Multidimensional Analysis of Categorical Data: Log-Linear Models: An Introduction and Example from the Epidemiology of Drug Addiction

The method for analyzing log-linear models developed by L. A. Goodman (see SA 20:6/72F7979) is described & given a new application. It is used to identify higher-order interactions & correlations in the analysis of multidimensional cross-tabulations. Structural similarities with multipath variance analysis are pointed out. The procedure is demonstrated through a 6-dimensional table from the epidemiology of drug consumption in the canton of Zurich in 1971 & 1978. Analysis indicates that: the proportion of 19-year-olds with drug experience declined minimally from 1971 to 1978; males have considerably more drug experience than females; higher education leads to more experience; & the offspring of upper-income parents have more frequent drug experiences. It is concluded that the method: can also be applied to multidimensional cross-tabulations with 'structurally empty cells,' allows for the analysis of categorical as well as ordinal data, & can also be used for causal analysis. 6 Tables. HA Tr & Modified by S. Whittle.