Latent and observed variables as "regressors"
Posted on
suresh
Joined: 03/15/2010
Forums
Hello,
I have a set of 4 correlated continuous variables that I have observed, and also a fifth binary observed variable that has an effect on 2 of the 4 continuous variables.
I want to test the hypothesis that a single latent factor plus the fifth binary observed variable are together sufficient to explain the covariance matrix for the 4 correlated continuous variables.
How can I test this in OpenMx ?
Also, what category does this fall under (so I can google for it or look it up in textbooks) ? Is there a place where I could read up on this (ideally with examples from OpenMx, or LISREL or AMOS) ?
Thanks much !!!!!!!!
Suresh
Take a look at the books
Take a look at the books under the Resources tab at the top of this page.
I'd recommend reading Maruyama (1998) Basics of Structural Equation Modeling or Marcoulides & Schumacker (1996) Advanced Structural Equation Modeling: Issues and Techniques. That should get you started with the ideas.
Another way to get started is by looking at my class notes at http://openmx.psyc.virginia.edu/forums/openmx-help/teaching-sem-using-openmx/uva-introduction-sem-spring-2010
Log in or register to post comments
In reply to Take a look at the books by Steve
Thanks much for the links
Thanks much for the links !!
I managed to figure out how to do the analysis (in R with the sem package, and in LISREL) with just the one latent factor. What I am struggling with is to figure out the right way to now include the binary observed variable as an explanatory factor in the path diagram..
I will read as well.
Very best, Suresh
Log in or register to post comments
In reply to Thanks much for the links by suresh
The analysis is not all that
The analysis is not all that difficult in OpenMx. I would take the script from the OpenMx homepage http://openmx.psyc.virginia.edu/ and add a definition variable for the binary explanatory variable. To do this, take a look at, e.g., http://openmx.psyc.virginia.edu/docs/OpenMx/latest/DefinitionMeans_Path.html and see if you can construct a model in which the binary variable has an effect on the mean of each of the observed variables.
Log in or register to post comments
In reply to The analysis is not all that by neale
Thank you, yes, that seems
Thank you, yes, that seems like what I am looking for.
I also finally need to figure out how to do some missing-data imputation (like FIML in LISREL); I saw some previous posts that advised that other R packages for imputation could be combined with OpenMx for this. I will try to follow those leads..
Very best, Suresh
Log in or register to post comments
In reply to Thank you, yes, that seems by suresh
Your reply is a little
Your reply is a little confusing; OpenMx has built-in FIML functions, and FIML does not require any imputation. It simply calculates the likelihood (on a case-by-case basis) of those variables that have been observed. I hope that this is clear from the examples, but please let us know if it is not.
Log in or register to post comments
In reply to Your reply is a little by neale
A question like that came up
A question like that came up at a talk recently, from an epidemiologist used to using regression: I think from that framework, people often think that FIML involves imputation, and ask "can I see what it looked like in the raw data?" not knowing that it was the raw data...
Log in or register to post comments
In reply to A question like that came up by tbates
Yes, mea culpa ! I spoke to a
Yes, mea culpa ! I spoke to a genuine expert on factor analysis and in our brief discussion, he told me to use FIML to handle the missing data in LISREL, and I then mistakenly assumed that it uses some "reasonable" method to "fill in" the missing data.
I will go through the examples and other material now !
Log in or register to post comments
In reply to Yes, mea culpa ! I spoke to a by suresh
thank for the updates..
thank for the updates..
Log in or register to post comments