The proper methods for analysis of ordinal variables, such as Likert-type items, have been subject of controversy for decades. Among other issues, the controversy concerns the problem of how one can prove that one measures what one is supposed to measure (Prytulac, 1975), whether or not measurement scales are relevant in statistical analysis (Gaito & Yokubinas, 1986), and how the scale of measurement is to be determined (Nunnally, 1967). Specific to the ordinal scale of measurement are the problems whether or not measurements on such a scale can be calibrated (Kampen & Swyngedouw, 2000), whether or not they can be interpreted as rough measures of continuous (underlying) variables (Yule, 1912; Kampen & Weeren 2017), and whether or not they may be modeled by parametric techniques (Boneau, 1961).

In this Friday Methods Session, rather than choosing a side in the controversy, a systematic account of possible approaches to regression-type analysis involving independent and/or dependent ordinal variables is given. Topics that will be covered include ordinal dummy-coding, polychoric correlation, and (non-linear) ordinal response models. Special attention is given to the analysis of Likert scales. The approach will be pragmatic as well as scientifically justifiable. Theory will be illustrated with practical examples. Students are assumed to be familiar with the basic principles of statistical analysis (hypothesis testing, correlation, simple and multiple regression, ANOVA).