re: st: Mediation for a left-censored dependent variable and ordinal mediator variable

["Ariel Linden, DrPH" <ariel.linden@gmail.com>]

Hi Vijay,

I believe you are using the wrong modeling approach here. You need a
"specialized" program to correctly model an ordinal mediator and censored
outcome. Using -sem- or any other regression approach that assumes linearity
in these two models will likely lead to misspecification and thus erroneous
results.

You can consider using a user written program -khb- (findit khb)  than can
handle ordinal mediators, but does not appear to handle censored outcomes at
this time. Similarly, you could consider using the R version of -medeff-
(findit medeff), that can also handle a large array of mediator and outcome
data types (the Stata version is limited to regress, logit and probit). I am
not sure if this program can handle censored cases.

Perhaps another approach, albeit a manual process at this point, would be to
use a potential outcomes approach via stratification, weighting, or
principal stratification (see references below). I am particularly enamored
with Hong's stratification approach, and I think it would fit your needs the
best. However, it (like the rest of the approaches referenced below) is
somewhat labor intensive....

I hope this helps

Ariel

Hong, G. (2010). Ratio of mediator probability weighting for estimating
natural direct and indirect effects. 2010 Proceedings of the American
Statistical Association, Biometrics Section [pp.2401-2415], Alexandria, VA:
American Statistical Association.

Imai, K., Keele, L., & Tingley, D. (2010). A general approach to causal
mediation analysis. Psychological Methods, 15(4), 309-334.

Imai, K., Keele, L., & Yamamoto, T. (2010). Identification, inference and
sensitivity analysis for causal mediation effects. Statistical Science,
25(1), 51-71.  

Pearl, J. (2010). The mediation formula: A guide to the assessment of causal
pathways in non-linear models. Los Angeles, CA: University of California,
Los Angeles. Technical report R-363, July 2010.

Peterson, M. L., Sinisi, S. E., & van der Laan, M. J. (2006). Estimation of
direct causal effects. Epidemiology, 17(3), 276-284.

VanderWeele, T.J. (2009). Marginal structural models for the estimation of
direct and indirect effects. Epidemiology, 20, 18-26.

Jo, B. (2008). Causal inference in randomized experimentswith mediational
processes. Psychological Methods, 13, 314-336.

Jo, B., Stuart, E. A., MacKinnon, D. P., & Vinokur, A. D. (2011). The use of
propensity scores in mediation analysis. Multivariate Behavioral Research,
46, 425-452.



Date: Sun, 12 Feb 2012 23:00:57 -0500
From: Vijay Sampath <vijsampath@hotmail.com>
Subject: st: Mediation for a left-censored dependent variable and ordinal
mediator variable

I am trying to run a mediation analysis in Stata 12 using the "sem" command.
The dependent variable is a continuous variable which is left-censored. The
mediator is a ordinal variable with 5 levels and the independent variable is
a log-converted continuous variable.
I am getting significant results when I run the regressions separately: (a)
the tobit regressions on the dependent variable, and (b) ordered logit
regressions with the mediator as the dependent variable. I would like any
thoughts as to how to combine them. I did a Google search, which pointed me
to the MPlus command.
Thanks,
VJ 		

*
*   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/