I am translating an English questionnaire. Using AMOS, the CFA of the models that I analysed shows a fairly good construct validity (fitness level of GFI >0.9 and RMSEA
this is a tough one. Under ideal conditions, you would compare measurement invariance between similar groups, one using the English, one the other version of the questionnaire.
First of all, most reseachers I know will probably tell you not to use AMOS, but I cannot comment on that because I am not familiar with the programme.
Secondly, your fit indices are not that great to begin with, e. g. your CFI. Does AMOS provide you with more, e. g. TLI, SRMR, Chi-square?
Thirdly, I do not understand how you measure convergent and discriminant validity. As far as I know, your need an additional scale that is similar (convergent) and one that is dissimilar (discriminant) to your own.
AMOS does provide the other fitness index. AGFI reach the level of > 0.9.
Based on a book I read on SEM, the chapter of CFA gave the calculation of
Convergent validity = AVE is more than 0.50. Due to the low factor loadings, I can't achieved it.
How low it is? most of the factor loadings are around 0.3 to 0.5.
I just realized I had not calculated the discriminant validity yet. So that will be done later. Wondering what else can be done to improves the scores.
@Hazwani In your model, loadings of observed items with their latent variables are too low to establish the convergent validity and would create problems in divergent validity as well, because of low convergence, when comparing squared root of AVE a latent variable to its correlation with other latent variables.