Categorical data analysis in primary care research: log-linear models.

Abstract

Primary care researchers often wish to perform multiple variable analyses using variables measured at a nominal or ordinal level. This paper provides a step-by-step description of log-linear modeling, an approach uniquely well suited to explore and describe interactions among three or more nominal or ordinal variables. The method of log-linear analysis is illustrated with the use of an example from a primary care research project in which the relationships among hypertension, diet, and sodium were examined. The advantages and disadvantages of log-linear models and logistic regression are compared and available computer software programs discussed.