| |
 |
|
|
Science Forum Index » Statistics - Education Forum » Question re: inadmissable solutions in AMOS
Page 1 of 1
|
| Author |
Message |
| srbraithwaite@gmail.com |
 Posted: Mon Mar 17, 2008 8:22 am |
|
|
|
Guest
|
I'm attempting to fit growth curve models in AMOS for variables with
three time points using condition (which is dichotomous) as an
exogenous predictor. I believe I have everything set up so the model
should work. I adjusted the ICPT and SLOPE so they are free to vary
(since estimating means and intercepts in AMOS changes the mean of
latent vars to 0). The loadings from ICPT to the 3 observed variables
are all set to 1 and the loadings from SLOPE to the indicators are
fixed at the number of weeks from baseline (0, 6 and 40). I adjusted
the observed indicators so their intercept is 0 and their disturbance
terms are constrained to be equal.
Here's the problem. When I attempt to correlate the disturbance terms
for ICPT and SLOPE, it says that the solution is inadmissible because
the matrix between the disturbance terms is not positive definite.
When I remove the correlation between the disturbance terms for ICPT
and SLOPE, it says the solution is inadmissible because the variance
(or disturbance term) for SLOPE is negative. This has happened for
more than one of the variables I am attempting to model.
What am I doing wrong? |
|
|
| Back to top |
|
| Ryan |
| Posted: Mon Mar 17, 2008 1:00 pm |
|
|
|
Guest
|
On Mar 17, 2:22 pm, "srbraithwa...@gmail.com"
<srbraithwa...@gmail.com> wrote:
Quote: I'm attempting to fit growth curve models in AMOS for variables with
three time points using condition (which is dichotomous) as an
exogenous predictor. I believe I have everything set up so the model
should work. I adjusted the ICPT and SLOPE so they are free to vary
(since estimating means and intercepts in AMOS changes the mean of
latent vars to 0). The loadings from ICPT to the 3 observed variables
are all set to 1 and the loadings from SLOPE to the indicators are
fixed at the number of weeks from baseline (0, 6 and 40). I adjusted
the observed indicators so their intercept is 0 and their disturbance
terms are constrained to be equal.
Here's the problem. When I attempt to correlate the disturbance terms
for ICPT and SLOPE, it says that the solution is inadmissible because
the matrix between the disturbance terms is not positive definite.
When I remove the correlation between the disturbance terms for ICPT
and SLOPE, it says the solution is inadmissible because the variance
(or disturbance term) for SLOPE is negative. This has happened for
more than one of the variables I am attempting to model.
What am I doing wrong?
You might find the following links helpful in your search to solve the
problem:
http://www2.gsu.edu/~mkteer/npdmatri.html
http://oregonstate.edu/dept/hdfs/papers/lgcgeneral.pdf |
|
|
| Back to top |
|
| Richard Ulrich |
| Posted: Mon Mar 17, 2008 11:00 pm |
|
|
|
Guest
|
On Mon, 17 Mar 2008 11:22:03 -0700 (PDT), "srbraithwaite@gmail.com"
<srbraithwaite@gmail.com> wrote:
Quote: I'm attempting to fit growth curve models in AMOS for variables with
three time points using condition (which is dichotomous) as an
exogenous predictor. I believe I have everything set up so the model
should work. I adjusted the ICPT and SLOPE so they are free to vary
(since estimating means and intercepts in AMOS changes the mean of
latent vars to 0). The loadings from ICPT to the 3 observed variables
are all set to 1 and the loadings from SLOPE to the indicators are
fixed at the number of weeks from baseline (0, 6 and 40). I adjusted
the observed indicators so their intercept is 0 and their disturbance
terms are constrained to be equal.
Just a couple of comments that suggest my ignorance
about the problem, but which may inspire someone else -
I don't know about the tool you are using, AMOS, but "three
time points" is pretty minimal for doing *any* growth
modeling.
This is a poor model/design in one regard - If (0, 6, 40) weeks
were reasonably expected to show equal-interval effects,
distances should be modeled as "equal" and not as the
number of weeks. For my curiosity -- What are you hoping
to get from this that you would not get from linear slopes
in ANOVA, for each of the several variables?
Quote:
Here's the problem. When I attempt to correlate the disturbance terms
for ICPT and SLOPE, it says that the solution is inadmissible because
the matrix between the disturbance terms is not positive definite.
When I remove the correlation between the disturbance terms for ICPT
and SLOPE, it says the solution is inadmissible because the variance
(or disturbance term) for SLOPE is negative. This has happened for
more than one of the variables I am attempting to model.
What am I doing wrong?
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Tue Oct 07, 2008 3:58 am
|
|