I am examining a loyalty program's (LP) effectiveness on customer loyalty across members and non-members. I first hypothesised members will perceive higher levels of switching costs (SC), and exhibit higher levels of attitude, commitment and satisfaction than non-members. This is then followed by next set of hypotheses predicting all those variables will partially mediate the relationship between LP's membership and customer loyalty (e.g. higher SC -> higher customer loyalty).

How would I be able to test my second set of hypotheses (mediation effect)? Do I simply do a multiple regression on SPSS by commanding only outputs of LP members ('rule selection variable') and examine the 'unstandardised coefficients' of all those predictors on customer loyalty?

Given that there's a categorical variable (LP membership) in my hypotheses, are Baron and Kenny (1986)'s mediation steps still applicable in my case?


Baron and Kenny's seminal work still applies, but a more current approach is that taken by e.g. David MacKinnon in his book "Introduction to Statistical Mediation Analysis" and articles.

The essential idea is to look at three regressions:

DV ~ IV + M
IV ~ M

Mediation involves the first equation having a different parameter estimate for the IV than the second equation and there being a relationship in the third equation.

The language isn't entirely consistent and some people use mediation to mean the effect of the IV in the second equation is much less than in the first.

Some people insist that the mediator be on a causal pathway from the IV to the DV.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.