Post-EstimationPenalization: more ‘pep’ for estimates of genetic covariancematrices

Optimum ψ (for population values known) – Penalty reduces average loss by ∼ 60− 70% – 2-step method (PEP) almost as effective as penalised estimation (MVP) Pick ψ so that ΔL ≈ −2χ 5%,1 – change in logL not significant – under-penalisation, but achieve substantial proportion of reduction in loss possible – results for PEP more variable than for MVP Little difference between types of penalty Results Distribution over replicates P:none P:eigenvalues P:correlations