**Abstract** The present paper explores the variable nature of speakers’ (probabilistic) grammar by analysing variation in a well-known syntactic variable, namely the English dative alternation, across World Englishes. While the numerous constraints that impact the choice of dative variant are well-known, the extent to which the regional variability of these constraints in specific registers or lexical items can be generalized to the language as a whole has largely remained understudied (but see, for instance, Bresnan and Hay 2008; Bresnan and Ford 2010; Röthlisberger, Grafmiller, and Szmrecsanyi 2017). Building on recent work (Röthlisberger et al. 2017), the current study takes a comprehensive large-scale comparative perspective across nine varieties, 14 different registers and including 86 alternating verbs (N= 13,171) to showcase how restrictions in the dataset (e.g. to specific registers or verbs) can result in misleading generalizations. Relevant observations were annotated for approximately 10 probabilistic constraints. Results of a mixed-effects regression analysis indicate that the factors regionally variable across the whole dataset might not be regionally variable across smaller speaker groups or registers and vice versa, thereby challenging the methodological set-up of earlier studies on the English dative alternation. In doing so, the present paper not only confirms the variable nature of probabilistic grammars but also stress the importance of combining an aggregate perspective with more fine-grained analyses to grasp more fully the cross-lectal variability of speakers’ grammatical knowledge.