Abstract
Variation Theory, a subfield of sociolinguistics, underwent a scientific revolution in 2009, when a logistic regression analysis was replaced as an analytical tool by a generalized linear mixed model (GLMM). We first give a detailed theoretical and historical background of the revolution, followed by a demonstration of GLMM analysis that involves a reanalysis of the variation and change of the innovative potential suffix in Tokyo Japanese proposed by Matsuda (1993). Our GLMM reanalysis with two random intercepts (the speaker and the verbal stem) showed that, while the language internal (grammatical or phonological) factors mostly remained statistically significant, only a single explanatory variable, (normalized) age of the speaker, emerged as a significant external (social) factor. It also revealed that two verbal stems are exceptions in the sense that they behave quite differently from what is expected from the combinations of relevant factors. We conclude that the GLMM is better suited than the fixed-effects only logistic regression analysis to the studies of language variation and change, where overdispersion and large individual differences are more prevalent than generally assumed.