Our goal is to build a three-level hierarchical linear model based on complex survey design in Stata. I am struggling to combine the svy and meglm commands when our nested variables for the model do not relate to our stratified sampling key variables.

Background on the survey design: Our data is from a stratified sampling design. Our primary strata are 10 intervention regions. Intervention areas mapped onto small areas layers (SALs) covering targeted areas. Across the 10 regions, there were ~3000 SALs targeted for intervention. We selected a simple random sample of SALs in the intervention areas in each region. A systematic random sample of available households within each sampled SAL was selected to obtain required sample size (sample size per region was designed to be proportional to the population in the sub-region areas selected for intervention). The svyset command is as follows:

Code:
 svyset sal, strata(region) fpc(sal_tot) weight(sample_weight) vce(linearized) singleunit(missing)
I understand that meglm is compatible with svy in Stata: after specifying svyset, we can call svy: meglm. However, after calling the following lines:

Code:
 svyset sal, strata(region) fpc(sal_tot) weight(sample_weight) vce(linearized) singleunit(missing)

svy: meglm outcome gender alcohol drugs poverty age || relational_factor1: || relational_factor2:
I get the following error message:

Code:
 hierarchical groups are not nested within sal
an error occurred when svy executed meglm
All examples online that combine svy and a hierarchical model have a survey design whose "levels" (clusters, strata, primary sampling units, etc.) are the same levels used in the hierarchical model. For example, say that schools were sampled, then classrooms from each school were sampled, then students from each classroom were sampled - the hierarchical model examples based on this survey design simply seek to create nests where level 1 is student, level 2 is classroom, and level 3 is school.

Our research question for this model, though, is different - we are looking at individual- (level 1), interpersonal- (level 2), and community-level (level 3) factors predicting outcome. These variables do not have anything to do with the SALs or regions that make up our complex survey design. Based on my research, I am thinking that I might need to specify weights at each higher level in the svy: meglm call. However, I am not sure how to calculate these weights, given that the relational_factor1 and relational_factor2 variables are not SAL- or region-based... Any help?