You are here

Joint test of indirect effect x moderator

5 posts / 0 new
Last post
pascofearon's picture
Offline
Joined: 05/07/2013 - 10:57
Joint test of indirect effect x moderator

Hi Mike,
Hope you're well. I just wondered whether you had a suggestion for getting a single test of a moderator on an indirect (a*b) path using osmasem. Right now I can easily get the indirect effect itself, and estimates of it at high- and low- levels of the moderator, plus regressions of the respective paths on the moderator, but not a joint test of the indirect effect's variability as a function of the moderator... any tips?

Thanks!

Best wishes,

Pasco

Mike Cheung's picture
Offline
Joined: 10/08/2009 - 22:37
Hi Pasco,

Hi Pasco,

I don't think that this is possible as ab is not a parameter in the model. Moderators can only be used to predict parameters (a and b), not ab.

Best,
Mike

pascofearon's picture
Offline
Joined: 05/07/2013 - 10:57
Joint test of indirect effect x moderator

Thanks Mike - that's what I suspected.

Thanks for the speedy reply!

Best,

Pasco

AdminNeale's picture
Offline
Joined: 03/01/2013 - 14:09
It can be done, IMO

Hi

It seems to me that one could moderate the entire indirect effect, ab. Usually we think about moderating individual paths, so a definition variable, say age, could affect parameter a, replacing it by a + beta_a*age. Path b could be similarly moderated to become b + beta_b*age. Equating the two moderations beta_a and beta_b would be somewhat equivalent to moderating the product ab by age. Note, however, the existence of other components to the mediation: (a+beta_a*age)(b + beta_b*age) is now the mediation effect (which will vary across values of age). The expansion is a*b + beta_a*age*b + a*beta_b*age + beta_a*beta_b*age*age, so it's partly quadratic with respect to age whenever both betas are non-zero. Even if the betas are equated.

Mike Cheung's picture
Offline
Joined: 10/08/2009 - 22:37
Hi Michael,

Hi Michael,

Yes, the path coefficients a and b can be separately modeled by a moderator (m) as you suggested. But this model is different from ab = beta0 + beta1m. I have tried hard to figure out if it is possible to do the latter one.

Mike