the stress values at nodes are never uniquly defined. The only location where the "true" stresses are calculated are the integration points. Any stress or strain value at nodes are a result of a postprocessing, and there are many ways to do that. However, if the gradients of stresses are not large, the difference between the "nodal stresses" and the "element stresses" should not be huge. This difference is actually used as an error estimator to check if the mesh is fine enough. I would need to see a plot in order to make further comments. Feel free to share!