2
$\begingroup$

I was wondering if you can help me. I want to run a pathway model using composite variables. Each composite variable is made up of 3 or 4 observed items. To create the composite variable to draw my model in AMOS do I simply compute the variable in SPSS by adding up the observed items relevant to each composite variable. Or do I use the data imputation option in AMOS to create my composite variable? Is there a preferred method

$\endgroup$
  • 2
    $\begingroup$ If you are asking how to use AMOS, that would be off-topic here. You don't need a composite variable, however. Add a latent variable that all the observed items measure. $\endgroup$ – gung - Reinstate Monica Dec 15 '14 at 14:07
  • $\begingroup$ @gung, consider converting to an answer. $\endgroup$ – StasK Dec 15 '14 at 15:01
1
$\begingroup$

Don't use a composite variable in a structural equations model. Instead, add a latent variable that the manifest variables (i.e., the items) all measure. The composite variable isn't what the items measure, it is just (presumably) a better measure of the latent variable in question. If you were to add a composite variable, it would decrease the ability of your model to extract the information in your data, as it forces the loadings to be 1, whereas the latent variable will allow the loadings to be estimated from the data.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ it would almost certainly degrade your model's fit -- yes and no; you will have fewer moments to fit, and the model structure will be different. Consider an SEM with two latent variables, regressed on one another, and with a sufficient number of indicators. The full model may not fit well because of cross-loadings or measurement error correlations. On the other hand, if you make composite variables of all of the indicators of each latent variable, then you will end up with a regression model in two variables that has zero SEM degrees of freedom and hence a perfect fit ... $\endgroup$ – StasK Dec 15 '14 at 15:38
  • $\begingroup$ ... so the composite variables is a way to sweep lack of fit under the carpet. ('twas too long for one comment) $\endgroup$ – StasK Dec 15 '14 at 15:38
  • $\begingroup$ @StasK, it seems like that would improve the fit by biasing the model. I have changed the answer; see if you think it's better now. $\endgroup$ – gung - Reinstate Monica Dec 15 '14 at 15:40
  • 1
    $\begingroup$ Thank you -- it is more specific now as to what the mechanics of the composite variables are. $\endgroup$ – StasK Dec 15 '14 at 15:42
  • $\begingroup$ @StasK Thank you so much for your helpful insight. You are right a pathway model would be an almost perfect fit of a structural model that uses latent variables. I am writing my PhD and I need to explore, discuss and explain that I have used both pathway models and latent structural models so that I can demonstrate that I know the pitfalls for both and to show the examiners I understand how to interpret the models. $\endgroup$ – user63869 Dec 16 '14 at 9:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy