Moderate censoring, no effect (medians 8 and 8): control values below the LOD 28%; p<0.05 with LOD/2 substitution 5.1%, Mann-Whitney 4.7%, left-censored model 7.0%; median ratio estimate: substitution 0.99, model 0.99; median control log-SD after substitution 0.83 (true 0.80)
Moderate censoring, twofold effect (8 and 16): control values below the LOD 28%; p<0.05 with LOD/2 substitution 51.7%, Mann-Whitney 48.8%, left-censored model 58.7%; median ratio estimate: substitution 2.07, model 2.00; median control log-SD after substitution 0.83 (true 0.80)
Heavy censoring, no effect (medians 5 and 5): control values below the LOD 50%; p<0.05 with LOD/2 substitution 4.7%, Mann-Whitney 4.3%, left-censored model 5.7%; median ratio estimate: substitution 1.01, model 1.01; median control log-SD after substitution 0.73 (true 0.80)
Heavy censoring, twofold effect (5 and 10): control values below the LOD 50%; p<0.05 with LOD/2 substitution 50.2%, Mann-Whitney 47.2%, left-censored model 54.6%; median ratio estimate: substitution 1.98, model 2.00; median control log-SD after substitution 0.73 (true 0.80)