st: CFA

[AL <tayku943@gmail.com>]

Dear Statalisters,

I am currently building a 3-factor measurement model (with a sample
size of 387) and my understanding is that all latent variables
including residuals (measurement errors) must be scaled i.e.
constrained to 1. I think I have managed to do this using the SEM
builder but I am not sure how to incorporate the constrains into
commands.

sem (Trauma -> health, ) (Trauma -> hrtraum, ) (Trauma -> natural, )
(Trauma -> witness, ) (Trauma -> childhood, ) (health <- _cons@1, )
(Adversity -> stresspov, ) (Adversity -> stressconf, ) (stresspov <-
_cons@1, ) (Distress -> k10tot, ) (Distress -> epdstot, ) (Distress ->
ptsdavg, ) (k10tot <- _cons@1, ), covstruct(_lexogenous, diagonal)
latent(Trauma Adversity Distress ) cov( Trauma@1 Trauma*Adversity
Trauma*Distress e.health@1 e.hrtraum@1 e.natural@1 e.witness@1
e.childhood@1 Adversity@1 Adversity*Distress e.stresspov@1
e.stressconf@1 Distress@1 e.k10tot@1 e.epdstot@1 e.ptsdavg@1)
nocapslatent

I wonder if there is another way of doing this e.g.:

sem (Trauma@1 --> health hrtraum natural witness childhood)
        (Adversity@1 --> stressconf stresspov)
        (Distress@1 --> k10tot epdstot ptsdavg)

My question is how do you constrain each measurement error to 1 on
each of the indicators as well as cross-factorial covariance in this
command?

I think this model is misspecified evidenced by poor model fit -- it
failed the chi square test (P=.000) with a value of 6000 (is this
possible?) and other fit statistics (RMSEA, CFI, TLI, SRMR) as well.
Also, the modification indices indicate that adding covariances
between errors for certain indicators might improve the model
substantially (is it possible to have a modification index of over
2000?). Some of the indicators have reasonably strong loadings on each
of the factors (convergent) and low cross factorial loadings
(divergent).

Much appreciated,
Alvin
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/