SEM with Dichotomous Dependent Variable

[Ricardo Gonçalves Silva]

Dear Users,

My question concerns with the estimation of a recursive two-equation SEM
model, where one of the dependent variables is dichotomous. I search the
groups' archives and found some useful guidelines, however, I also would
like to known if LISREL can handle this kind of models and, if not, there
is some way to impose a restriction on the parameters set in order to get
predicted values inside the [0,1] interval.

Thanks

Ricardo

[Cameron McIntosh]

Hi Ricardo,

What exactly is your dichotomous DV? If it is truly categorical (e.g., disease present versus disease absent), then no, LISREL does not really have appropriate methods for handling this type of model. On the other hand, if your DV is a coarsely categorized version of a continuous variable (e.g., acceptable versus unacceptable), then LISREL does offer some ways around the problem.

Thanks,

Cam

Cameron N. McIntosh, MA
Analyst
Health Analysis and Measurement Group / Groupe d'analyse et de mesure de la santé
Statistics Canada / Statistique Canada
24-A R.H. Coats Building
Ottawa, ON
K1A 0T6

Tel: (613) 951-3725
Fax: [phone]



-----Original Message-----
From: Structural Equation Modeling Discussion Group [mailto:***@BAMA.UA.EDU] On Behalf Of Ricardo Gonçalves Silva
Sent: August 12, 2005 1:52 PM
To: ***@BAMA.UA.EDU
Subject: SEM with Dichotomous Dependent Variable


Dear Users,

My question concerns with the estimation of a recursive two-equation SEM model, where one of the dependent variables is dichotomous. I search the groups' archives and found some useful guidelines, however, I also would like to known if LISREL can handle this kind of models and, if not, there is some way to impose a restriction on the parameters set in order to get predicted values inside the [0,1] interval.

Thanks

Ricardo

[Michael Gillespie]

Guys,

There is an alternative way of viewing a dichotomous dv, particularly one related to disease. That, of course, is when the disease occurs--for example, age at first diagnosis. (One could treat the onset of other diseases and/or subsequent death from other diseases as competing risks.) This kind of analysis is the mainstay of epidemology. You cannot deal a dv conceptualized this way in LISREL, but you can in Mplus (and, probably GOLD) if you're willing to treat time as discrete rather than continuous. The advantages of Mplus are: (a) possibly modelling the IV's, particularly if one wants to treat them as fallible measures of factors and/or (b) using both the IV's and the DV to define latent classes. I don't know whether you would regard this kind of analysis as structural equation modelling. I would.

Mike Gillespie

________________________________

From: Structural Equation Modeling Discussion Group on behalf of Cameron McIntosh
Sent: Fri 8/12/2005 2:12 PM
To: ***@bama.ua.edu
Subject: Re: SEM with Dichotomous Dependent Variable



Hi Ricardo,

What exactly is your dichotomous DV? If it is truly categorical (e.g., disease present versus disease absent), then no, LISREL does not really have appropriate methods for handling this type of model. On the other hand, if your DV is a coarsely categorized version of a continuous variable (e.g., acceptable versus unacceptable), then LISREL does offer some ways around the problem.

Thanks,

Cam

Cameron N. McIntosh, MA
Analyst
Health Analysis and Measurement Group / Groupe d'analyse et de mesure de la santé
Statistics Canada / Statistique Canada
24-A R.H. Coats Building
Ottawa, ON
K1A 0T6

Tel: (613) 951-3725
Fax: [phone]



-----Original Message-----
From: Structural Equation Modeling Discussion Group [mailto:***@BAMA.UA.EDU] On Behalf Of Ricardo Gonçalves Silva
Sent: August 12, 2005 1:52 PM
To: ***@BAMA.UA.EDU
Subject: SEM with Dichotomous Dependent Variable


Dear Users,

My question concerns with the estimation of a recursive two-equation SEM model, where one of the dependent variables is dichotomous. I search the groups' archives and found some useful guidelines, however, I also would like to known if LISREL can handle this kind of models and, if not, there is some way to impose a restriction on the parameters set in order to get predicted values inside the [0,1] interval.

Thanks

Ricardo