Multilevel Modeling

6. Multilevel Data Structures


It is well known that once groupings are created (consisting of individuals), even if their origins are essentially ‘random,’ individuals end up being influenced by their group membership. Such groupings can be spatial (e.g., areas) or non-spatial (e.g., communities). Hierarchies are one way of representing the dependent or correlated nature of the relationship between individuals and their groups. Thus, for instance, we can conceptualize a two-level structure of many level-1 units (e.g., individuals) nested within fewer level-2 groups (e.g., neighborhoods/places) as illustrated in Figure 2. Since individual outcomes are anticipated as being dependent upon the neighborhoods in which they live, responses within a neighborhood are more alike than different. When dependency is anticipated in the ‘population’ or ‘universe,’ they represent population-based or naturally occurring hierarchies.

Figure 2

Figure of two level structure as described in text.

The importance of identifying and specifying the ‘higher’ levels is critical for multilevel research. Researchers must a priori specify why they think that there will be variation in the outcome at these levels over and above variation at the individual-level. Such thinking naturally leads to considerations of which levels to include in the model. For example, do we expect variation at the level of small neighborhoods (e.g., census blocks) or larger neighborhoods (e.g., census tracts)? The most common multilevel model is a two-level hierarchic nested modeling with many level-1 units within a smaller number of level-2 units, as exemplified in Figure 2. A multilevel structure can be cast, with great advantage, to incorporate a range of circumstances where one may anticipate clustering (Subramanian, Jones et al., 2003).
Subramanian, S., Jones, K., et al. (2003) Multilevel methods for public health research. In: I. Kawachi and L. Berkman (Eds.). Neighborhoods and Health. New York: Oxford Press. 65-111.