- 1. Confirmatory Factor Analysis in AMOS (Lecture 7) | www.pietutors.com
- 2. What is multilevel structural equation modelling? by Nick Shryane
- 3. SEM Series Part 5a: Confirmatory Factor Analysis
- 4. CFA Demo Part 4 (modification indices and correlated uniquenesses), using AMOS program
- 5. 48th & 49th International Cat Show {February 2015} Promo

This is Lecture 7 from PIE TUTORS's Online Structural Equation Modeling Training Series. This video is brought to you by:- PIE TUTORS a statistical consulting ...

+PIE TUTORSÂ thank you very much for the response. I wanted to know about the possibility in amos. And, regarding your question, i understand it won't serve any purpose. Â

DilshadIn amos we don't have that options.What purpose will the SE of the standardized loadings and the error terms will serve?

Structural equation modelling is a family of statistical models that encompasses regression-, path- and factor analysis. For more methods resources see: ...

very good. does anybody know if I have a model of; all my IVs are in level
1 and 2, than mediators in level 1 and finally, the DV is at level 2 so, in
short is 1,2:1:2. I know is a bit uncommon. please...Thanks.

thank you very much! is there any specific tutorial about multilevel
confirmatory factor analysis?ï»¿

Fantastic ! Very clear explanations... as well as the IRT video

lucid and informative presentation. very helpful.

In this video I show how to do the CFA, including invariance tests, common method bias, and some model fit.

Hello James,
I'm using SPSS Amos for my thesis and part of that is explaining how Amos
(CFA) works. But there is one small thing that I can't seem to find in
scientific articles or books. With a CFA you first look at how the model
fits the data. So you look at the regression and covariance weights. At
around 3:09 you mention that covariance between latent variables above .8
is alarming. Now I don't have any covariance above .8, but I do have to
explain why it matters that the covariance between latent variables
shouldn't be above .8. Could you explain this to me and maybe also give me
a source to cite? It would help me a lot!ï»¿

+Meneer This is because it is easier to theoretically justify these covariances. However, you can covary errors from different factors if you can justify it. Some common reasons for this would be if the two items were immediately beside each other in the survey, or if the items are worded very similarly.

+James Gaskin Thanks a lot, now I understand. I have one more question though, hope you could answers this one as well. When you don't have a great model fit in a CFA, you look at the modification indices for the covariances. But you only covary error terms that are part of the same factor, and not error terms of different factors or error terms with observed/latent variables. Why is this?

+Meneer Not sure of a citation, but the reason is that the covariance represents the shared variance between two factors. If they share that much variance, then it is hard to say that they are two different factors (and not the same factor).

Hi James, thanks a lot. I have a question please, What if the results of
EFA were different from those of theory? we conduct CFA basing on which of
them? Let's assume that we have a concept "attitude" which theoretically
has three dimensions (cognitive, affective & behavioral), but the results
of EFA shows only one factor. In this case should we ignore EFA & treat
Attitude as a second order factor withe three dimensions? Or we should
respect the EFA results & treat it as a single one order factor? I have
such problem in my research & I couldn't figure out what to do.ï»¿

+Mohamed Hasan 2nd order factors sometimes need to be run as their own EFA. This is what I usually do. Then in the CFA, model it as 2nd order.

Dear James
I have two questions
1- I want to apply a pre-post test for my research and I want to assess the
validity and reliability of my questionnaire. how can i do that with EFA?
can I do with CFA like you did in CFA section and do invariance test?
2- in in variance test procedure and configural in variance test ,
shouldn't we assess the model fit for both groups? because you assessed
only women group after groupingï»¿

+hadi yasrebdoost Usually a pre-post is for identifying differences after a treatment. The t-test or ANOVA is best for this. You can also do a homogeneity of variance test (Levene's test) to see if they are the same at the measurement level.

+James Gaskin could we do pre-post test like moderating effect? because if we do like the we should apply an invariance test . and for the last question should we do two efa ( for pretest and post test separately or we should do it for all of the data all together.?

+hadi yasrebdoost Do validity just like you would with any other dataset using EFA, CFA. To do a pre-post test, you would do a t-test or anova. In the invariance test, model fit is shown for the model, rather than for each separate group. See my configural invariance video for this.

Hello, James
Sorry to bother you again, I have 3 group moderator and I was doing the
invariance test but one of the group the "Note for Model" shows me this
message "The following covariance matrix is not positive definite" .. How
Can I deal with it? is there any solution? (size of this group is 84
comparing to the others 125, 111)
Many thanksï»¿

+Riad Cheikh Then you might try Bayesian estimation rather than ML. This might fix it.

the problem just incurred after putting constraint on error variance because it was negative (I constraint the error variance to b 0.001 which made the Standard loading to be 1 for this "1st order factor" )

+James Gaskin I can't constraint the regression weight "aaa" because the factor itself is part of the 2nd order factor and it has its own error variance... is there another solution?

+Riad Cheikh It that is the standardized loading, then it is too high. You might consider constraining the regression weight to equal the other regression weights on the factor by giving the constraint of "aaa" to all of them.

+James Gaskin I did constrained the error variance to 0.0001, which made the the loading of this item to be 1.00 in factor, how can I interpret this loading (isn't it Haywood case problem loading >1.00)?

+James Gaskin Thx a lot James, I really appreciate all your help and answers ... just for double check when you said constrain error variance, it is same technique as constraining Latent variable ( put 0.001 on variance)

+Riad Cheikh When you get that error, it is usually because an error variance is negative. You might need to constrain it to be some small positive number (e.g., 0.001). In the invariance tests, it would be best to exclude the model with all data on it.

for Configural test, is okay to have all the groups + a group with all data on (like; example: Male, female , both gender )?if you fail this test, how can you reported? what is the consequences?I appreciate your help a lot Dear James

Hello,
Can you tell me what source/article you recommend when stating to never
free mi's between uniqueness and factors?
Thanks
Kind regardsï»¿

I really need your help to coach all about Amos

Next Cat Show: We are happy to announce that our next Cat Show which will take place on Saturday, 7th February 2015. The WCF gave us special permission ...

Sign up for free
**to join this conversation on fsaved.com.**

Already have an account? Sign in to comment

Already have an account? Sign in to comment

Dilshad Manzarcommented on 05 Feb 2015