Data Snooping and Multiple Outlier Testing

摘要

Data snooping, using Studentized or un-Studentized one-dimensional conventional hypothesis tests, is described as a special case of general multivariate linear hypothesis tests. The derivation of test statistics is based on the ideas of Allen J. Pope. Special attention is given to the concepts of internal and external reliability of networks, as defined by W. Baarda, for un-Studentized data snooping. Finally a heuristic procedure for the detection of multiple (or simultaneous) outliers in one adjustment, called 'iterated data snooping,' is described. The results of an experiment with simultated errors in the observations are given.