Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

re: st: mediation analysis with proportional data


From   "Ariel Linden" <[email protected]>
To   <[email protected]>
Subject   re: st: mediation analysis with proportional data
Date   Thu, 13 Feb 2014 10:11:37 -0500

Hi Sandra,

Indeed, linear models for mediation (as well as everything else) are not
appropriate for binary Y and X variables. I suggest you consider to
different programs: (1) -medeff-, (2) -khb-. Both are user-written programs
and can be downloaded from SJ or SSC (findit medeff, findit khb).

Additionally, you may want to read the paper referenced below in which we
ran simulations under many different X-Y scenarios, and found that both
these programs performed very well.

I hope this helps

Ariel

* Linden A, Karlson KB. Using mediation analysis to identify causal
mechanisms in disease management interventions. Health Services and Outcomes
Research Methodology 2013;13:86-108.

Date: Wed, 12 Feb 2014 17:36:56 +0000
From: "Sandra Virgo" <[email protected]>
Subject: st: mediation analysis with proportional data

Hello all
 
I'd like to know the right type of mediation analysis to use with my
proportional data. 
I am using Stata 13 for Windows. 
 
My original regression analysis of these data used a generalised linear
model with a logit link i.e. a fractional logit as detailed here:
http://www.ats.ucla.edu/stat/stata/faq/proportion.htm 
The DV is a proportion between 0 and 1, as is the IV and some of the
covariates. There are a few actual 0 and 1 values. 
 
I now wish to do a mediation analysis with the same data, with the same IV
and DV, but testing whether one of the covariates (a proportion between 0
and 1) is a mediator. I also wish to keep the other covariates in. I have
read the sem mediation instructions
http://www.ats.ucla.edu/stat/stata/faq/sem_mediation.htm but of course I
have proportional variables so cannot fit a linear model. 
 
Therefore I have used gsem and used a logit link function and binomial
family as in the regression analysis. But a colleague has informed me that
the output is not valid, and he thinks that there isn't a mediation analysis
programme which will suit my data. 
 
Should I give up using gsem for these data? If so, is there a better
mediation programme for my data? 
 
Best Wishes
Sandra Virgo
PhD student
London School of Hygiene & Tropical Medicine


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


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index