[Ctsem-mail] Some ideas of what might be going wrong?

[E. L. (Lisanne) de Moor]

Dear all,

For a project, I'm examining the bidirectional self-esteem and satisfaction
with social contacts. Now, for a robustness check I wanted to see if my
findings also (generally) hold for a more specific kind of satisfaction --
romantic satisfaction. For this, I could use the same dataset, the same
self-esteem data, and only the satisfaction variable (which is on the same
scale as the original and also has a similar mean and variance) is
different.

However, when running my models on this new variable, I get unlikely
estimates for the random intercept model (the model with no random
intercept, so basically the "CLPM" model, seems fine). Specifically, in the
drift matrix the autoregressive effect of self-esteem and, to a lesser
extent satisfaction, is much lower. The cross-lagged effects seem to be
inflated (to .3 and .5 where they were < .05 in the original model).
Finally, the manifest variance for satisfaction is much smaller than it was
in reality (now only < .001). No error messages are produced.

In short, I expected this model to run pretty similar to the original model
as only one variable is different, but ended up with a model that does not
seem to fit at all. I have examined my data several times, but cannot
detect what might be the cause of this weird fit. Therefore, I was
wondering if any of you have any ideas on where the problem may be located
or what I could do to identify the problem.

Many thanks in advance!

Best,
Lisanne de Moor
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/ctsem-mail/attachments/20200508/0513130c/attachment.html>