Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Testing mediation between 10 independent variables and 1 DV, with "mixed"?

    Dear all,
    I have a dataset and my model assumes that one variable mediates between 10 independent variables and 1 dependent variable. All 12 are observed. Furthermore, data was collected in two countries, so I have some nesting.
    Let's say that X represents my set of 10 independent variables, and Y represents my dependent variable, and M represents my mediator.
    Until now, I've been using the "mixed" regression command 4 times (Y on X, Y on M, M on X, Y on M and X) to determine mediation, as a first cut. But of course, this doesn't calculate the indirect effect properly (i.e. re: Mackinnon Type II errors). And not exactly state-of-the-art.
    As you can see, I'm a bit new to testing mediation in STATA.
    I explored a bit and found the medeff command but it can't handle nested data, right?
    I could really use some advice. (I'd like to stay away from SEM if possible.)
    Regards, -Chih

  • #2
    Hmm... Just now found the ml_mediation command. Anybody know if this is the best route for the situation I've described above?

    Comment


    • #3
      I don't think this is a classic multilevel analysis problem as your data are not "nested" in the sense you imply. You have two countries which should be treated as a fixed effect. Thus, at least to start, standard methods are fine. So, you don't need a random intercept model but you do need to deal with a possible interaction with a categorical variable. Interactions with mediators are usually referred to as "moderated mediation." You can find a nice discussion of this issue at http://www.ats.ucla.edu/stat/stata/faq/modmedcat.htm. That discussion shows you how to use Stata's sureg command to accomplish what I think you need to do. The references there and a little Google work will lead you to many other discussions.

      The medeff program you mention extends the original work on mediation in important ways. The following paper, cited in the medeff help file is a very good discussion of more modern methods of causal inference with respect to mediation. Imai, Kosuke, Luke Keele and Dustin Tingley (2010) A General Approach to Causal Mediation Analysis, Psychological Methods 15(4) pp. 309-334.
      Richard T. Campbell
      Emeritus Professor of Biostatistics and Sociology
      University of Illinois at Chicago

      Comment


      • #4
        Hi Richard,
        Thanks for this advice. I will start my analysis using the standard methods, and follow with the kind of hunting you describe! Regards, -Chih

        Comment

        Working...
        X