The best is to see initial details of the construction of any discrimination diagram. Application of the diagram for drawing meaningful conclusions requires that you plot data of rocks of similar composition. I mean, if a discriminant diagram is designed through use of basaltic rocks, it should be used for rocks of basaltic composition. It may not be appropriate to draw conclusions by plotting rhyolite or peridotite analyses on it. I think that Mullen was dealing with basaltic rocks and applying the diagram to ultrabasic rocks is a choice you can decide yourself.

Yes, Qasim Jan is correct: this diagram was constructed for basalts and cannot be used for peridotites or serpentinized. The best tool for peridotites is mineral chemistry, either najor elements (e.g., Choi et el., 2008, Contributions to Mineralogy and Petrology) or trace elements from LA-ICPMS (e.g., Jean et al, 2010, CMP; or Shervais et al, 2011, GSA Special Paper 480). Spinel is best all around but need pyroxene minor elements as well because of “overlap” in spinel compositions.

Dear Cheolwoo: most tectonic setting discrimination diagrams were designed for basaltic rocks, providing no fractioning, no cumulates and the smallest alteration possible. Basalts are usually primitive magmas, generated by partial fusion of mantle rocks, and they are quite widespread in all tectonic settings, the efectiveness of such diagrams could be very useful. As Jorge Costa says, they can also be applied to low grade metabasalts, and so we have done in metavolcanic rocks in Venezuela. As he also states, there is no tectonic setting for mantle rocks, their mineralogy and trace elements are used mostly to find out the depth from which they were exhumed to the surface, and perhaps the degree of hydrothermal alteration, including serpentinisation, which has affected them during their emplacement. Regards. Sebastian (Central Univ. of Venezuela, Caracas).

I recommend John's answer