Greetings, I would be really grateful for a bit of advice. I want to demonstrate discriminate and convergent validity of the latent variables in my model. As I have a predetermined idea about which items should load onto which factors I understand that confirmatory factor analysis (CFA) is the most appropriate way of testing construct validity (I am using AMOS). Once I have tested and modified the model to obtain a suitable fit, I intend to use the handy Stats tool on James Gaskin's statwiki page to obtain the AVE scores etc
http://statwiki.kolobkreations.com/index.php?title=Main_Page
My main question is: in order to test construct validity, do I need to add covariances between the exogenous variables? I am not interested in the extent to which the 3 exogenous variables are related except in relation to the issue of discriminant validity.
many thanks for any help you can offer. Nick
Thanks. It just that the excel tool developed by James Gaskin I am using to help calculate the validity scores doesn’t seem to work without the correlations output from AMOS which I can only get by covarying the IVs
Sorry for my previous answer but as those variables are exogenous then it is indeed necessary to introduce covariances. It is the same as in multiple regression. Have a look at
For testing the CFA you need to covary all latent variables ( affactivecom, trust, techqual and loyalty ) then check the correlation box ( ✔ ) at analysis properties window on AMOS to get results of correlations. after running your model, you can use outputs of AMOS on Gaskin's Excel.
Best ,
Alireza
Hello, again! I think that confirmatory factor analysis should be used prior to the specification of the structural model in order to test for the validity of the measurement model. Thus, I believe that the variables ( affectiveCom, Trust, Techqual), if they belong to the same instrument, should be tested together prior to the path analysis (I.e. the regression of the three variables to loyalty).
The discriminant validity is based on the correlation of latent variables (it is recommended that they be less than .70) and the factor loads of the exogenous variables in relation to each dependent variable (greater than .50).
Thanks all. Most helpful. Please excuse my ignorance - I am away from my institution and trying to learn all this on my own.
Ioannis Katsantonis are you suggesting to first run the CFA with just the independent variables in the model? ( as in the attached picture added to this post). At this stage I am more interested in testing the validity of the constructs. However if I exclude the dependent variable (loyalty) how then can I demonstrate the validity of this construct?
Nick W Williams, you can run (a) either separate CFAs for each construct and if they indicate adequate model-data fit then construct validity is established OR (b) you could run a pooled CFA with all the constructs connected with covariances, either approach is acceptable.
My previous suggestion was to conduct CFA with the exogenous only IF they originate from a single measure/scale etc.
What you present, though, in the first path diagram of the post is a structural model which presents the impact of the IVs on the DV. Construct validity is established when testing the measurement model.
Among the exogenous variables a covariance is needed to test for multicolinearity and discriminant validity (I.e. the extent of the distinctness of the exogenous is sufficient to affect the endogenous).
With regards,
John
thanks for your help to date Ioannis Katsantonis Alireza shabani shojaei Cristian_Ramos-Vera. Much appreciated.
Could you kindly advise on an important question I am grappling with. Would you advice co-varying the dependent variables in with the dependent variables? (I have attached picture to demonstrate what I mean). Otherwise how does one determine construct validity of the dependent variables? Or is this less important than determining validity of the independent variables. Again sorry for any ignorance demonstrated.
I’m afraid I don’t follow you. However for testing the measurement model ( CFA ) we should to co-vary between ALL variables ( depended and independent latent variables ), which already you have done the same at your model.
Hi Nick. It seems that you have already some good answers here: the main point is that you should do first a CFA, before replacing the covariances between some of your latent variables with regressions. I have a new course on Udemy that does these procedures in several exercises. https://www.udemy.com/course/confirmatory-factor-analysis-with-amos/?couponCode=9FE45F1E896A4522869C
- Badges
- Science topic
- Similar topics
- Education