great explanation i have been searching for a simple videos like this that explain the concepts behind SEM method and eventually i found this ! so thank you very much sir!
I don't think there is a "minimum" correlation necessary. Theory will determine your model more that retrospective correlations....Saying that, if you have really low correlations you are most likely not going to find significance in your model.
Hello, Mr. Collier. May I ask if you have encountered a correlation between 2 constructs that is greater than 1 in SEM? I did. How do I explain this? Is it ok to consult?
So you have what is called a "Heywood" case. This is an error in the analysis. You have multicollinearity between your two constructs. In essence, the constructs will not distinguish between one another because they are so similar. Your best bet is to drop one from the analysis and move forward.
@@joelcollier9387 Thank you Sir for replying. May I ask one more question? I also got an R^2 that is greater than 1. Path coefficients can be greater than 1, right? But can R^2 be greater than 1?
@@rubymoran9359 Standardized path estimates can not be greater than 1. Unstandardized path estimates can be greater than 1. With a R2 greater than 1, that says you are explaining more than 100% of the variance. Subsequently, This is an invalid result. My best guess is you have multicollinearity between constructs that is making your analysis go haywire.
great explanation i have been searching for a simple videos like this that explain the concepts behind SEM method and eventually i found this ! so thank you very much sir!
Clear and simple explanation.
Hi Joel! This is a great explanation! Really, really helpful to understand the underlying mechanisms of SEM!
Thank you Professor!
Thank you for the video!
Really insightful
Thats the explanation of observed cov matrix then how the explanation of estimated covarusnce matrix?
Hi Joel, I was wondering if you have any resources regarding the minimum correlation strength required between variables to run an SEM?
I don't think there is a "minimum" correlation necessary. Theory will determine your model more that retrospective correlations....Saying that, if you have really low correlations you are most likely not going to find significance in your model.
Hello, Mr. Collier. May I ask if you have encountered a correlation between 2 constructs that is greater than 1 in SEM? I did. How do I explain this? Is it ok to consult?
So you have what is called a "Heywood" case. This is an error in the analysis. You have multicollinearity between your two constructs. In essence, the constructs will not distinguish between one another because they are so similar. Your best bet is to drop one from the analysis and move forward.
@@joelcollier9387 Thank you Sir for replying. May I ask one more question? I also got an R^2 that is greater than 1. Path coefficients can be greater than 1, right? But can R^2 be greater than 1?
@@rubymoran9359 Standardized path estimates can not be greater than 1. Unstandardized path estimates can be greater than 1. With a R2 greater than 1, that says you are explaining more than 100% of the variance. Subsequently, This is an invalid result. My best guess is you have multicollinearity between constructs that is making your analysis go haywire.
@@joelcollier9387 Thank you for your reply, Sir. I have two groups but this happens to only 1 group. 😔
Brilhant