|
|
 |
| How many factors to choose in EFA |
 
|
| Message/Author |
|
|
| Yue Ma posted on Monday, November 20, 2017 - 4:18 pm
|
|
|
Hello,
I am using EFA on 7 indicators, here is what I got:
SUMMARY OF MODEL FIT INFORMATION
Model / # of Parameters / Chi-Square / DF / P-Value
1-factor: 21, 255.255, 14, 0.0000
2-factor: 27, 12.945, 8, 0.1138
3-factor: 32, 1.338, 3, 0.7202
Models Compared / Chi-Square / DF / P-Value
1-factor against 2-factor: 233.179, 6, 0.0000
2-factor against 3-factor: 15.873, 5, 0.0072
Should I choose 2-factor model given that the p-value = 01138 is not significant for the 2-factor model, or should I choose 3-factor model given that 2-factor against 3-factor p=0.0072 is significant?
I also wonder how to count the number of parameters for EFA
Thank you! |
|
|
|
|
The 2-factor model fits so no need to use more factors.
EFA parameter counting is shown in our video and handout for Short course Topic 1 on our web site. See the Short Course video web page:
http://www.statmodel.com/course_materials.shtml |
|
|
| Yue Ma posted on Monday, November 20, 2017 - 8:40 pm
|
|
|
| Thank you, Dr. Muthen! |
|
|
| Kathy Xiao posted on Tuesday, April 10, 2018 - 5:22 am
|
|
|
Hi Dr. Muthen,
I ran a EFA from 1->5 factors. The results shows from 1 to 4 models, the p values are all significant. For 5-factor model , it shows “NA”. And the “n-1 factor against n factor” comparison are all significant. Inspection of eigenvalues shows only the 1st factor has >1 eigenvlaue.
In this case, shall I choose the 1-factor model? Can I tell if alternative models (2,3,4 factor mods) are less fit than 1 factor model?
Thanks very much !! |
|
|
|
|
| These general analysis strategy questions are better suited for SEMNET. |
|
|
| Back to top |
|
|
