Home Forums Default Forum Mediation

Viewing 2 posts - 1 through 2 (of 2 total)
  • Author
  • #1696


    I am currently undergoing undergraduate research and was recommended to look up mediation in Andy’s book (unhelpfully I wasn’t told which one – I only have ‘Discovering Statistics Using SPSS – 3rd ed, which ,whilst usually brilliant, doesn’t help! – and I am a little confused, so if anyone could help that would be great!

    My limited understanding (from the omniscient google ) is that a mediator variable is a sort of predictor or casual variable between an IV and a DV. – X leads to Y which in turn leads to Z ergo X leads or is related to Z. Is this right?

    Secondly, my research involves both a experiment and a correlation, can I use a mediator variable in both types of research?

    Also, how would I report or treat a Mediator Variable in my design and results sections? Would I  treat it as an IV? What type of design would it be?  I currently have it as;

    A 2x2factorial within groups design, this includes my main iv and the iv that I want to have as a mediator variable. So, I have 2 Ivs, each with 2 levels. In the correlation exp I will have only 1 IV, whilst in the exp condition, I will have 2 Dvs but that was all written before I was told to look at mediator variables so now I am confused do I just have a 2 factor within groups design with 1 mediator variable? Is it acceptable to write it like that?

    Finally, what would you recommend be the best way to analyse this? By an ANOVA, regression or something else? I will be using SPSS.

    I have looked at Baron and Kenny (1986) as well as David Kenny’s site- but it hasn’t helped my confusion!

    Sorry for the long, and probably confusing post, I lack the ability to condense! I will of course ask the teacher concerned but they can be rather lax at replying to emails and never seem to be in their office!!!

    Many thanks,


    Rafael Garcia

    A mediator is a suspected causal intermediate link between two variables. So, yes you’re right.

    Mediation is tested in the analyses, and can be tested in experiments or observational studies.

    You treat is as both a DV and and IV (see below).

    I don’t entirely understand your description of your design. You have two binary predictors and a continuous outcome?

    (I might get grief for this) What I would suggest is a series of sequential regressions.

    If you follow a strict Baron & Kenny (1986) -like procedure, you’d do the following:

    Step 1:  Show that the initial variable is correlated with the outcome.  Use Y as the criterion variable in a regression equation and X as a predictor (estimate and test path c in the above figure). This step establishes that there is an effect that may be mediated.

    Step 2: Show that the initial variable is correlated with the mediator.  Use M as the criterion variable in the regression equation and X as a predictor (estimate and test path a).  This step essentially involves treating the mediator as if it were an outcome variable.

    Step 3:  Show that the mediator affects the outcome variable.  Use Y as the criterion variable in a regression equation and X and M as predictors (estimate and test path b).  It is not sufficient just to correlate the mediator with the outcome; the mediator and the outcome may be correlated because they are both caused by the initial variable X.  Thus, the initial variable must be controlled in establishing the effect of the mediator on the outcome


    I’d suggest doing Steps 2 and 3.

    (2) establishes an effect from X->M, while (3) establishes an effect of M->Y after controlling for any effect of X. If a researcher insists on using (1) then there may be a problem in the fully-mediated case of a small indirect effect. 

    Dummy code your two binary predictors (that just means assign one value 0 and the other 1, and in SPSS make them Interval), and do the following two regressions:

    Mediator_predictor = Main_predictor 

    Outcome = Mediator_predictor + Main_predictor (using a Type 1 or sequential Sums of Squares).

    Admittedly, this will bias your regression weights in favor of the mediator, but I consider that appropriate. You should report the beta (standardized) regression weights, they are more easily interpreted. I hope that helps…


Viewing 2 posts - 1 through 2 (of 2 total)
  • The forum ‘Default Forum’ is closed to new topics and replies.