I'm running a mediation in SPSS as per Baron and Kenny's guidelines (using regression). X is a dichotomous variable; M and Y are continuous.

  • Step 1) X-->Y (r = .07, p = .03)
  • Step 2) X-->M (r = .45, p < .001)
  • Step 3) M-->Y (r = .31, p < .001)
  • Step 4) X-->Y (beta = -.09, p =.01); M-->Y (beta = .35, p<.001)

Based on previous research, X is consistently linked to Y; though, M is theorized to account for most of this relationship. I'm struggling with the interpretation, specifically why the relationship reversed and remained significant. Does it have something to do with X being dichotomous and/or the initial relationship from X-->Y being small?


1 Answer 1


Have tried doing a mediation analysis with regression? Typically the way mediation is concluded is by running three regressions (instead of correlations):x predicting y, x and mediator predicting y, and x predicting mediator. If you find that x significantly predictors y by themselves, but that it does not significantly predict y when the mediator is added, and x also significantly predicts the mediator, then you can conclude mediation. Also, as far as the interpretation goes it depends on what your variables are, and is highly subjective. I hope that helps at least some.

  • $\begingroup$ Thanks for your comment, costebk08. I got the same results via regression. Based on that definition, I believe I can conclude mediation. However, I am still perplexed on the reversal of the "causal" X variable here. I'm beginning to think it may just be an artifact of how we measured the X variable. However, I'd be interested to hear if 1) anyone else has experienced this and can provide an example, and/or 2) can direct me to literature on findings like this? Thanks. $\endgroup$
    – jsd
    Jun 30, 2015 at 21:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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