Hi All,
Am getting this error when I run the attached model for CFA in Amos. Will appreciate if you can help me in resolving this error.
Deepali Bhardwaj, your model has to many parameters and AMOS can not find a solution. It is if you ask AMOS to give a value for X in the formula X+Y = 2. This has many solutions. If you fix Y to 1, X is 1 as well and you have a solution. Your model is very complex and as far as I can see, you only fix one parameter to 1 and the rest are free. You might want to fix to 1 one path from all latent variables to the observed variables for all latent variables. Fixing only one path, might AMOS tell you to fix another parameters. It will continue to do so until AMOS can find a solution. A simple method is the counting rule (Kaplan, 2000, 21) if s = total number of variables. The number of non-redundant elements in the population variance/correlation matrix is 1/2 (s(s+1). If t is the number of parameters, t must be
Hi Deepali,
beyond Peter's elaborate explanation which I 100% concur with, I could imagine that especially latent variable #8 (you have strange routine to label latents :) could make problems. As this latent has only one indicator it could be necessary to fix not only the loading but also the error variance.
Beyond that, I agree with Peter: Reducing the number of items not only reduces trouble but also forces you to select those indicators that really and as closely as possible represent your imagined latent.
This paper should be of some value:
Hayduk, L. A., & Littvay, L. (2012). Should researchers use single indicators, best indicators, or multiple indicators in structural equation models?. BMC medical research methodology, 12(1), 159.
Best,
Holger
Hi Deepali
There are some excellent answers already, but I'll add my tuppence worth.
In my experience most identification problems arise through trying to set the metric of the latent variables. There are two ways to do this. First, the reference indicator method involves fixing a 1 in the first factor loading for each latent variable. This means that the variance of the latent variable is a free parameter that will be estimated. When a latent variable is endogenous this is the way that you have to proceed. Second, you can estimate all the factor loadings but fix the variance of the latent variable, usually at 1 (this helps with interpretation). So there are 2 choices, and one method, not both, must be used for identification problems (the issue of identification has been beautifully explained by Peter).
A latent variable with one indicator is a special case. You can fix the loading at 1 and error variance at 0, this makes the indicator have a reliability of 1, and this is usually an unrealistic assumptions. Alternatively you can fix a 1 in the factor loading and a value for the error variance that implies a particular reliability. This is good webpage to read up on this http://reifman-sem.blogspot.com/2012/04/we-recently-discussed-how-to-handle.html
Mark
Thank you all for your inputs. Figured that there was only one item attached to factor 6. Have resolved it an now the model is working.
Hi Deepali, you need to put a "1" on one of the arrows connecting your latent variables to their indicators. You have done that with latent variable 6, but not with the others (see the one above the first arrow on the oval marked as "6"). You may what to do my new course on CFA: https://www.udemy.com/course/confirmatory-factor-analysis-with-amos/?couponCode=9FE45F1E896A4522869C