Post Your Answer
4 years ago in Likert Scale , SEM Analysis , Structural Equation Modelling By Shabana
How can the error correlations be justified during SEM?
I want to know what kind of error justifications (mostly theoretical) are considered to be a good example while doing the structural equation modelling. As per my study, this is not rare and happens pretty often when researchers face trouble/issue while correlating the errors so that the quality and fairness of their model can be improved.
In a few studies, I came across that mostly the researchers end up using measurement errors which are not exactly the same as what is required but similar to the areas they assess. In many cases, researchers have also failed to properly connect the errors with the proper justifications. The in-depth analysis or not, sometimes researchers fail to write down the desired justification.
I heard that having longitudinal studies or priori justifications are good for providing justifications in such cases, does it really work?
In my research work, the modification indices of both the items I used is above 20, and I have also got a great model fit after the error correlation. However, I am worried about how to justify it now as in both the cases, the terms and wordings seem to be similar.
Â
Can anyone please suggest some way? Will it be enough if I provide justification on the wording? Will it prove the errors?
You can also share articles or other study material that can help me with my project. I will be really grateful.
Thank you.
Â
All Answers (2 Answers In All)
By Malvika Mathur Answered 4 years ago
In such a case, I guess I believe the face value becomes more important. As a general principle, if the method variance has been taken care of and the correlation becomes intact at face-value, then it becomes simpler to handle such errors.
For this to happen, there are a few musts that have to be there:
Two or more items in your collective data should have the same or similar options to respond.
These items or variables should differ from other variables belonging to the same latent variable.
Same likert scale can be used but the result should be different. Like some items can be portrayed through positive words and other items with negative words.
After reversing the items which are negatively worded, they should not affect the tendency shown by the positive worded items.
Even after the reverse, the negative worded elements should remain more correlated to each other.
Items administered at the same situation should be more correlated than the items administered at different situations.
Â
Like this, when you allow the residuals to be correlated with each other nicely and clearly, it becomes easier to provide theoretical justifications and these occasional errors can be handled well too.
Replied 4 years ago
By Shabana
Reply to Malvika Mathur
By Jaafar Answered 4 years ago
However, I still hope that you pay attention to different studies. I will also look into this topic and will share articles or research papers, if I come across any.Â
Replied 4 years ago
By Shabana
Reply to Jaafar
Related Questions