Do components of smooth tests of fit have diagnostic properties?

Smooth goodness of fit tests were introduced by Neyman (1937). They can be regarded as a compromise between globally consistent (“omnibus”) tests of fit and procedures having high power in the direction of a specific alternative. It is commonly believed that components of smooth tests like, e.g., skewness and kurtosis measures in the context of testing for normality, have special diagnostic properties in case of rejection of a hypothesisH ₀ in the sense that they constitute direct measures of the kind of departure fromH ₀. Recent years, however, have witnessed a complete change of attitude towards the diagnostic capabilities of skewness and kurtosis measures in connection with normality testing. In this paper, we argue that any component of any smooth test of fit is strictly non-diagnostic when used conventionally. However, a proper rescaling of components does indeed achieve the desired “directed diagnosis”.     

THE CHOICE OF CELLS IN CHI–SQUARE TESTS

The choice of cells in chi–square goodness of fit tests is a classical problem. Some recent results in this area are discussed. It is shown that the likelihood ratio of alternatives w.r.t. null distributions plays a key role when judging different procedures. The discussion centers on the case of a simple hypothesis, but location–scale models and tests of independence in contingency tables are also considered.     

THE DISTRIBUTION OF GENERAL QUADRATIC FORMS IN NORMA

For a general quadratic form in normal variables a representation in terms of independently distributed standard normal variables is derived. The necessary and sufficient conditions for such a quadratic form to have a non–central chi–squared distribution can be found easily using this representation.     

INVARIANCE OF THE UNORDERED BEST ESTIMATOR IN THE T,–CLASS FOR MIDZUNO'S AND FOR IKEDA–SEN'S SAMPLING PROCEDURES

MIDZUNO'S sampling procedure is considered where the first (n – 1) draws are carried out with simple random sampling without replacement and the nth draw with varying probabilities. It is shown that for this scheme, the best estimator in the HORVITZ–THOMPSON (1952) T_t–class of linear estimators exists and rejects the last draw. When MURTHY'S technique of unordering of an ordered estimator is employed, the rejected draw is restored and the unordered estimator is obtained. Surprisingly, this unordered estimator is the same as the unordered best estimator in the T₁–class, derived for IKEDA–SEN'S sampling procedure.