AP Statistic for Identification of Outliers in Multi-Response Experiments with Correlated Errors

In multi response experiments where correlated errors are present, identification of outliers becomes a complex yet crucial task. Outliers not only distort the estimation of model parameters but also jeopardize the validity of statistical inferences drawn from the data. In the present study the statistic given by Andrews and Pregibon (AP) (1978) for detection of influential observations in linear regression is suitably modified for detection of outliers in multi – response experiments with correlated errors. The statistic was developed considering the data structure of auto – regressive order 1. The outlier detection is based on mean – shift method, in which the expected value of the outlying observation is different from the expected values of other observations. The developed statistic is successful in detection of outliers when tested upon simulated datasets.