Whoa wait. His rationale for 5 teams being outliers seems pretty weak. If I'm reading this right the outliers make up around 17% his data points as well. Seems like a lot of data to disregard as "outliers".
This seems more like data shaping than noise reduction.