We received the following error. To rectify this we removed the error terms which showed negative variance i.e (e16, e20, e24). Can someone please tell me a way to fix it?
@Sanyam Ahuja Your model is underspecified as you have only two observed items under all your latent variables. Although, you can make it run by putting regression weight on both paths from each latent variable to their respective observed items.
For this and your related query (about the negative variance) what you might consider doing is abandoning all of the latent variables, summing the two related indicators together (e.g., AT1, AT2), and reformulating the model as an ordinary path analysis.
In general, it's recommended to have at least three indicators (manifest variables) per latent variable. So, for future exploration of this type of model, perhaps you could look for or create additional measures for each intended latent variable.
Thank you David Morse id and Imran Anwar for your responses. Imran's solution was a quick fix but we can't use that for our research paper. David we considered your solution and considering that we don't have many indicators for all our latent variables we are left with this option I feel. But we wanted to still apprise you about the 2 approaches we were thinking about :-
1) we remove the 2 latent variables and the model was fine with zero negative variance. The chi square value of the model appeared to be 497
or
2) we combine the indicator variables for the respective latent variable and then follow this video https://youtu.be/xAVHnSMxW0c (if possible can you please check this video out and tell us whether we are in the right track).
and is it possible to apply ordinary path analysis for technology acceptance models because during our literature survey all we came across was SEM and CFA.
@Abhinit Mahajan Indeed, what I suggested was a quick fix but not an appropriate way for covariance based SEM and path analysis is a way better choice when using AMOS to test your model ( as suggested by @David Morse). However, you could also use Smart-PLS as it works perfectly fine with two observed items under a latent variable.