Methodological issues in categorical data analysis: Categorization, linearity, and response effects

The “General Linear Reality” view of the social world endorsed by analysis models assuming (underlying) continuous variables that are normally distributed is still prevailing in most of social and behavioral research. In this article both assumptions are discussed, arguing that it might often be better to treat (latent) variables as fundamentally categorical and showing that the assumption of linear relationships may mislead researchers and lead them to accept response effects that may not be there or are of a rather different nature.
Keywords: categorical/categorized variables, underlying latent variables, Pearson-Yule dispute, ordinal/linear relationships, (extreme) response effects, measurement equivalence