenced)
Example Two:
How Do People Learn About Politics?
October 13, 1988 Presidential Debate
George H. W. Bush vs. Michael Dukakis
Day Before: 134 people interviewed
Day After: 63 reinterviewed
- Cued Recall
- Recognition Memory
- Attitude Importance
Analysis: Hierarchical Linear Modeling
Participant
Taxes Capital Defense
Punishment Spending
CR RM CR RM CR RM
Effect of Importance on Knowledge
b=.46
SE=.23
p<.05
N=366
The Usual Approach
Stop There.
Smile.
Fold Up Tent.
Go Home.
.11*
.04
.31**
.44
Laboratory Experiment Conditions
Unpaced
A woman should have the right to obtain an abortion if she so chooses.
The death penalty is an acceptable punishment for convicted murderers.
We should build more nuclear power plants to provide an alternative source of energy.
The U.S. should send more troops to Saudi Arabia to oppose Iraq’s invasion of Kuwait. In order to protect American jobs, we need more restrictions on the number of goods imported.
Candidate A
Laboratory Experiment Conditions
Unpaced
Paced
“You will have 20 seconds to read five statements made by each candidate.
After 20 seconds, the statements will disappear, so read as quickly as you can.”
Laboratory Experiment Conditions
Unpaced
Paced
Elaboration Time
“You will have 20 seconds to read five statements made by each candidate.
After 20 seconds, the statements will disappear, so read as quickly as you can.
Once the statements disappear, you will have 45 seconds to think about what you read.”
Laboratory Experiment Conditions
Unpaced
Paced
Elaboration Time
Topic Labels
A woman should have the right to obtain an abortion if she so chooses.
The death penalty is an acceptable punishment for convicted murderers.
We should build more nuclear power plants to provide an alternative source of energy.
The U.S. should send more troops to Saudi Arabia to oppose Iraq’s invasion of Kuwait.
In order to protect American jobs, we need more restrictions on the number of goods imported.
Candidate A
Abortion
Death Penalty
Nuclear Power
Iraq
Imports
Measures
• Free Recall
• Recognition Memory
• Attitude Importance
Alternative Hypothesis
Actual
Knowledge
Volume
Perceived
Knowledge
Volume
Attitude Importance
.37*
.27*
.38*
.55** .04 .00
Conclusions
• You can test mediation with panel data
• You can test mediation via moderation manipulations in the lab
• You can test mediation with cross-sectional data with instrumental variables and 2SLS
Broader Conclusion
• Test mediation (and moderation)!• Two examples where this was not done:
Valentino, N. A., Hutchings, V. L., & White, I. K. (2002). Cues that matter: How political ads prime racial attitudes during campaigns. American Political Science Review,96, 75-90.
Tali Mendelberg The Race Card: Campaign Strategy, Implicit Messages, and the Norm of Equality (Princeton Univ Press, 2001).