[Mediation-information] Causally dependent mediators

[Angel Rodriguez]

Dear subscribers,


I'm interested in using mediation for the following conceptual model applied to survey data:

infant deprivation (binary) --> education level (ordered categorical) --> economic position (continous) --> aging well (dichotomous outcome)

I assume that, apart from the mediation effects, each link has direct effects on the outcome.

I've read in "Unpacking the Black Box of Causality: Learning about Causal Mechanisms from Experimental and Observational Studies" (Imai et al 2011), that studying education as the main mediator wouldn't be a problem, because the sequential ignorability assumption would be met due to the fact that the second mediator is downstream (figure 6d of that paper). That would be my first model.

What should I do to know if the effect of education is partly or completely mediated by economic position? I understand I cannot use economic position as a main mediator in the model deprivation --> economic position --> aging well, because there would be an upstream mediator (education), which violates the sequential ignorability. I should use multimed then (my second model): 


R> Xnames <- c("age","gender", "region")
R> set.seed(2014)
R> m.med <- multimed(outcome = "agingwell", med.main = "economic", med.alt = "education",
+ treat = "deprivation", covariates = Xnames,
+ data = framing, sims = 1000)
R> summary(m.med)


If I get a large mediating effect of "education" in my first model and a negligible effect of "economic" in the second, could I conclude that all the mediating effect of "education" on the relationship between "deprivation" and aging well is driven by "economic"?

I suppose I cannot analyse multilevel data with multimed.



Thank you very much.

Angel Rodriguez-Laso
Research Project Manager
Matia Instituto Gerontológico




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/mediation-information/attachments/20140826/d4eb975b/attachment.html>