Hi there
I have a model with 5 latent variables and the related indicators. Now I want to constrain certain parameters i.e., factor loadings and error variances to be positive and factor correlations to values between 0 and 1. I specified parameter names (e.g., f1 for the factor loading of indicator 1) and then put something like "f1>0" in the parameter constraints box. However, this doesn't work. It appears that in Amos one can constrain parameters to specific values (e.g., f1=1) but not to a specific range.
Am I missing something here? How can I tell AMOS to constrain parameters to a range of values instead of a single value? Any help or advice is greatly appreciated.
Why you want to constraint in the first place?
Constraint will not solve anything.
As far as I know, you can only set them to a specific values or equal to each other.
If your model is reflective in that case you have to constrain one variable, else wise AMOS will not run. For this go to object properties and click parameter and put 1 in the box. this process will constrain the specified relation.
Regards
Thank you very much for your responses. Meanwhile I figured out that it is, in fact, not possible to impose such constraints in AMOS. However, with MPLUS and R (lavaan) it works. BTW: simultaneously constraining certain parameters is a reasonable strategy in certain applications (e.g., http://proceedings.aom.org/content/2015/1/13078.short?related-urls=yes&legid=amproc;2015/1/13078)
@Armin, It is very much possibly to apply constraint in AMOS. Read Aurbuckle and Wothke (2004) guide to use AMOS
@Jeevan: of course it is possible to constrain parameters to specific values. The question here was whether they can be contrained to a range of values (e.g., >.00 and
Armin,
To best of my knowledge , we can use only discrete starting values as constraints to run the model. The basic purpose of fixing the constraint is to estimate the free parameters and to find out the difference between the sample covariance matrix and implied covariance matrix. Numbers of implied matrices are generated in different iterations and this process continues till it is not possible to estimate the parameters and produce an implied covariance matrix. The convergence between the sample covariance matrix and implied covariance matrix results in the model fitness. Robust fitness results when the difference in the two matrices is as small as possible. The range of values used as a constraint results in almost same level of model fitness, hence there is no need to use range of values. Please refer a book on Introducing LISREL by Diamantopoulous and Siguaw.
Armin, Why do you want to use range of values as constraints??
Hope this is useful..
Hardip
you can also do it in EQS with inequalities
/ine
(V1,F3)>0.07;
(V2,F4)>0.182;
(V3,F5)>0.122;
(V4,F6)>0.182;
(V5,F7)>0.135
Armin Pircher Verdorfer the link you posted appears to be broken. Would you be able to link the source for your comment again? Thank you very much
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling