I am using AMOS to test a model.The CFA model fit is ok, but I cannot get a satisfying path model fit. Because I have missing data, I cannot use the Modification indices function to see what could be the possible ways to improve model fit. Has anyone got any idea about how to improve the model fit in AMOS? At the moment, I am trying to draw co-variances between the possible items, but the fit still does not meet the standard. Are there any other ways to improve model fit? Many thanks in advance.
First, it is important to know how you treated the missing values (there are different treatments of missing value).
Secondly, it's not clear how you can draw co-variances between items without looking at Modification Indices.
As I know, you first need to fix the missing value using any of different alternatives available (you may get some idea here: http://handbook.cochrane.org/chapter_16/16_1_2_general_principles_for_dealing_with_missing_data.htm)
You should draw co-variances by looking at the modification indices in order to improve model fit. You should also look at loading of the individual item. Dropping items with poor loading (also considering the theoretical importance of the item) always helps improve the model fit.
Hi Khan, thank you very much. I already dropped the items with poor loading. I guess I need to impute the missing value. I may use EM or MI to fix it, but one of the important variable has 20% missing. A big headache for me. That is why I thought just leave it.
Do you think add more control variables or change the control variables will help improve the model fit?
Dear Lili:
Perhaps I understand your problem. But its difficult to comment on your control variables without looking at the model.
At the same time, 20% missing value is significant. To me, the best way to improve the model is to fix the missing value. Not sure, if you did your survey online, is it possible to get the responses again by resending the survey to those respondents? As I mentioned before, its also not justified to draw co-variances without looking at modification indices. Apparently, fixing missing values is required.
Hope this clarifies.
Dear Lili,
Good day,
I think you have to ignoring the missing data and just analyzing the available data,
Regards,
Dr. Alhaddad
Dear Lili,
Try to ignore the missing data, and tell me if it's works,
Regards,
Dr. Alhaddad
You can remove the entries or individuals with missing data. However, this can reduce your sample size. You may need to test for the adequacy of the sample.
Regards
If you remove the individuals with missing values from your sample, please test those two groups (complete / missing) for significant differences. Otherwise, you might miss an important facet of your sample (e.g. individuals with a specific set of characteristics drop out early).
http://orm.sagepub.com/content/17/4/372.short
Check the CFA first for factor loading of each latent variable.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS