I want to estimate a multilevel model with 2 DVs and compare the coefficients of a contextual predictor. The participants are ethnic minority members. One DV is contacts with natives and the other contacts with other minorities. I have data from 50 municipalities, and my contextual predictors are percentage of natives in the municipality (for predicting contacts with natives) and percentage of ethnic minorities (excluding one's own ethnic group) (for predicting contacts with other minorities).
For % of natives all minority members in the same municipality get the same value (=50 higher-level units). For % of other minorities the values differ per ethnic group. This means that my higher level has to be ethnic group in municipality. I have 4 ethnic groups, which adds up to 200 higher-level units. I made an id variable with 200 values and used it for specifying the nesting.
If I estimate a model like this it gives me coefficients for both contextual predictors and their respective relations with the two DVs, so technically it works. However, the estimates for the relationship between % of natives and contact with natives are based on fewer values than the estimates for the relationship between % of other minorities and contact with other minorities. Is this a problem and can I at all compare the coefficients? If not, what would be an alternative in MPlus? Thank you
I am not convinced that your approach of working with 200 higher-level units is correct. You may think of how it would differ from doing a separate 2-level analysis for each ethnic minority, in which you would retain the original number (50) of higher-level units.
For within-cluster groupings such as ethnic minorities within municipalities, see also the new web note on our web site:
Asparouhov & Muthen (2012). Multiple group multilevel analysis. Web note 16.
For more feedback, you may also want to post on Multilevelnet.