<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
<HTML>
<HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
<META NAME="Generator" CONTENT="MS Exchange Server version 6.5.7638.1">
<TITLE>Causally dependent mediators</TITLE>
</HEAD>
<BODY>
<!-- Converted from text/plain format -->
<BR>

<P><FONT SIZE=2>Dear subscribers,<BR>
<BR>
<BR>
I'm interested in using mediation for the following conceptual model applied to survey data:<BR>
<BR>
infant deprivation (binary) --> education level (ordered categorical) --> economic position (continous) --> aging well (dichotomous outcome)<BR>
<BR>
I assume that, apart from the mediation effects, each link has direct effects on the outcome.<BR>
<BR>
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.<BR>
<BR>
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):<BR>
<BR>
<BR>
R> Xnames <- c("age","gender", "region")<BR>
R> set.seed(2014)<BR>
R> m.med <- multimed(outcome = "agingwell", med.main = "economic", med.alt = "education",<BR>
+ treat = "deprivation", covariates = Xnames,<BR>
+ data = framing, sims = 1000)<BR>
R> summary(m.med)<BR>
<BR>
<BR>
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"?<BR>
<BR>
I suppose I cannot analyse multilevel data with multimed.<BR>
<BR>
<BR>
<BR>
Thank you very much.<BR>
<BR>
Angel Rodriguez-Laso<BR>
Research Project Manager<BR>
Matia Instituto Gerontológico<BR>
<BR>
<BR>
<BR>
<BR>
</FONT>
</P>

</BODY>
</HTML>