In this chapter I discuss the contribution which variable rule analysis makes to my study of the relative system, and show how some hypotheses about the types of rules needed to describe the variation can be tested using the Cedergren–Sankoff variable rule program. I will be particularly concerned with examining the kinds of arguments which have been advanced in favor of the claim that the program (and variable rule analysis) gives new and reliable evidence for judging certain linguistic issues.

Cedergren and Sankoff's (1974) work on the probabilistic component of variable rules has had a major impact on quantitative analysis in sociolinguistics. The ramifications of this shift in emphasis from frequencies to probabilities are enormous; yet, they have scarcely been treated cogently in their relation to the construction of a sociolinguistic theory which claims to have an empirical base. Labov (1975a: 228), for example, has maintained that the ‘new tools of probability theory’ give us a ‘dramatic increase in power and perception’. The claim about increased power is true, though, ironically perhaps, not in the way Labov seems to think (cf. Romaine 1981a).

Probability theories might be considered more powerful with respect to their domain of application because, strictly speaking, they cannot be confirmed or disconfirmed by empirical findings.