st: Sobel test for mediation with a mixed linear model - GLLAMM???

[Lance Erickson <lance_erickson@byu.edu>]

Also not sure exactly what the question is, but there have been some recent publications about mediation in a multi-level context. See...

MacKinnon, David P. 2008. Introduction to Statistical Mediation Analysis. New York: Lawrence Erlbaum Associates. (Chapter 9 is explicitly on multi-level models)

and

Zhang, Zhen, Michael J. Zyphur, and Kristopher J. Preacher. 2009. "Testing Multilevel Mediation Using Hierarchical Linear Models: Problems and Solutions." Organizational Research Methods 12:695-719. (If I understand your question right, I believe this paper might explain why you are getting confusing results.)

I'm sure there are more that I haven't run across.

Best,
Lance


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Ariel Linden, DrPH
Sent: Wednesday, March 16, 2011 3:35 PM
To: statalist@hsphsun2.harvard.edu
Subject: re: st: Sobel test for mediation with a mixed linear model - GLLAMM???

Hi Nicola,

The traditional Baron-Kenny approach uses an OLS approach, while you are
proposing a multi-level model. First off, you do not provide us with
justification for this approach, so it is difficult to provide feedback. Why
would you not use the OLS approach? Do you have linear outcomes in both Y
and M models? Are you looking at multiple mediation/outcome periods???

If you choose not to use the traditional BK approach, the literature
suggests structural equation modeling or even non/semi-parametric
approaches. 

Again, we can't help you unless you help us...

Ariel  

Date: Tue, 15 Mar 2011 19:55:44 +1100
From: Nicola Man <n.man@unsw.edu.au>
Subject: st: Sobel test for mediation with a mixed linear model - GLLAMM???

Would anyone have a clue on how to conduct a Sobel test for a multilevel
linear model?

One common method is to conduct three separate regressions (according to
Baron and Kenny):
1) y= constant + c'x + bm + error
2) y= constant + cx + error
3) m= constant + ax + error

Where y=dependent variable (i.e. outcome), x=independent variable and
m=mediator variable

Then obtain the product of the coefficients ab and standard errors to
estimate indirect effect.

I can do these mixed effect regressions with xtreg, but the problem is c-c'
should equal ab and I am not getting this using xtreg (so I think this is
not correct).  I suspect it could be something to do with the random error
structure, and wondering if it is possible to get the models simultaneously
fitted in gllamm which might overcome this problem?

Help appreciated! Thanks!

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/