Mean and variance quality control for multiple correlated levels of replicated control samples.

摘要: Mean and variance rules for quality control are more powerful than based on individual values. An algorithm applying such is described that controls type I errors (false alarms), while allowing multiple levels of samples, correlation between levels, small numbers preliminary values, replication samples autocorrelation arising from random effects. Based ANOVA empirical approximations the maintains a low per-batch probability errors. Three statistics computed, z m , b w which shown by simulations to be primarily sensitive concordant shift in discordant an increase variability, respectively. Simulations also show Gaussian distribution analytical errors, error likely within range 0.0045-0.0071 two four where there 20-100 batches inter-level correlations zero 0.8. This partial separation out-of-control alarms into three components provides assistance with troubleshooting do multivariate schemes Hotelling's T 2 retaining comparable power.