In variationist linguistics, the usual way of proceeding is to first identify some variable phenomenon in the data and then attempt to model the quantitative patterns in the corpus as accurately as possible. This still leaves a more general question: what kinds of quantitative patterns are predicted to be possible and what kinds of quantitative patterns are excluded by the underlying theory? In this talk, I note that Optimality Theory predicts two kinds of quantitative patterns: those that are independent of rankings and could not be otherwise, and those that depend on rankings and are predicted to vary from one case to the next. I illustrate this distinction from the metrical phonology of Finnish. In Finnish, metrical pressures yield a range of segmental effects, including lenition, fortition, and phonotactic gaps. I show that a small number of metrical constraints suffice to derive the observed quantitative patterns, independently of rankings. I then show that the same metrical problems are resolved differently in different morphological constructions, suggesting that these constructions involve different rankings.