I am conducting a CFA through AMOS, my scale was based on 67 items all items chosen from literature or Alpha value is also good. on the basis of CFA results, i have removed 16 items on criteria of .50 below factor loading through that i have to delete two full constructs from my model.
Now all factor loading are above .50 of 51 items in my scale. My sample size was 405.
After deletion of poor factor loading my model fit values are as;
Constructs 51
Chi-Square (CMIN) 2410.469
CMIN/DF 2.004
CFI .847
IFI .805
RMSEA .050
Before deletion values
Constructs 67
Chi-Square (CMIN) 5299.169
CMIN/DF 2.514
CFI .725
IFI .728
RMSEA .061
Please suggest any suggestion how to improve these values or any other way to tackle it.
You can check with the results of Modification Indices and you have to find the highest Modification Indices values of your items and connect them by covariance. Then automatically your model fit will be improved.
All the best
Amna Farzand Ali though modification indices can help in improving model fit in CFA, it is important to note that "cross border" error term correlations are not allowed. From cross border mean do not follow the software if it suggests to connect item errors from Variable A to Variable B and so on.
Your loadings would be acceptable only with reasonable AVE (0.50 and above) and CR (0.70 and above) values.
Most importantly, I suggest you to clean your data by checking and deleting for OUTLIERS (in your case item-wise). Extreme outliers not only reduce the model fit but also badly impact the relationships in the model. Best of luck.
Dear Amna Farzand Ali
I think this link will be helpful to some extent
https://www.youtube.com/watch?v=JkZGWUUjdLg
Hi Amna, if your aim is to get the fit statistics to reach some pre-determined level then just keep adding correlated errors until you achieve 'acceptable' model fit. That's the easy bit. The difficult part will then be trying to convince people that the analysis you did was 'confirmatory' in nature - this will be difficult because what you will have done is to move into a 'model generation' strategy. You will not have confirmed the structure of the items.
Have you tried to understand why the proposed model does not fit? it may be helpful to look at item level descriptive statistics - is there evidence of floor/ceiling effects or missing data on particular items. This information can be used for an initial cull of items. If the scale is multi-dimensional you may want to start by testing each dimension separately. You may also want to consider if your method of estimation is appropriate - if your response scales have a few categories then ML (which I assume you are using) may not be appropriate.
I hope this helps
Mark
Thanks for the valuable suggestions its working.
Can someone guide me if a statement is measuring a heart of construct but having low factor loading. In this case what to do?
A similar question was asked on RG, which I think will be helpful
https://www.researchgate.net/post/What_are_the_tactics_to_improve_model_fit_indices_in_CFA_in_R_Lavaan
There are two methods to it. Either drop an item below the loadings of .50 or covary the redundant items based on model fit indices. Best method is to covary the redundant items as it is better to retain the item. I think to solve validity issues is difficult (next step) than model fit of measurement model. All the best
your results are quite good just try to covary the highest modification indices.
Hi Amna,
I am also conducting CFA not in Amos but in Mplus. I want to ask a question. You wrote that you deleted items below .50 because of the low factor loadings. Did you delete items or factor loadings below .50 together or one by one?
Best,
- Badges
- Science topic
- Similar topics
- Psychology