I have dyadic data with n=1024 customers linked specifically to 97 employees so a 1:n rather than 1:1. The samples are imbalanced.
I have measured multiple behavioral constructs for each customer and employee. In this analysis I am focusing on the latent variables A, B and C.
A is measured at the customer level with eight items. B and C are measured at the employee level with four items.
I am presuming that a multi-level model using customers at the BETWEEN level and employees at the WITHIN level can be used to model this.
An extract from my MPlus code is:
CLUSTER IS y4; ! The employee number %WITHIN% A by a1 a2 a3 a4 a5 a6 a7 a8; %BETWEEN% B by b1 b2 b3 b4; C by c1 c2 c3 c4; C ON B; Queries: 1. How then do I model A ON B?
I presume just having A ON B will work in the BETWEEN level.
Do I also set up A as a WITHIN variable (say A_E) and regress A_E ON B? Are there any variances or means that need to be fixed in this case for this to run?
2. I have a few occasions where the variance of a specific item (such as a2) at customer level is zero for specific employees. Is the only solution to remove all this particular employee-customer pairings?