|
|
 |
| Seemlingly inconsistent CFA fit stati... |
 
|
| Message/Author |
|
|
| Anh Hua posted on Wednesday, August 07, 2019 - 11:40 am
|
|
|
Hi all,
I ran a CFA model with 8 factors and 42 items on a large sample size of approximately 6000. Because it's a likert scale measure, I chose WSLMV as the estimator for my CFA.
What's interesting about the results is that while RMSEA (.05) and SRMR (.05) were within the acceptable range of <.08, the chi square of the model was significant (p <.001) and CFI and TLI were around .70, which is too low for acceptable model fit.
Can someone kindly explain why this is the case (i.e, why CFI and TLI were not in the acceptable range but RMSEA and SRMR were)? My understanding is that, because p value of chi square was <.001, that's evidence for model misspecification, but what does the acceptable RMSEA and SRMR values and the low CFI/TLI values tell me about my model fit?
If I had to guess as to why the CFI/TLI were low, it's probably because I have way too many factors and way too many items. Also, since the baseline model assumes that the covariation of the 42 items is 0, the fact that I have low CFI/TLI suggests that the covariation of the items in my scale are not highly correlated enough? Does this reasoning seem correct?
I would really appreciate any feedback on how to make sense of my results and on my thinking about the CFI/TLI. Thank you for your time!
Anh |
|
|
|
|
| Good question for SEMNET. Indices often disagree. But with CFA for that large of a model, there seems to be many occasions for cross-lagged loadings - try an EFA to see. |
|
|
| Anh Hua posted on Friday, August 09, 2019 - 1:32 pm
|
|
|
Thank you, Dr. Muthen! This sounds like a great suggestion. I will go back to conducting an EFA to see if there're any cross loadings and I suspect that there will be!
Sincerely,
Anh |
|
|
| Back to top |
|
|
