Teleseismic shear-wave splitting is a widely-used technique for measuring oriented fabric in the crust and upper mantle; such fabric is an important marker for current or past deformation. The technique yields both the orientation (fast direction) and cumulative intensity (split time) of the net fabric. However, published splitting results, particularly split times, can have puzzling inconsistencies that make mapping splitting over large areas challenging; semivariograms of compiled splitting results show a lack of spatial coherence in split time measurement when studies using different methods are combined. I present modelling work that demonstrates that these inconsistencies result from an inherent bias in splitting measurement, particularly pronounced for split time, that is sensitive to details of the data processing methods and is amplified by averaging single-event measurements. With a correct choice of averaging method (error-surface stacking), this bias can be mitigated sufficiently to allow split time to be mapped over large areas, as demonstrated using compiled data from western Canada. The results show strong spatially-coherent variations along the strike of the Cordillera, which may represent regions of dominant vertical vs. horizontal flow in the upper mantle, driven by complex Cordilleran active tectonics.