Contents - Index

Singular value partitioning 

There are three options for Singular Value Partitioning (S.V.P.)

• Entry-focused: The singular value is entirely partitioned into the entry eigenvectors, i.e, alpha = 1, beta = 0. This SVP is needed for accurate comparison among the entries. This is also called row metric preserving, leading to "JK biplot".

• Tester-focused: The singular value is entirely partitioned into the tester eigenvectors, i.e, alpha = 0, beta = 1. This SVP is needed for accurate visualization of the relationship among the testers. This is also called column metric preserving, leading to a "GH biplot".

• Symmetrical: The singular value is symmetrically partitioned into the entry and the tester eigenvectors (alpha = 0.5, beta = 0.5). This SVP is most often used in AMMI analysis and other biplot analysis, but it is not ideal for visualizing either the relationship among entries or that among the testers.

Note: 

• There are numerous other possible ways to partition the singular values, and their interpretations are not defined.
• All singular value partitioning methods are equally valid in approximating the two-way table and in showing the which-won-where pattern.

Related topics: Singular value decomposition.