A mediator in psychology is a variable that explains the process through which one thing affects another. Instead of simply knowing that stress leads to drinking problems, for example, a mediator tells you the “how” or “why” behind that connection. Maybe stress increases feelings of distress, and that distress is what actually drives the drinking. In this case, distress is the mediator.
The concept is central to psychological research because it moves beyond surface-level observations (“X causes Y”) and into the mechanism underneath. Mediators describe the psychological process that creates a relationship between two variables, and they are always dynamic properties of individuals: emotions, beliefs, behaviors, or perceptions that shift in response to some trigger and then influence an outcome.
How a Mediator Works
A mediator sits in the middle of a causal chain. The predictor variable causes changes in the mediator, and the mediator then causes changes in the outcome. Think of it as a domino in between two others. The first domino (the predictor) doesn’t knock over the third (the outcome) directly. It knocks over the middle one (the mediator), which then knocks over the third.
A well-known example from workplace psychology illustrates this clearly. Researchers studying the link between job conditions and alcohol use found that work pressures and lack of control didn’t lead to drinking directly. Instead, those work features increased work-related distress, and that distress, in turn, increased drinking. Work distress was the mediator that explained how job conditions translated into a behavioral outcome.
Another example comes from health psychology. In a study of women living with HIV, researchers hypothesized that stress would lead to psychological distress, but that the link ran through avoidance coping, meaning behaviors like denial or mental disengagement. The data supported this: stress predicted greater use of avoidance coping, avoidance coping predicted greater psychological distress, and the indirect pathway through coping was statistically significant. Avoidance coping partially mediated the relationship between stress and distress.
Full Mediation vs. Partial Mediation
When a mediator completely accounts for the relationship between a predictor and an outcome, that’s full mediation. In statistical terms, the direct effect of the predictor on the outcome drops to essentially zero once the mediator is included. The entire influence travels through the middle variable.
More commonly, researchers find partial mediation. This means the mediator explains some of the relationship, but a direct link between the predictor and the outcome still exists, just smaller than before. In the HIV and coping study above, avoidance coping was a partial mediator. It explained a meaningful portion of why stress led to distress, but stress also had a remaining direct effect that didn’t pass through coping. Most real-world psychological processes involve partial mediation because human behavior is complex and rarely funnels through a single pathway.
How Researchers Test for Mediation
The most influential framework for testing mediation comes from a 1986 paper by Baron and Kenny, which laid out four conditions. First, the predictor must be significantly related to the outcome. Second, the predictor must be significantly related to the proposed mediator. Third, the mediator must be significantly related to the outcome even after accounting for the predictor. Fourth, when the mediator is included in the analysis, the predictor’s effect on the outcome should shrink compared to what it was without the mediator in the model.
If all four conditions hold, you have evidence for mediation. The degree to which the predictor’s effect shrinks tells you whether you’re looking at full or partial mediation.
In more recent years, researchers have moved toward a technique called bootstrapping to test whether the indirect effect (the path running through the mediator) is statistically significant. Older methods assumed the indirect effect followed a normal, symmetric distribution, but it usually doesn’t. Bootstrapping generates thousands of simulated samples from the data to build a more accurate confidence interval. The current consensus is that bias-corrected bootstrapping is the best method for testing indirect effects.
How Mediators Differ From Moderators
Mediators and moderators are easy to confuse, but they answer fundamentally different questions. A mediator answers “how” or “why” something happens. A moderator answers “when” or “for whom.”
A moderator changes the strength or direction of a relationship between two variables without explaining the mechanism. For instance, work stress increases drinking problems for people who rely heavily on avoidant coping styles like denial, but stress shows no relationship with drinking for people low in avoidant coping. In this case, coping style is a moderator. It doesn’t sit in the middle of the causal chain. It was already a characteristic of the person before the stress occurred, and it altered how strongly stress affected drinking.
There’s also a key timing difference. A mediator must change after exposure to the predictor variable, because the predictor is supposed to cause changes in the mediator. A moderator, by contrast, is already present before or at the same time as the predictor. Gender, personality traits, and pre-existing beliefs are common moderators precisely because they’re stable characteristics that exist before an intervention or event takes place.
To put both concepts side by side with a clinical example: imagine a therapy program reduces hospitalization rates for people experiencing a first psychotic episode. If the therapy works by improving patients’ ability to manage symptoms, then symptom management is a mediator. If the therapy works better for women than men, then gender is a moderator. The mediator tells you the internal process through which the therapy helped. The moderator tells you who benefited most.
Why Mediators Matter in Practice
Identifying mediators is what moves psychology from knowing that something works to understanding why it works. This distinction has real consequences. If a therapy reduces anxiety, and research identifies that the mediator is a shift in how people interpret ambiguous situations, clinicians can focus more attention on that interpretive shift during treatment. They can design shorter or more targeted interventions that zero in on the active ingredient rather than delivering a broad program and hoping the right component lands.
Mediators also help explain when interventions fail. If a program is supposed to reduce depression by increasing social connection, but participants’ social connection levels don’t actually change, the mediator analysis reveals that the mechanism never engaged. The program didn’t fail because the theory was wrong about social connection mattering. It failed because the program didn’t successfully move the mediator.
In everyday terms, understanding mediators is like understanding why a medicine works rather than just knowing that it does. That deeper knowledge makes it possible to improve the treatment, adapt it for different populations, and predict when it’s likely to succeed or fall short.