On Methods: What are mediation and moderation?

Screen Shot 2013-03-01 at 8.46.07 AM



By Adar Ben-Eliyahu

One of the burning questions for practitioners, researchers, and policy makers is – how does mentoring work? We have all seen that mentoring is correlated with good outcomes, but correlation does not explain what leads to those outcomes, why or how. A correlation can tell us if the relations are positive, where higher quality mentoring leads to more adaptive outcomes, or negative, where the reverse is true.

Mediation – Why? How?

Many of us are interested in why or how mentoring leads to adaptive outcomes. Maybe it’s because the mentee has learned a valuable skill, has changed the way s/he thinks about the future, or has begun to think differently about the other adults in his or her life. These factors, which help explain the association between the input to the output, are called mediators and could be presented like this:


Moderation – When? For whom?

We might also be interested when or for whom these associations matter most. For example, our findings might be slightly different for girls and boys. This would mean that gender moderates the correlational model we are testing.


Examples from Mentoring Research


Nearly 15 years ago, Rhodes, Grossman, and Resch (2000) tried to understand what mediates the effect of mentoring on grades and school attendance. Their theory was that mentoring “works” because it helps improve mentees’ relationships with their parents. A simplified model of this finding is:


With improved parental relationships, children/youth experience higher levels of self-worth and competence, and this in turn leads to improved academic outcomes such as grades and attendance. Parental relationships are the mediator that connects mentoring to improved academic outcomes.


Many times when we find significant and meaningful relations between characteristics/variables, we are interested in whether this  holds true across different groups. Moderation answers the question of when and for whom.This question was addressed by Grossman, Chan, Rhodes, and Schwartz (2012)  in their study of match length and rematching in mentoring.

Specifically, Grossman et al. investigated whether the well-known associations between length of match and academic outcomes would hold when youth who had a broken match were re-matched. In other words, does being re-matched versus being in the same match continuously, influence the correlational statistical model, presented in the following figure:


Grossman et al. found that rematching moderated the relations between mentoring relationship length and academic outcomes, such that improved academic achievement was related to mentoring length only for youth in intact matches.


As we investigate mentoring, we can use different statistical techniques to understand why, how, when, and for whom mentoring works. It is important to consider these mediation and moderation effects so that we can not only understand the processes at play, but make recommendations for tailoring mentoring programs and supports for mentors.