Thanks Samuel. In terms of SES as a confounding variable, this is complicated, as we discussed in class, since SES does not "cause" race/ethnicity. This points to the need to be thoughtful about what question we are most interested in and how the modeling we perform, including how we incorporate SES, affects which questions we are actually answering.
Effect modification is an additional option for SES in a model.
Your example for multilevel modeling is interesting - it would be informative to perform that kind of analysis and then look for explanatory variables for state-level variations.