Showing papers in "The Statistician in 1989"
3,152 citations
TL;DR: The Prediction Matrix Role of Variables in a Regression Equation Effects of an observation on a regression Equation Assessing the Influence of Multiple Observations Joint Impact of a Variable and an Observation Assessing the Effect of Errors of Measurements.
Abstract: The Prediction Matrix Role of Variables in a Regression Equation Effects of an Observation on a Regression Equation Assessing the Influence of Multiple Observations Joint Impact of a Variable and an Observation Assessing the Effect of Errors of Measurements A Study of Model Sensitivity by the Generalized Linear Model Approach Computational Considerations. Appendix: Summary of Vector and Matrix Norms, Proofs of Three Theorems.
617 citations
295 citations
TL;DR: The Condorcet Jury Theorem as discussed by the authors states that if n = 2m+1 jurists act independently, each with probability p>1/2 of making the correct decision, then the probability h 2m + 1 (p) that the jury (deciding by majority rule) makes the correct decisions increases monotonically to 1 as m increases to infinity.
Abstract: The «Condorcet Jury Theorem» theorem states that if n=2m+1 jurists act independently, each with probability p>1/2 of making the correct decision, then the probability h 2m+1 (p) that the jury (deciding by majority rule) makes the correct decision increases monotonically to 1 as m increases to infinity. Condorcet argued therefore that there are situations in which it is advisable to entrust a decision to a group of individuals of lesser competence than to a single individual of greater competence. Some generalisations of this theorem will be presented in which (a) voter competencies vary, and (b) there is a dependence between voter decisions. Furthermore, the concept of an indirect majority system will be discussed and compared with a simple majority system
250 citations
152 citations
130 citations
TL;DR: The book of plane answers to complex questions the theory of linear models, as an amazing reference becomes what you need to get.New updated as mentioned in this paper The latest book from a very famous author finally comes out.
Abstract: New updated! The latest book from a very famous author finally comes out. Book of plane answers to complex questions the theory of linear models, as an amazing reference becomes what you need to get. What's for is this book? Are you still thinking for what the book is? Well, this is what you probably will get. You should have made proper choices for your better life. Book, as a source that may involve the facts, opinion, literature, religion, and many others are the great friends to join with.
116 citations
86 citations
TL;DR: Equality of News Z- and T- tests, Balanced ANOVA; Homogeneity of Variance Tests; and Contingency Table Analysis.
Abstract: CHAPTER 1 Introduction CHAPTER 2 General Concepts CHAPTER 3 The Pivotal Case Intraclass Correlation CHAPTER 4 Equality of News Z- and T- tests, Balanced ANOVA CHAPTER 5 Correlation Coefficients CHAPTER 6 Linear Regression CHAPTER 7 Homogeneity of Variance Tests CHAPTER 8 Binomial Tests CHAPTER 9 Contingency Table Analysis CHAPTER 10 Conclusions
TL;DR: In this paper, an account of the methods of collecting and processing data about the occurrence and consequences of road accidents, and numerous statistical tables as Examples of the findings are given, with a focus on what can and cannot be found in statistics that are routinely collected -special research studies are outside its scope.
Abstract: The book is an account of the methods of collecting and processing data about the occurrence and consequences of road accidents, and gives numerous statistical tables as Examples of the findings. Chapters 1 and 2 contain aggregate statistics from many countries. Chapters 3 to 7 deal with various aspects of data collected by the police about road accidents, and chapters 8 to 13 discuss how medical sources can supplement the police data, especially in respect of the nature of injury sustained. Finally, chapter 14 examines insurance data. The book concentrates on what can and cannot be found in statistics that are routinely collected -special research studies are outside its scope, though naturally reference is made to some. (TRRL)
TL;DR: In this article, the authors demonstrate that naive application of the usual derivative formula to estimate changes in probabilities caused by changes in independent variables in probabilistic choice models can yield nonsensical results.
Abstract: This note demonstrates that naive application of the usual derivative formula to estimate changes in probabilities caused by changes in independent variables in probabilistic choice models can yield nonsensical results. This can occur because the usual derivative formula does not constrain the estimated change in probability. When the probabilities are large, the usual derivative expression can lead to violations of the constraint that probabilities must sum to one. A better approach is to use the cumulative probability functions to calculate the difference in probabilities. Since this answer is based on the cumulative probability function, the answer cannot be implausibly large in absolute value. This alternative calculation eliminates the inconsistencies that are possible when the derivative expression
TL;DR: The use and interpretation of regression diagnostics are discussed along with recent developments in this area, including their incorporation into the major statistical packages.
Abstract: In the past 10 years, many techniques have been developed for identifying unusual data and evaluating their effects on least squares regression. The rapid acceptance of regression diagnostics by practitioners is evidenced by their incorporation into the major statistical packages. As with other statistical techniques, the widespread availability of regression diagnostics is accompanied by the potential for their misuse. In this article, the use and interpretation of regression diagnostics are discussed along with recent developments in this area. Two real data examples are used to demonstrate these techniques.
