A mediator variable explains the how or why of an (observed) relationship between an independent variable and its dependent variable.
In a mediation model, the independent variable cannot influence the dependent variable directly, and instead does so by means of a third variable, a ‘middle-man’.
In psychology, the mediator variable is sometimes called an intervening variable. In statistics, an intervening variable is usually considered to be a sub-type of mediating variable. However, the lines between the two terms are somewhat fuzzy, and they are often used interchangeably.
Watch the video for examples of mediator variables, plus an overview, or read on below:
Mediator Variable Examples
A mediator variable may be something as simple as a psychological response to given events. For example, suppose buying pizza for a work party leads to positive morale and to the work being done in half the time.
- Pizza is the independent variable,
- Work speed is the dependent variable,
- The mediator, the middle man without which there would be no connection, is positive morale.
Although we may observe a definite effect on work speed when and if pizza is bought, the pizza itself does not have the power to affect work rates: only by affecting morale of the workers can it make an actual difference.
Full Mediation and Partial Mediation
Full mediation is when the entire relationship between the independent & dependent variables is through the mediator variable. If you take away the mediator, the relationship disappears. Since the real world is a complicated place with many interactions, this is less common than partial mediation.
Partial mediation happens when the mediating variable is only responsible for a part of the relationship between independent & dependent variables. If the mediating variable is eliminated, there will still be a relationship between the independent and dependent variables; it just won’t be as strong.
Mediational hypotheses, by definition, include full (complete) mediation. In other words, the independent variable has zero effect on the dependent variable; the causal relationship depends entirely on the mediator.
Baron and Kenny’s Four Steps
Baron and Kenny (1986), Judd and Kenny (1981), and James and Brett (1984) outlined the following steps to identify the mediational hypothesis. If the steps are met, then variable M is said to completely mediate the X-Y relationship. The steps are
- Show that a the independent variable (X) is correlated with the mediator (M).
- Demonstrate that the dependent variable (Y) and M are correlated.
- Demonstrate full mediation on the process. The effect of X on Y, controlling for M (i.e. controlling for paths a and b in the image at the top of this page), should be zero. If the results for this step are anything but zero, then there is partial mediation.
The authors state that three regression analyses are needed:
- X as the predictor variable and M as the outcome variable.
- X as the predictor variable and Y as the outcome variable.
- X and M as the predictor variables and Y as the outcome variable.
The procedures come with some hefty explanations, which are beyond the scope of this article. I recommend reading Baron and Keny’s original text. Or, as an excellent (plain English) alternative, read Paul Jose’s Doing Statistical Mediation and Moderation: Methodology in the Social Sciences, which includes Baron and Kenny’s steps starting on page 20.
Mediator versus Moderator variables. Retrieved from http://psych.wisc.edu/henriques/mediator.html on June 26, 2018.
Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182. Retrieved June 26, 2018 from http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.169.4836&rep=rep1&type=pdf on June 26, 2018 Butler, Adam. Mediation Defined. Retrieved from https://sites.uni.edu/butlera/courses/org/modmed/moderator_mediator.htm on June 26, 2018