The mechanisms underpinning the formation of a focused volcanic arc above subduction zones are debated. Suggestions include controls by: (i) where the subducting plate releases water, lowering the solidus in the overlying mantle wedge; (ii) the location where the mantle wedge melts to the highest degree; and (iii) a limit on melt formation and migration imposed by the cool shallow corner of the wedge. Here, we evaluate these three proposed mechanisms using a set of kinematically-driven 2D thermo-mechanical mantle-wedge models in which subduction velocity, slab dip and age, overriding-plate thickness and the depth of decoupling between the two plates are systematically varied. All mechanisms predict, on the basis of model geometry, that the arc-trench distance, D, decreases strongly with increasing dip, consistent with the negative D-dip correlations found in global subduction data. Model trends of sub-arc slab depth, H, with dip are positive if H is wedge-temperature controlled and overriding-plate thickness does not exceed the decoupling depth by more than 50 km, and negative if H is slab-temperature controlled. Observed global H-dip trends are overall positive. With increasing overriding plate thickness, the position of maximum melting shifts to smaller H and D, while the position of the trenchward limit of the melt zone, controlled by the wedge's cold corner, shifts to larger H and D, similar to the trend in the data for oceanic subduction zones. Thus, the limit imposed by the wedge corner on melting and melt migration seems to exert the first-order control on arc position.