Just as in all scientific work, we strive in variation studies for suitable levels of abstractness and thereby generality. Although it is not always appreciated, even the classical studies variation, e.g. those on the second sound shift in Continental Germanic (see Bloomfield), abstract a good deal away from details in the fieldworkers' raw data. As soon as we abstract into categories, we face the problem of defining the categories. (In a less vegetarian era, categorization was compared to butchery, in particular the need to carve nature "at its joints" (Plato, Phaedrus 265d-266a).)
The problem crops up in phonological, lexical and morpho-syntactic research, but it is especially virulent where categorizations are relatively unconstrained, i.e. where there is little rigor. Rigor constrains the sorts of analyses (and categories) one accepts and is therefore often resisted by critically oriented practitioners, who can spot gaps and flaws in any rigorous scheme. The way forward is to accept rigorous classifications for the purpose of comparison, and these must be modest if they're to be accepted.