why is -2LL value of nested model smaller than that of full model, and how to find the best start value

1 reply [Last post]
mailyn's picture
Joined: 05/19/2010

when I simulated the model, I found the -2LL value of some nested models (which had less estimate parameters than full model) were smaller than that of full model, without any "warning" in script. I think it is abnormal, and I have no idea why it happened, and could you give me some information or suggestion to solve this problem.
Thank you!

question 2:
In Mx, there are some order to help me to find the best start value, like "option THard=-n" or "option jiggle", so, is there any order in OpenMx with the same function. Could you give any clue?
Thank you again!

Steve's picture
Joined: 07/30/2009
I suspect that these may

I suspect that these may actually be one question. Unexpected -2LL may occur if the optimizer is stuck in a local minimum. Trying again with several starting value is one way to see if this is the problem.

One way to use R to generate your starting values might look like this:

rstart <- rnorm(1, mean=0, var=.5)
somePath <- mxPath(from=x, to=y, values=rstart, free=TRUE)

You can use R variables in specifying values= arguments in mxMatrix and mxPath.