ABSTRACT
The models discussed so far assume a continuous dependent variable and a normal error distribution. If the dependent variable is a scale in which the responses to a large number of questions are summated to one score, the data generally approximate normality. However, there are situations in which the assumption of normality is clearly violated. For instance, in cases where the dependent variable is a single dichotomous variable, both the assumption of continuous scores and the normality assumption are obviously violated. If the dependent variable is a proportion, the problems are less severe, but the assumptions of continuous scores and normality are still violated. Also, in both cases, the assumption of homoscedastic errors is violated. This chapter treats multilevel models for these kinds of data.
