peer reviewed ; In syntactic alternation studies, the level of abstraction at which to investigate the variant distribution at issue is commonly determined a priori. For instance, the English conative alternation may be lexically delineated only by the preposition at, with the subject, verb and object slots in principle kept open (Perek 2015). In other occasions, an alternation may be limited to only the instances of a single verb and a single preposition, as in Pijpops et al. (2018), or researchers may focus on even more concrete lexical combinations, as in Lehmann and Schneider (2012). Still, the most appropriate level of abstraction at which to investigate a specimen of language variation is not always a given. To illustrate this, we will look at the alternation between the direct and prepositional object in Dutch, as in (1)-(2). (1) Voor iedereen dien je (naar) een individuele oplossing te zoeken. for everyone ought you (to) an individual solution to search (Sonar corpus, id: WS-U-T-B-0000000143.p.7.s.3, Oostdijk et al. 2013) ‘You ought to search for an individual solution for everyone.’ (2) Maar Caroline weerstaat (aan) de verleiding. but Caroline withstands (to) the temptation (WR-P-P-G-0000539169.p.2.s.4) ‘But Caroline withstands the temptation.’ We ask at which level of abstraction this variation functions. That is, should we consider occurrences such as (1)-(2) as instantiating a single alternation between an abstract transitive and prepositional construction, or should we look at each preposition, verb, or even object separately? In particular, we will focus on the influence of lectal predictors. To investigate this, we propose a corpus-based procedure that involves two steps. The first step defines the several levels of abstraction, ranging from lexically specified syntactic slots to entirely open slots. In second step, we fit regression models at each level. From our first results, it appears that the influence of lectal predictors seem to be tied to specific lexical elements. For instance, we find ...