How to analyze categorical dependent variables with more than two levels
What is it about?
As the state of the field advances empirically, sociolinguists are increasingly expected to utilize statistics in their data analysis. Some researchers have limited formal statistical training, and even for the more experienced researcher, the focus of model construction is often on the independent variables, e.g. interactions or multicollinearity issues. However, dependent variables with three or more variants require careful consideration. Building on Paolillo (2002), I show that identical binomial logistic regression models yield disparate results given differential treatment of a complex dependent variable. I conclude by offering concrete, hands-on advice for linguists working with their data in R with the goal of promoting judicious analyses among Hispanic sociolinguists.
Why is it important?
This chapter shows just how much the results of statistical analyses can differ when multi-level dependent variables are treated differently, and it encourages sociolinguists to carefully consider their treatment of dependent variables.