It really comes down to the materials and the phenomenon they experience. Here are some notes from the manual on Failure theories that might help.
The reliability of this failure theory depends on the accuracy of calculated results and the representation of stress risers (peak stresses). Stress risers play an important role if, for example, yielding at local discontinuities (e.g., notches, holes, fillets) and fatigue loading are of concern. If calculated results are suspect, consider the calculated stresses to be nominal stresses, and amplify the nominal stresses by an appropriate stress concentration factor Kt. Values for Kt are available in many strength of materials handbooks.
If fatigue is not a concern, localized yielding will lead to a slight redistribution of stress, and no real failure will occur. According to J. E. Shigley (Mechanical Engineering Design, McGraw-Hill, 1973), "We conclude, then, that yielding in the vicinity of a stress riser is beneficial in improving the strength of a part and that stress-concentration factors need not be employed when the material is ductile and the loads are static."
Alternatively, localized yielding is potentially important if the material is marginally ductile, or if low temperatures or other environmental conditions induce brittle behavior.
Yielding of ductile materials may also be important if the yielding is widespread. For example, failure is most often declared if yielding occurs across a complete section.
The proper selection and use of a failure theory relies on your engineering judgment. Refer to engineering texts such as Engineering Considerations of Stress, Strain, and Strength by R. C. Juvinall (McGraw-Hill) and Mechanical Engineering Design by J. E. Shigley (McGraw-Hill) for in-depth discussions on the applied theories.