On 27 March 2011 12:12, jouba <[email protected]> wrote:
> I am a new user of the function sem in package sem and lavaan for
> structural
> equation modeling
> 1. I dont know what is the difference between this function and CFA
> function, I know that cfa for confirmatory analysis but I dont know what
> is the difference between confirmatory analysis and structural equation
> modeling in the package lavaan.
>
Confirmatory factor analyses are a class of SEMs. All CFAs are SEMs, some
SEMs are CFA. Usually (but definitions vary), if you have a measurement
model only, that's a CFA. If you have a structural model too, that's SEM.
If you don't understand this distinction, might I suggest a little more
reading before you launch into the world of lavaan? Things can get quite
tricky quite quickly.
> 2. I have data that I want to analyse but I have some missing data I must
> to
> impute these missing data and I use this package or there is a method that
> can handle missing data (I want to avoid to delete observations where I
> have
> some missing data)
>
No, you can use full information maximum likelihood estimation (= direct ML)
to model data in the presence of missing data.
> 3. I have to use variables that arnt normally distributed , even if I
> tried
> to do some transformation to theses variables t I cant success to have
> normally distributed data , so I decide to work with these data non
> normally distributed, my question my result will be ok even if I have non
> normally distributd data.
>
Depends. Lavaan can do things like Satorra-Bentler scaled chi-square, which
are robust to non-normality, and corrects your chi-square for (multivariate)
kurtosis.
> 4. If I work with the package ggm for separation d , without latent
> variables we will have the same result as SEM function I guess
>
Not familiar with ggm. I'll leave that for someone else.
> 5. How about when we have the number of observation is small n, and what
> is
> the method to know that we have the minimum of observation required??
>
>
>
>
Another very difficult question. Short answer: it depends. Sometimes you
see recommendations based on the number of participants per parameter, which
is usually around 5-10. These are somewhat flawed, but it's better than
nothing.
Again, I should reiterate that you have a hard road in front of you, and it
will be made much easier if you read a couple of introductory SEM texts,
which will answer this sort of question.
Jeremy
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
[[alternative HTML version deleted]]
______________________________________________
[email protected] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
