ABSTRACT
It is not uncommon for an analyst to have to consider hierarchy in data. Sometimes this hierarchy can be explicit and visible in the data and other times there is a belief that the data represents some latent, invisible factors at play. The first part of this chapter covers modeling situations where the data has an explicit hierarchy. This can happen when observations are of individuals who are members of groups, or where multiple observations are taken of the same individual over time. The concept of mixed (hierarchical or multilevel) models that have fixed effects and random effects are introduced. Data from a speed dating experiment is used to illustrate a mixed binomial logistic regression model and time is spent illustrating the difference when interpreting a mixed model versus a standard fixed effects model. The second part covers a common situation where large, often disorganized, survey instruments need to be modeled. Using the example of a survey on voting preferences, structural equation modeling is introduced as a method used to confirm a smaller set of unmeasured latent variables which explain the responses to the measured survey items (confirmatory factor analysis), and then explain the outcome of interest against the latent variables (structural modeling).
