 |
 |
Thank you again, Jason. The ab/c ratio is 65.8%. Is this very high? By the way, how can I determine which mediator is most important (or at least more important than another one) if I have 2 mediators? Supposed we did collect M' data If we conducted 2 mediation analyses, say (a) X--Y mediated by M and (b) X--Y mediated by M', and the ab/c ratio in (a) is higher than that in (b), we can then say that, as a mediator, M is more important than M'? Best, Mark 2013/1/17 Jason Steffener <[log in to unmask]> > Just to be clear: > > c' is the direct effect of X on Y > c is the total effect of X on Y > ab is the indirect effect of X on Y > > c = c' + ab > > The test of whether age-related differences in behavior are > significantly related to age-related differences in fMRI activity is > the test of whether ab is significant. The best way to test this is > with BCa confidence intervals on a bootstrap test. > > If ab is significant then yes you can say that X has and significant > indirect effect on Y via M. You can then look at the ratio of ab/c to > determine how much of the age effect is via M. I do not think that you > have any grounds to state that M is the most important mediator unless > the ab/c ratio is very high. > > Jason > > > > On Thu, Jan 17, 2013 at 12:59 PM, Tseng Mark <[log in to unmask]> wrote: > > Sorry a correction: c' is not zero. c' = –0.029. > > > > > > 2013/1/17 Jason Steffener <[log in to unmask]> > >> > >> Dear Mark, > >> I think you should avoid the discussion of full versus partial > >> altogether in relation to other unmeasured variables because this > >> distinction is based solely on a significance threshold. Paths c and > >> c' are not needed for their to be a significant indirect effect (ab) > >> and it is the significance of your indirect effect that is of interest > >> to you. This is another reason for avoiding the full/partial > >> discussion. > >> > >> > >> To address your specific question, is c' zero? Or is it just > >> not-significant? If it does not go to zero after accounting for M, > >> then there is still an effect of X on Y in the presence of M. It may > >> be that M' may be that effect. So I do not think you have any basis > >> for saying that M' has no effect on X, M or Y especially since M' is > >> not measured. > >> > >> best, > >> Jason. > >> > >> On Thu, Jan 17, 2013 at 12:20 PM, Mark <[log in to unmask]> wrote: > >> > Results showed that path > > > > >