Dear all, I would like to know your opinion on the following issue I am struggling with.
I am estimating a multigroup model with several factors all correlated among them (they represent different dimensions of the same concept). I have a good fit which points to metric invariance, and I would like to comment to what extent the covariances among the factors differ in the two groups.
As far as I know, to do that I should ask for the confidence intervals of these covariances, and compare them in the 2 groups (an overlap should suggest that the difference in that covariance in teh 2 groups is not statistically signifcant). The issue here is that when I ask for the bootstrap (that I think it is the way to obtain confidence intervals) Amos gives me an error message saying "an attempt was made to compute the correlation between two variables, one of whose estimated variances failed to positive. This attempts was made because 'Standardized estimates' in the 'Analysis properties' window was checked, or because the Standardized method was used".
Do you have any clue on how to fix this problem?
Thanks a lot in advance, best, Giorgio
Hello Giorgio,
You could treat the question as a nested models problem: The unrestricted model allows all covariances to be independently estimated, whereas in the restricted model, you fix each covariance (between a given pair of variables/factors) to be equal for all groups. The change in model-data fit indices (starting with the chi-square) represents a test of the hypothesis that groups have comparable covariances.
Good luck with your work.
In order to check for significance of invariance across paths/covariance while doing Multi-Group SEM, you need to add group differences plugin to AMOS.
Dear David Morse and Imran Anwar , thanks for your feedbacks.
David Morse: I see your point, but it is not exactly what I have in mind. Indeed, I already did look at the fit indices for the models with increasing invariance, and the chi square test suggests that the 'accepted' one is the metric invariance. From this, I would like to run the multigroup with metric invariance and report the values for each covariance, to evaluate for each of them the significance of their difference in the two groups. In order to do that, I think I need the confidence intervals, but as I said when trying to estimate them with bootstrap Amos does not estimate due to negative variance... do you have tip on this?
Imran Anwar I did not know this plugin, where can I find it and how does it work?
Thanks a lot again.
https://drive.google.com/file/d/0B3T1TGdHG9aENVBmSDZzSUNSR1k/view?usp=drivesdk
Above is the link for estimand to compare paths using chi-square statistics in AMOS. Moreover, there are many more plugins and estimands for AMOS.
To know how to analyse in AMOS, refer to the video 👇
https://youtu.be/Pakg3PlppuY
- Badges
- Science topic