Henry R. Neave

A Development of Tukey's Quick Test of Location

Abstract A simple test for recognising differences in mean is described. It has its origins in a test proposed by Tukey in 1959, and effectively ignores that piece of data which most seriously detracts from the significance of Tukeys statistic. Heuristic discussions and a reference to empirical evidence are given concerning the resultant improvement in power. The test is quick to execute both in formation of the statistic and in deciding on its significance, since in a wide class of situations the 5%, 1% and .1% critical values can be taken as 10, 13 and 17 respectively. Tables of critical values for small samples are however included, as are graphs based on asymptotic considerations.

A Comparison of Lag Window Generators

Abstract Spectrum estimates depend for their specification on a truncation point and a “covariance averaging kernel” or “lag window,” which is derived from a “lag window generator.” A comparison of such generators is presented, based upon their negated derivatives, which are shown to be the approximate weightings in a particular type of representation of the spectrum estimate. It is clearly shown why the well-known generators of Parzen and Tukey are superior to most (but not all) of their rivals. These still have weaknesses, however, and the article concludes with a suggestion how these may be overcome.

“I shall teach...the theory of a system, and cooperation”

Contends that Deming’s philosophy for management is imperfectly understood. Unlike just a few years ago, very many people have now heard of W. Edwards Deming, and know something of his philosophy for management. But “something” is, in most cases, far too little. His work is sometimes dismissed as “just statistics” or even as no more than statistical process control. Another common false impression is that his philosophy is “summarized by the 14 Points”. And one of the few things to continue to irritate him until his death in December 1993 was any attempt to connect him with “TQM”! Agrees that all such diminutions trivialize his legacy to us. His work was not only much larger than any of these imply: it was genuinely of a higher dimension. He said of the two books by his great mentor, Walter Shewhart, that a “century will pass before people in industry and in science begin to appreciate the contents of these great works”. The ominous truth is that the same could well be true of Deming’s own two great books on management. Fears that we cannot afford to wait that long.

Quick and Simple Tests Based on Extreme Observations

This is a survey of a number of quick and simple distribution-free tests (quick to execute and simple to understand) that have been developed over the last thirty years to solve a variety of problems. The typical test statistic consists of counting the ..

A Quick and Simple Technique for General Slippage Problems

Abstract In a recent article we described some quick and simple tests for the k-sample slippage problem. Those tests were designed for the alternative that precisely one population had slipped. In this article we present a method of testing which has power for all slippage alternatives and which makes a diagnosis of the type of slippage that has occurred. Critical values are obtained either from a table (in the case of small samples), from graphs or from asymptotic approximations, which turn out to be very accurate.

A New Look at an Old Test

Fishers exact test for 2 × 2 frequency tables is presented in a new format which, we believe, is easier to use and which permits a substantial saving in tabulation-space of critical values. Tables are provided here for total frequency N ≤ 25. Tables for N ≤ 100 are obtainable on request to the author

Percentage Points of Some Non-Parametric Tests for Independence and Empirical Power Comparisons

Abstract A class of nonparametric tests based on the third quadrant layer ranks has recently been studied by Woodworth [12] in connection with the problem of testing for independence in a bivariate distribution. In this article exact one-sided rejection regions are tabulated for the normal score layer rank test which is asymptotically locally most powerful for positive dependence in the bivariate normal distribution. The cut-off points are tabulated for sample sizes n = 4(1)9 and significance levels α = .10, .05, .025 and .01. Normal and Edgeworth approximations for the significance probabilities are also given. A simplified version of the normal score test is proposed, and its rejection regions are tabulated. These tests are compared with the correlation coefficient test, Kendalls τ test, Fisher-Yates (normal-score) correlation test and Spearmans rank correlation test for independence by means of Monte Carlo evaluation of power employing 10,000 trials from two different types of bivariate distributions...

Extending the Frequency Range of Spectrum Estimates by the Use of Two Data Recorders

Any device being used to record a time series has a natural maximum reading frequency which it is impossible or impractical to exceed, and this limits the absolute frequency range over which a spectrum estimate may be made. It is shown that if one may use two such devices, which do not have the same sampling period, to measure the the same process, then spectrum estimates may be made over a wider absolute frequency range than would normally be possible. Upper bounds to the asymptotic variance of such estimates are calculated and discussed. The situation is treated as a particular type of missing data problem, and part of the work presented here applies to a larger class of missing data problems (those with periodic amplitude modulating function.)