TL;DR: It is proven that a sign test which takes ties into account may be uniformly most powerful, while omitting ties as proposed by several authors may lead to anti-conservative results.
Abstract: This note addresses the problem of ties in the sign test for discretised data. It is proven that a sign test which takes ties into account may be uniformly most powerful, while omitting ties as proposed by several authors may lead to anti-conservative results.
TL;DR: The experience in the use of centrally held general practitioner lists as a sampling frame in a major survey of cardiovascular risk factors is reported and there was considerable variation in the extent of sampling frame problems, especially of foreign elements.
Abstract: The experience in the use of centrally held general practitioner lists as a sampling frame in a major survey of cardiovascular risk factors is reported. These lists contain details of sex and date of birth and thus permit age-sex specific sampling. The lists were held as alphabetical card indexes, year of birth card indexes and computerised registers so that different procedures had to be used to sample from each type. Particular benefit was obtained from the use of microcomputers to assist in sampling and data entry at remote sites. There was considerable variation in the extent of sampling frame problems, especially of foreign elements.
TL;DR: In this paper, the authors compare sieve sampling with random and cell sampling of monetary units in terms of the behavior of the Stringer and cell bounds for two methods of assigning errors to the constituent monetary units of a line item.
Abstract: Sieve sampling has been proposed as a method of selecting a sample of monetary units in auditing. The primary purpose of this paper is to compare sieve sampling with random and cell sampling of monetary units in terms of the behavior of the Stringer and cell bounds for two methods of assigning errors to the constituent monetary units of a line item. A simulation study based on two actual accounting populations was conducted
TL;DR: A cumulative sum control chart scheme is designed for the detection of the outbreak of an epidemic to demonstrate the use of CUSUM chart in the non-manufacturing sector.
Abstract: A cumulative sum control chart scheme is designed for the detection of the outbreak of an epidemic to demonstrate the use of CUSUM chart in the non-manufacturing sector. The designed schemes are then applied on data on diabetic disease to illustrate the detection of outbreak of epidemic and also a retrospective analysis is carried out and the local means are computed
TL;DR: In this article, generalized binomial models are proposed to model overdispersed and under dispersed binomial data and goodness-of-fit tests are described to test whether the binomial distribution is sufficient or whether a more general model is needed to evaluate the historical control assumption.
Abstract: Active control equivalence studies, with the goal of demonstrating therapeutic equivalence between a new and an active control treatment, are becoming more widespread due to current therapies that reflect previous successes in the development of new treatments. An important feature of such studies is the reliance on an implicit 'historical control assumption'; that is, whether the active control would have performed significantly better than a placebo, had a placebo been used in the trial. A quantitative approach to examine this assumption is described in our paper. Generalised binomial models are proposed to model overdispersed and underdispersed binomial data. Goodness-of-fit tests are described to test whether the binomial distribution is sufficient or whether a more general model is needed to evaluate the historical control assumption. These methods are illustrated using one-week mortality data following myocardial infarction. The steps we describe to establish the superiority of an active control relative to placebo are recommended in the design and interpretation of active equivalence control studies.
TL;DR: In this paper, a quadratic loss function based on the intrinsic differences among the categories is introduced for estimating the cell probabilities of a multinomial distribution, such loss functions include many commonly used losses as special cases.
Abstract: For estimating the cell probabilities of a multinomial distribution, a quadratic loss function based on the intrinsic differences among the categories is introduced. Such loss functions include many commonly used losses as special cases and can be interpreted as divergence measures induced by quadratic entropy. For this class of loss functions and a single population, the Bayes estimator, the minimax estimator and a pseudo-Bayes estimator are obtained. Also, for estimating the cell probabili- ties of several multinomial populations, empirical Bayes and pseudo-Bayes estimators are discussed and illustrated with an example.
TL;DR: In this paper, the authors describe the small sample behavior of some members of k-class estimators in a number of specific cases by the means of a Monte Carlo experiment and show that the OLS estimator of structural coefficients (predictor of the mean of consumption variable) is the best in the sense that its bias and dispersion are the smallest.
Abstract: The purpose of this paper is to describe the small sample behaviour of some members of kclass estimators in a number of specific cases by the means of a Monte Carlo experiment. The k-class estimators are applied to the consumption function of Klein's Model I with data from the Greek economy. Specifically we consider some forms of normal and non-normal disturbance distributions and report on a Monte Carlo study of the small sample properties of Ordinary Least Squares (OLS), Two-Stage Least Squares, Nagar's Unbiased to order T-1 and Limited Maximum Likelihood (LIML) estimators. The results show that the OLS estimator of structural coefficients (predictor of the mean of consumption variable) is the best in the sense that its bias and dispersion are the smallest, and the LIML estimator (predictor) is the worst for almost every form of disturbance distribution examined. In addition, the alternative estimators (predictors) seem in general to have the lowest dispersion when the disturbances follow the lognormal distribution and the highest when the disturbances follow the normal distribution.