There are a number of fitness indices out there and AMOS supplies a number of them. Like other commentators the Chi-sqaured should always be reported. The RMSEA is usually reported and depending on your field of research should usually be below 0.05 but some journals will permit 0.08 depending on the field. SRMR is also required since it is a different type of fit statistic and values again below 0.05 are very good but again 0.08 is also permissible. After that you have indices such as CFI, GFI, AGFI, TLI etc ... generally a score of 0.95 is very good and 0.90 acceptable. However not all of the scores can be on the low end of the scale.
Before you even get to fitness indices you need to ensure that each item loads well onto its hypothesized factor in the CFA part of the model and these CFAs should be done first before doing the model itself. Poor CFAs will lead to poor model fit ... Each CFA should ideally be specified individually to check for measurement. This is the now pretty standard two step approach advocated by Anderson and Gerbing (1988).
The important values in CFA are factor loading for every items, the R-Square for every item, and also the correlation between latent constructs in the model.
More importantly, you need to assess for the Fitness Indexes which reflect how fit is your measurement model to the data from the field.
There are a number of fitness indices out there and AMOS supplies a number of them. Like other commentators the Chi-sqaured should always be reported. The RMSEA is usually reported and depending on your field of research should usually be below 0.05 but some journals will permit 0.08 depending on the field. SRMR is also required since it is a different type of fit statistic and values again below 0.05 are very good but again 0.08 is also permissible. After that you have indices such as CFI, GFI, AGFI, TLI etc ... generally a score of 0.95 is very good and 0.90 acceptable. However not all of the scores can be on the low end of the scale.
Before you even get to fitness indices you need to ensure that each item loads well onto its hypothesized factor in the CFA part of the model and these CFAs should be done first before doing the model itself. Poor CFAs will lead to poor model fit ... Each CFA should ideally be specified individually to check for measurement. This is the now pretty standard two step approach advocated by Anderson and Gerbing (1988).
First of all, you should access convergent validity and discriminant validity of your constructs. items of a construct is said to be convergent when its C.R is greater than 0.70 and AVE value is larger than 0.5. For discriminant validity, inter construct correlation must be less than square root of AVE. Regarding recommended values to qualify your model as acceptable, CFI and TLI should be atleast 0.90, while SRMR should be less than 0.05 and RMSEA should be smaller than 0.08.
The important values in CFA are factor loading for every items more than 0.6 if the questionnaire adopted from literature. and more than 0.5 if you develop the questionnaire. also the correlation between latent constructs in the model must to be less than 0.9.
for convergent validity and reliability C.R is greater than 0.70 and AVE value is larger than 0.5
after that you need to assess for the Fitness Indexes
CFI, GFI, AGFI, TLI, NFI >90. RMSEA should be smaller than 0.08. . Chisq/df