In the SEM model I used the total score of my dependent variable (As an observed V.), however I run CFA for this variable using its items and the result of model fit was good. Would you please let me know whether it is acceptable or not?
I worry if the items are too many for CFA.
Thank you,
I think the key question is whether your 18 items do indeed form a single scale, as opposed to something more complex. Since you have an acceptable level of fit, that essentially says that your single scale model is good enough.
I'm not sure why you are thinking in terms of discriminant validity, which would mean that a construct was unrelated to some other concept, as predicted beforehand.
It is ok to run a CFA on 18 items, but it might be more convincing to run a CFA with more than one scale so you can show that the fit is better with items loading on their own scales than with all items loading on a single factor.
Thank you dear Prof. Spector..
Would you please explain? You mean "the measurement model" with all scales in my research?
Regards,
Farima,
did I understand you correctly that you wish to use the total score of the variable and that you seek for confirmation that the 18 items are uni-dimensional, which means that they are all measuring the same construct? It is of course a good idea to investigate convergent validity. Paul suggests to investigate discriminant validity, too, which would also be fine.
As Paul already pointed out, it is o.k. to run a CFA on 18 Items, but you would need a sample size appropriate for your estimation method. If you are using maximum likelihood estimation then you would need at least N = 100, better N = 200.
HTH
Karin
Dear Prof. Schermelleh-Engel
Thank you for your reply.
Yes, it is exactly what I wish. My sample size is 305 after deletion of outliers.
So, in order to investigate discriminant validity, more than one scale for each variable is needed. Am I right?
Regards,
Farimah,
yes, you would need to include more items, your 18 items and additionally other items measuring a different construct. In this case you would analyze a model with two correlated latent variables.
It all depends on what your aims are in your research.
Kind regards,
Karin
Dear Farimah,
In CFA, you should not take the composite score of your items. Do CFA with all your 18 items and assess the Factor Loading, Fitness Indexes, and proceed into computing Average Variance Extracted (AVE), Composite Reliability (CR) etc. The whole process is called Validating the measurement model of a latent construct.
I think the key question is whether your 18 items do indeed form a single scale, as opposed to something more complex. Since you have an acceptable level of fit, that essentially says that your single scale model is good enough.
I'm not sure why you are thinking in terms of discriminant validity, which would mean that a construct was unrelated to some other concept, as predicted beforehand.
Thank you dear Prof. Morgan.
Yes, these 18 items form a single scale to measure marital satisfaction. The original scale had 15 items and I did some cultural changes base on literature and increased to 18.
Indeed, I run CFA before putting the variable in the mediating model to determine the factor structure of my dataset and make sure that the added 3 items did not put me in trouble in the final modelings.
Thank you again,
Regards,
You can't test the mean score of the 18 items as the dependent variable and the 18 items as the dependent variables. It is like calculating Body mass Index (BMI) from weight and height and testing whether BMI depends on weight and height. Of course it will
Dear Dr. Karuthan
Thank you for your reply.
I am a little confused. Would you please explain more what you mean?
Regards
- Badges
- Science topic
- Similar topics
- Philosophy Of Science
- Models