Re: ESEM from MPLUS in lavaan

[yrosseel]

On 02/08/2013 09:01 PM, Pete Parkers wrote:
> Hallo everyone,
>
> do you know whether there is a simple possibility to use lavaan in a way
> it is described in "Marsh, H.W., Muth�n, B., Asparouhov, A., L�dtke, O.,
> Robitzsch, A., Morin, A.J.S., & Trautwein, U. (2009). Exploratory
> Structural Equation Modeling, Integrating CFA and EFA:

I don't think this is possible using lavaan (for now).

Yves.

[Pete Parkers]

I´m looking forward to this version. And once again thank you for your good work and also your really helpful support in this group!

Best

Pete

Am Samstag, 9. Februar 2013 19:51:42 UTC+1 schrieb Yves Rosseel:

On 02/08/2013 09:01 PM, Pete Parkers wrote:
> Hallo everyone,
>
> do you know whether there is a simple possibility to use lavaan in a way

> it is described in "Marsh, H.W., Muth�n, B., Asparouhov, A., L�dtke, O.,

[Katerina Bohle Carbonell]

Hello,

I might have missed it, but is there already a script/tutorial on how to do esem via the lavaan package? Thank you.

Regards,

Katerina

[yrosseel]

On 05/16/2014 06:58 AM, Katerina Bohle Carbonell wrote:
> Hello,
>
> I might have missed it, but is there already a script/tutorial on how to
> do esem via the lavaan package?

I'm afraid not. If somebody else would like to explore this, please do so.

Yves.

[Seongho Bae]

Why we should use the ESEM technique in SEM?

The ESEM core technique is rotation criterion; especially 'geominQ' criterion, I think.

If I have to use the ESEM technique, I rather than use Full-Information Item Factor Analysis in mirt package in R. 

Best wishes,

Seongho Bae.

2014년 5월 16일 금요일 오후 1시 58분 10초 UTC+9, Katerina Bohle Carbonell 님의 말:

[Edward Rigdon]

     The core of the ESEM technique in Mplus is Bayesian relaxation of the strict CFA assumption that “all other loadings are fixed at 0.”  In ESEM, these other loadings are given prior distributions which may be centered on 0 but with small variances so that estimates in individual replications need not be 0 exactly.  Thus ESEM retains the prior information that links particular observed variables with particular factors and generally excludes other links, but with a Bayesian “fudge factor”

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To post to this group, send email to lav...@googlegroups.com.
Visit this group at http://groups.google.com/group/lavaan.
For more options, visit https://groups.google.com/d/optout.

[Phil Chalmers]

I believe nearly the exact same exploratory design can be accomplished in the full information approach via IRT methods, which Seongho mentioned is possible with mirt. If you fit a confirmatory item factor analysis model and observe a poor fit (say via something like the M2 statistic, or similarly with lavaan's WLSMV approach) then you can free up additional slope/loading parameters to see if the fit improves. Improvement can be judged both from a likelihood ratio and model fit standpoint to see if estimating these additional parameters are really helpful. 

However as you said --- and what appears to be the primary inspiration for ESEM in the first place --- is if there is a priori knowledge that these additional exploratory slopes should be close to zero then users can just include an informative Bayesian prior parameter distribution in the model estimation for these parameters (say, a Gaussian one with a mean of 0 and standard deviation of .2, or even stronger depending on the knowledge of the researcher). The same effect is achieved, and you get the so-called 'fudge factor' all while avoiding the ad-hoc rotation methodology implemented in Mplus. Plus, different priors and weights can be used for each parameter, so in that sense it may even be more flexible. So I think Seongho is correct in that respect. Cheers.

Phil

[Seongho Bae]

Dear Edward,

Yes, your observation is right. The core of the ESEM technique in Mplus is a Bayesian relaxation of the strict CFA assumption, theoretically. In practical perspectives, CFA assumption is too strict to estimating loadings, Fit indices, and too high (overestimated) correlation between hypothesized factors for applied psychologists. So, I saw some researchers to want a more flexible estimation of their factor models to construct validation with the ESEM technique in Mplus.

I believe the "mirt" can reach the purpose of the ESEM technique. And I fully agree with Phil's opinion. If I'm the Mplus user, be naturally exultant over, I'll use the ESEM technique. Because, it's very convenient to find a number of factors or make rational arguments for my factor models for the Mplus users. But, I'm the R user. In the R packages, We already have an excellent package for estimating factor structure with exploratory and confirmatory design in the full information approach via IRT methods that is "mirt". It can be a good alternative Mplus which Phil already mentioned. Why we should use the ESEM technique in lavaan? I can agree implantation of the ESEM technique in the lavaan, but I recommend to use the "mirt" package until the state of the current development of the lavaan. The mirt package very useful to know state of Item response characteristic the each item. Knowing Item response characteristics very important in some research topics. Cheers.

Best wishes,

Seongho Bae.

2014년 7월 10일 목요일 오전 12시 44분 34초 UTC+9, Edward Rigdon 님의 말:

[Edward Rigdon]

Seongho--

     No doubt, the rise of multiple indicator modeling tools in R is the most significant software development in this field since the emergence of the Mplus package.  Mplus gave researchers access to cutting edge tools, but the price excludes some researchers from using it.  Developments in the R environment are giving almost every researcher access to most of the same tools.  Two years ago, my doctoral class emphasized Mplus ( on the factor side).  This year, the focus is on tools in R.

     (On the factor modeling side I think the biggest software developments across time have been:

LISREL (factor-based structural equation modeling)

COSAN (nonlinear constraints)

AMOS (true graphical interface, not just a shell around a Fortran routine)

Mplus (merging latent class analysis with factor analysis, plus Bayesian methods and a continuing rush of groundbreaking developments, while emphasizing ease of use)

Tools in R

)

Again, this is only on the factor modeling side.

     If I have left something out, please let me know.  I have been asked to update an encyclopedia entry on "SEM Software" from a few years back, and I would appreciate other perspectives.
--Ed Rigdon

Sent from my iPad

[Seongho Bae]

Dear Edward Ridgon.

Do you know CEFApack? Here you are. It's available to use TARGET rotation. I was checking out your webpage, but It's not listed on your website.

http://faculty.psy.ohio-state.edu/browne/software.php

--
Seongho Bae.

2014년 7월 13일 일요일 오전 2시 47분 3초 UTC+9, Edward Rigdon 님의 말: