The Principle of End Weight maintains that constituents will occur in order of increasing weight (Behagel 1909; Quirk et al. 1985; Wasow 2002), where "weight" has been contentiously defined in terms of syntactic (Hawkins 1994), processing (Gibson 1998; 2000), or phonological complexity (Zec and Inkelas 1990; Anttila et al. 2010; a.o.). This talk presents a systematic investigation of the behavior of five previously proposed end weight measures, including the number of syntactic nodes, words, lexical stresses, syllables, and discourse-new referents, in constituent ordering in two constructions of spoken American English: the genitive (the car’s wheel ~ the wheel of the car) and the dative (give the dog the bone ~ give the bone to the dog). The measures of end weight were examined using two relatively novel statistical techniques -- information-theoretic model averaging (Burnham and Anderson 2004) and conditional random forests analysis (Strobl et al. 2009) -- that better handle highly collinear data than standard regression analyses. Of the weight measures tested, we find the number of syntactic nodes and lexical stresses to be the most highly predictive of construction choice, while lower-level phonological measures are not as reliable in capturing the weight-driven effects of high-level constituent ordering.