On regression modeling in varieties research

Abstract

One particularly prominent methodological development in linguistics is what has been termed the “quantitative turn”: Not only are more and more studies using statistical tools to explore data and to test hypotheses, the complexity of the statistical methods employed is growing as well. This development is particularly prominent in all kinds of corpus‐linguistic studies: 20 years ago chi‐squared tests, t‐tests, and Pearson's r reigned supreme, but now more and more corpus studies are using multivariate exploratory tools and, for hypothesis testing, multifactorial predictive modeling techniques, in particular regression models (and, increasingly, tree‐based methods). However welcome this development is, it, and especially its pace as well as the fact that few places offer rigorous training in statistical methods, comes with its own risks, chief among them that analytical methods are misapplied, which can lead imprecise, incomplete, or wrong analyses. In this paper, I will revisit a recent regression‐analytic study in the research area of English varieties (on clause‐final also and only in three Asian Englishes) to: highlight in particular three fundamental yet frequent mistakes that it exemplifies; discuss why and how each of these mistakes should be addressed; reanalyze the data (as far as is possible with what is available) and show briefly how that affects the analysis's results and interpretation.