Growth model with no intercept
Message/Author
 giovanna caprara posted on Wednesday, October 31, 2007 - 2:50 am
I have a variable measured four times
age 14, 16, 18 and 20
I have done a series of GCM to find the best one. in particular the first model tested was a no-growth model (only intercept) The second was a linear model. In this model we fixed the factor loadings on the slope at 0, 1, 2, 3. The third model examined a non-linear growth.
The chi square difference test revealed that the best model is the non linear one.
yt = n0 + n1 xt + et ; with x3 and x4 as parameters to be estimated;

Now a referee told me that it is crucial that the authors evaluate the following submodel:
yt = n1 xt + et, with x2, x3, and x4 as parameters to be estimated and x1 =1.

this is the input i have wrote:
s by MD14@1 MD16*1 MD18*2 MD20*3;
md16 with md18 (4);
md14 with md16 (2);
md18 with md20 (2);
MD14 (8);
MD18 (10);
model female:

md16 with md18 (5);
md14 with md16 (1);
md18 with md20 (1);
MD14 (8);
MD18 (10);

Is this model nested with the linear one?
and is the requested model make sense?
Thank you
 Linda K. Muthen posted on Wednesday, October 31, 2007 - 8:18 am
This looks like a growth model with no intercept growth factor. I am not sure if it is identified and it does not make sense to me. I would ask the referee to give more information.
 giovanna caprara posted on Wednesday, October 31, 2007 - 8:48 am
thank you ,
this model is identified with a good fit.
The reviewer cited the work of Meredith and Tisak (1990). My problem is that I don't think it is nested with the other models i tested. Is it right?
 Linda K. Muthen posted on Wednesday, October 31, 2007 - 9:17 am
Even if the model is nested, with an implied zero variance of the intercept growth factor, difference testing cannot be used because a variance of zero in on the border of the admissible parameter space.