A few years ago I became interested in claims made about the epistemological status of sociolinguistic methodology and, in particular, the so-called empirical foundations of a sociolinguistic theory. My concern with the nature of sociolinguistic methods and data grew out of some of the difficulties I encountered in trying to present a sociolinguistic description of some variables in Scottish English (cf. Romaine 1975). Some of the problems (e.g. continuous vs. discrete variation, levels of abstraction in the construction of sociolinguistic grammars, probabilistic rules) still bother me. I deal with them again here, but this time with reference to another descriptive problem, namely, variation in the relative marker in Middle Scots. This also leads to consideration of some new issues, e.g. the scope of sociolinguistic theory and the relevance of sociolinguistic methods to problems in historical syntax.

It will become apparent that I am using the term ‘sociolinguistic’ primarily in a narrow sense, i.e. to refer to the work which has derived from Labov (1966). I have concentrated on Labov's research program because it has been so influential; supported by a substantial body of empirical research, it represents one of the most concrete proposals yet made for a sociolinguistic theory. However, I also discuss Bailey's work; and I attempt to show that much of the controversy between the so-called quantitative (Labovian) and dynamic (Baileyan) paradigms results from a misunderstanding of the ontological status of some of the arguments and explanations which can be supported on the basis of sociolinguistic or variable data.