Re: Re: st: sem mediation analysis - categorial mediator

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

Hi Sabrina,

You can perform mediation analysis in Stata for categorical mediatiors using
two different user-written programs. (1) -khb- (findit khb) is based on the
traditional sem framework, and -medeff- (findit medeff) uses the potential
outcomes approach.

In addition you may find this article helpful in describing the issues that
John and Billy highlight in their previous responses to you.

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.

I hope this helps

Ariel

Date: Wed, 19 Mar 2014 12:30:45 -0500
From: William Buchanan <william@williambuchanan.net>
Subject: Re: st: sem mediation analysis - categorial mediator

The big problem with nominal/ordinal mediators in SEM is related to the
scaling and interpretation of the path coefficients that are estimated and
their comparability to the other coefficients in the model.  Mplus offers
two different solutions, but the underlying premise is generally the same
for both and involves a latent indicator that represents the continuous
probability for membership in the different levels/categories of the
observed mediator.  If you create a latent indicator for your mediator you
should be in better shape, but Stata 12 only offers ADF as an estimator for
categorical DV.  If you have access to Stata 13, you could specify the model
using the appropriate GLM link and family functions to better estimate the
model parameters.  

Also, if you only have a the mediator between the IV and DV your path
coefficient from the mediator to the DV will be biased.  Think of it in the
same context as instrumental variables where you need an additional
exogeneous variable to identify the model and deal with the endogeneity
problem.

HTH,
Billy

Sent from my iPhone

> On Mar 19, 2014, at 11:55, John Antonakis <John.Antonakis@unil.ch> wrote:
> 
> Hi:
> 
> I think that the help file of the author of -cmp- is very useful and lays
out how to use cmp. After you go through it, why don't you try and estimate
the model and then show us what you did (your code) and what errors you got
and then maybe someone can help you.
> 
> Best,
> J.
> 
> __________________________________________
> 
> John Antonakis
> Professor of Organizational Behavior
> Director, Ph.D. Program in Management
> 
> Faculty of Business and Economics
> University of Lausanne
> Internef #618
> CH-1015 Lausanne-Dorigny
> Switzerland
> Tel ++41 (0)21 692-3438
> Fax ++41 (0)21 692-3305
> http://www.hec.unil.ch/people/jantonakis
> 
> Associate Editor:
> The Leadership Quarterly
> Organizational Research Methods
> __________________________________________
> 
>> On 19.03.2014 17:40, reitersa@cms.hu-berlin.de wrote:
>> thank you very much for your answer!
>> 
>> I installed cmp, but I don't understand how it works...
>> 
>> This is my STATA command for my (mediation analysis (without any control
>> variables):
>> 
>> sem (lebenszufriedenheit <- bildungsreich aeq_einkommen teilzeit rente
>> nicht_erwerb) (aeq_einkommen <- bildungsreich) ///
>> (teilzeit rente nicht_erwerb <- bildungsreich) if e(sample), stand
>> vce(cluster hid)
>> 
>> The last path is the one with the categorial dependent (mediator)
variable.
>> 
>> For me it would be enough to find literature where it is written that
>> STATA 12 cannot handle categorial mediators. Maybe you know something?
>> 
>>    Thank you very much, Sabrina
>> 
>> 
>> 
>>> Hi:
>>> 
>>> If your mediator is endogenous, which it probably is, then you need to
>>> use an instrumental variable estimator that can handle the type of model
>>> you have.  -sem- or -gsem- can do this for you. If you model two
>>> "discrete" equations where the disturbances are assumed to be
>>> orthogonal, the model is potentially misspecified.
>>> 
>>> Check out -cmp-, available from -ssc-; it can handle multinomials as
>>> mediators (multinomial probit in fact, that does not differ much from
>>> logistic).
>>> 
>>> Else (not a good thing to say here), Mplus can handle such models with
>>> the WLSMV estimator:
>>> 
>>> Muthén, B. O., du Toit, S. H. C., & Spisic, D. in press. Robust
>>> inference using weighted least squares and quadratic estimating
>>> equations in latent variable modeling with categorical and continuous
>>> outcomes. Psychometrika.
>>> 
>>> HTH,
>>> J.
>>> 
>>> __________________________________________
>>> 
>>> John Antonakis
>>> Professor of Organizational Behavior
>>> Director, Ph.D. Program in Management
>>> 
>>> Faculty of Business and Economics
>>> University of Lausanne
>>> Internef #618
>>> CH-1015 Lausanne-Dorigny
>>> Switzerland
>>> Tel ++41 (0)21 692-3438
>>> Fax ++41 (0)21 692-3305
>>> http://www.hec.unil.ch/people/jantonakis
>>> 
>>> Associate Editor:
>>> The Leadership Quarterly
>>> Organizational Research Methods
>>> __________________________________________
>>> 
>>>> On 19.03.2014 08:04, reitersa@cms.hu-berlin.de wrote:
>>>> I did an mediation analysis in STATA 12 with sem. My dependent variable
>>>> is
>>>> "life satisfaction" (continuous) and my mediator variable is
"employment
>>>> status" (categorial (four categories)). (And I use several independent
>>>> variables.)
>>>> Do you know how STATA is handling the path from the independent
variable
>>>> on the mediator? If I would not test this path with structural equation
>>>> modeling I had to test it with a multinomial logistic regression. So I
>>>> do
>>>> interpret my output coefficients of that specific path as relative risk
>>>> ratios or like the regression coefficients of the other paths (e.g.
>>>> mediator -->DV)?
>>>> Maybe you can recommend me some literatur on that topic. That would be
>>>> great!
>>>> 
>>>>    Thank you very very much, Sabrina


*
*   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/