Influence functions and efficiencies of the canonical correlation and canonical vectors estimates

Taskinen, Croux, Kankainen, Ollila, Oja2006Plug-in asymptotics

Taskinen, Croux, Kankainen, Ollila & Oja (2006) — influence of canonical correlations

Result

influence functions of canonical correlations and vectors for general scatter-matrix functionals; for the sample covariance the squared correlation has IF⁡(ρi2)=2ρiuivi−ρi2(ui2+vi2)\operatorname{IF}(\rho_i^2)=2\rho_iu_iv_i-\rho_i^2(u_i^2+v_i^2) in the canonical variates ui,viu_i,v_i.

Use

the per-coefficient form whose average over ii, divided by ρi2\rho_i^2, is O7.2's ψ\psi; confirms the canonical-correlation check of the determinant-route influence function. Centred scatter and a random second block; O7 is uncentred with a fixed partition indicator.

Verification

formula seen through the citing restatement in arXiv:1705.04194 (Sinha et al. 2017) — the paper itself was not retrievable (AUDIT-RETENTION-PLUGIN-VECTOR, 6 Sep 2026).

S. Taskinen, C. Croux, A. Kankainen, E. Ollila and H. Oja, Journal of Multivariate Analysis 97(2):359-384, 2006. Restates the influence function of a squared canonical correlation, IF(rho_i^2) = 2 rho_i u_i v_i - rho_i^2 (u_i^2 + v_i^2) in the canonical variates, the form O7.2's per-coefficient check uses. Registered 6 Sep 2026 (audit); formula seen only through a citing restatement (arXiv:1705.04194).

Cited by