A method is given to determine whether or not the distribution functions describing the two spin measurements in the spin-s Einstein-Podolsky-Rosen experiment are compatible with the existence of distributions describing three spin measurements (not all of which can actually be performed). When applied to the spin-1/2 case the method gives the results of Wigner, or of Clauser, Holt, Horne, and Shimony, depending on whether or not the two-spin distributions are assumed to have the forms given by the quantum theory. Generalizations (...) of the conditions of Wigner or of Clauser et al. to the spin-1 case are explicitly calculated. The spin-3/2 case is examined in some simple geometries to show that an apparently monotonic trend toward local realism as s increases from1/2 to1 is, in fact, violated when s increases from1 to3/2. The analysis is based on a novel representation of the modulus squared of a rotation matrix element. The structure of that matrix element responsible for the restoration of local realism in the classical (large s) limit is identified, but a rigorous treatment of the classical limit is not attempted. The higher-spin results are significantly stronger than those given by Mermin's spin-s Bell inequality. (shrink)