A popular method for studies of seismic wave speed anisotropy within the Earth is based on observations of birefringence in subvertically propagating shear waves. A well-recognized shortcoming of the method is that individual birefringence measurements, commonly interpreted as a path-average of anisotropy, do not reveal the potential complexity of anisotropic structure. Consequently, there is a necessary ambiguity in the interpretation of observed birefringence parameters (fast polarization directions and delays between components). In this paper, we explore the influence of observational geometry and the configuration of an anisotropic region on the birefringence parameters. We use full wavefield simulations in 3-D models to access these influences in a subduction zone setting. Our primary findings are that (a) modest wavefield complications associated with the non-planar geometry of a subduction zone can significantly bias the determination of the fast polarization and time delay associated with shear wave birefringence; (b) shear wave splitting associated with flow in the mantle wedge near the trench (e.g. the forearc) is likely to be weak, owing to short path lengths through an anisotropic region and (c) popular splitting estimators are vulnerable to waveform distortions caused by S-wave interaction with the dipping slab interface. We show that simultaneous interpretation of numerous birefringence observations from a variety of propagation directions is a more reliable way to constrain parameters of anisotropy at depth.