Disjoint 2 stage approach; HTMT; Measurement Model
Posted: Thu Jul 09, 2020 6:35 pm
Dear experts,
I would greatly appreciate your feedback on the following queries. I'm still new to PLS-SEM but I'm trying to learn from all the latest articles and posts in the forums.
I have a model that has three 2nd order constructs (reflective-reflective) and 4 first order constructs. I decided to go with the 2 stage disjoint approach. 1 of my variables, "Occupational Stress" has 12 sub-scales as per the validated OSI Scale(Srivastava & Singh, 1984). As per the 2 stage disjoint approach, I connected all the lower order constructs (without higher order in the path and as per relationship expected with the higher order). That model looks like a horrible spider web in the end. I'm confused about a few things:
1. Since the model looks very messy because of so many arrows, should I be concerned (even though this is stage1)?
2. I find that HTMT Inference violations are mainly there in my Occupational Stress Scale with values ranging between 1.004 to 1.259 (at 10% two tailed, bootstrapping at 5000). Example, Role Conflict and Role Ambiguity; Low status and Poor peer relations; etc. Having read the previous comments by experts here on HTMT and the article "A new criterion for assessing discriminant validity in variance-based structural equation modeling", I am confused because the sub-scales of OSI anyhow have few items each. So, item deletion becomes tough. Further, I remember reading that discriminant validity issues need to be handled creatively (as per Dr. Becker's comment in a post), hence checking if there any additional information which may guide me.
3. Can we proceed if Cronbach alpha and Rho is poor but CR and AVE thresholds are met? I'm facing this issue in sub-scales that have 2-3 items left after deletion.
Thanks in advance.
Navya Jain
I would greatly appreciate your feedback on the following queries. I'm still new to PLS-SEM but I'm trying to learn from all the latest articles and posts in the forums.
I have a model that has three 2nd order constructs (reflective-reflective) and 4 first order constructs. I decided to go with the 2 stage disjoint approach. 1 of my variables, "Occupational Stress" has 12 sub-scales as per the validated OSI Scale(Srivastava & Singh, 1984). As per the 2 stage disjoint approach, I connected all the lower order constructs (without higher order in the path and as per relationship expected with the higher order). That model looks like a horrible spider web in the end. I'm confused about a few things:
1. Since the model looks very messy because of so many arrows, should I be concerned (even though this is stage1)?
2. I find that HTMT Inference violations are mainly there in my Occupational Stress Scale with values ranging between 1.004 to 1.259 (at 10% two tailed, bootstrapping at 5000). Example, Role Conflict and Role Ambiguity; Low status and Poor peer relations; etc. Having read the previous comments by experts here on HTMT and the article "A new criterion for assessing discriminant validity in variance-based structural equation modeling", I am confused because the sub-scales of OSI anyhow have few items each. So, item deletion becomes tough. Further, I remember reading that discriminant validity issues need to be handled creatively (as per Dr. Becker's comment in a post), hence checking if there any additional information which may guide me.
3. Can we proceed if Cronbach alpha and Rho is poor but CR and AVE thresholds are met? I'm facing this issue in sub-scales that have 2-3 items left after deletion.
Thanks in advance.
Navya Jain