Multiple Variables in the Highest Level of Hierarchical Models of Education: Limits and Possibilities

• In Event: 207 - Section on Sociology of Education Roundtable Session (one-hour).
  In Refereed Roundtable: Table 18. Variation in the Transmission of Status

Abstract

This paper discusses methodological issues associated with including multiple variables at the highest level of multilevel models when there is a small number of cases at that level of the analysis. Using an example from my own work, I discuss the procedure for coming up with the most inclusive model possible given the difficulty of creating a model that is completely inclusive of my level-3 variables. This discussion highlights the fact that, after all, hierarchical linear models are essentially models that include many interaction terms, and when there are too many variables at Level-3 and not enough degrees of freedom, there just may not be enough variation to include all variables at the same time. I argue that this trend is not a surprising one, and is likely the reason that scholars publishing articles using hierarchical models tend to focus on only a single, or at most a few, level-3 variables.

Author

• Stephanie M. Arnett, Tulane University