ABSTRACT

Two types of categorical variables are distinguished: nominal and ordinal. This chapter describes methods that can be applied for analyzing both nominal and ordinal random variables. Special methods for analyzing ordinal data are available; they offer the potential of more power. The chapter covers some basic extensions relevant to contingency tables where there are more than two categories. Another extension covered in the chapter deals with regression where the dependent variable is binary. The chapter focuses on a generalization of the binomial probability function to situations where observations are classified into one of K categories. It takes up the common goal of measuring the association between two categorical variables. The p-value associated with the chi-squared test for independence might seem reasonable, but it is known to be unsatisfactory. Some of the measures of association that have proven to be useful in applied work are based in part on conditional probabilities.