ABSTRACT
An approach to robust sphering which involved ‘turning’ later attributes to make them ‘roughly orthogonal’ to earlier attributes has been described earlier. Here, ‘later’ and ‘earlier’ refer to position in a list of attributes ordered by judgment importance. The description was done as a step on the way to ‘Euclidean distances’ and ‘fractionation.’ This chapter explores what else we might use the process for and how does it relate, for instance, to upsweeping into subtables. Where we do analysis of variance separately for two attributes, we have one G×E table for attribute 1, and one G×E table for attribute 2 (these might be pure interaction subtables, or combinations with the G subtable). A regression coefficient has been found for regression of the entries of the latter on the corresponding entries of the former.
