The question of how data collected in the spin-\textonehalf{} Einstein-Podolsky-Rosen experiment become compatible with local realism for increasing detector error is investigated in two ways. (1) It is shown that data obtained in any experiment with three settings of one detector axis and any number of settings of the other will be compatible with local realism if the detector error is such that data obtained with any two settings of each detector axis are compatible with local realism. (The error rate must be more than 7.96% for this to be so.) (2) It is shown that a detector error rate of 14.64% is enough to mask nonclassical correlations in any experiment with any number of settings of both detector axes; this improves on the figure of 21.13% given by Mermin and Schwarz.