Hello,
I am testing a model using three different questionnaires, two of second-order and largely used on the literature and one already validated by me. I have about 1700 participants.
I'm running the model on AMOS and getting confusing results from the model fit indices: while RMSEA is 0,043(0,044 - 0,045, pclose=1,000) therefore acceptable, the values of CFI=0,880 and TLI=0,875 are lower than expected. What does it tell me about the model? How could I improve it?
Also, does anybody have a reference about the number of modification indexes I could use?
Thank you in advance.
Mayara Matos, it is not easy to give definite answers/advices without knowing more about your aim as well as your model. CFI and TLI will be influenced by the average level of correlation (ALC) in your data. The higher ALC, the closer CFI/TLI gets to 1. You can examine the factor loadings as well as the average variance extracted (AVE) in an attempt to technically increase CFI/TLI without having to use modification indices. Having said that, I personally do not consider CFI=0,880 and TLI=0,875 as poor values, particularly if you are estimating a complex model including second-order factors etc. If all the estimates are sensible and come out as expected, I do not favour modifying a model just for the sake of slightly improving CFI/TLI.
You can check the error covariance indexes, check the factorial loads.
Mayara Matos, it is not easy to give definite answers/advices without knowing more about your aim as well as your model. CFI and TLI will be influenced by the average level of correlation (ALC) in your data. The higher ALC, the closer CFI/TLI gets to 1. You can examine the factor loadings as well as the average variance extracted (AVE) in an attempt to technically increase CFI/TLI without having to use modification indices. Having said that, I personally do not consider CFI=0,880 and TLI=0,875 as poor values, particularly if you are estimating a complex model including second-order factors etc. If all the estimates are sensible and come out as expected, I do not favour modifying a model just for the sake of slightly improving CFI/TLI.
I totally agree with Mehmet Mehmetoglu and would have the same approach.
- Badges
- Science topic
- Similar topics
- Biological Science
- Histology
- Fixatives