摘要:

Some foundational issues in statistical inference, having a bearing on ‘influence’ are discussed.

ISSN:0266-4763    

DOI:10.1080/02664768700000015

出版商:Carfax Publishing Company    

年代:1987

数据来源: Taylor

摘要:

Statistical models are useful devices for understanding the essential features of a set of data. Models, however, are nearly always approximate descriptions of more complicated processes. Because of this inexactness, the study of the variation in the results of an analysis under modest modifications of a problem formulation becomes important. Components of a problem that are critical for the interpretation of the results are called influential. This paper contains a review of selected topics in influence assessment with emphasis on recent developments. Topics covered include a discussion of general ideas behind influence diagnostics, a comparison of case deletion diagnostics in linear regression, and the unifying approach to influence developed by Cook (1986).

ISSN:0266-4763    

DOI:10.1080/02664768700000016

出版商:Carfax Publishing Company    

年代:1987

数据来源: Taylor

摘要:

A Bayesian approach is presented for influential observations and predictive discordancy diagnostics. Tests of significance based on predictive distributions are devised to test the discordancy of an observation. Examples illustrating the methods are provided.

ISSN:0266-4763    

DOI:10.1080/02664768700000017

出版商:Carfax Publishing Company    

年代:1987

数据来源: Taylor

摘要:

If surveys are analysed using the randomisation distribution then model-based influence diagnostics have no role. Individual values can be influential in randomisation inference either when they are included in the sample or when they are not in the sample. In the former case diagnostics are useful, in the latter only prior knowledge can save the inference. Post-stratification unifies a variety of ad hoc procedures. Some ideas deducible from the influence function provide a justification for the Hajek ratio estimator which is an alternative to the Horvitz-Thompson estimator.

ISSN:0266-4763    

DOI:10.1080/02664768700000018

出版商:Carfax Publishing Company    

年代:1987

数据来源: Taylor

摘要:

Surveys of human populations invariably yield data sets in a near-standard cases-by-variables format, where the variables divide into a set of biographical questions and a set of questions relating to the survey's direct objective. Analysis consists of investigating the interrelationships between these two sets of variables as well as exploring variation within each set. An approach is described using correspondence analysis and cluster analysis to reveal the patterns of covariation in a large data set. Correspondence analysis elicits new continuous constructs from the data, whereas cluster analysis elicits discrete constructs, and the two techniques often complement each other. The descriptions which they provide can influence one's understanding of the data by revealing the true complexity of features that are not always apparent in a formal modelling approach. Examples are presented in the context of readership surveys.

ISSN:0266-4763    

DOI:10.1080/02664768700000019

出版商:Carfax Publishing Company    

年代:1987

数据来源: Taylor

ISSN:0266-4763    

DOI:10.1080/02664768700000020

出版商:Carfax Publishing Company    

年代:1987

数据来源: Taylor

摘要:

We propose a method for detecting outlier components in two-way contingency tables, using the hypothesis of homogeneity, based on the adjusted residuals.

ISSN:0266-4763    

DOI:10.1080/02664768700000021

出版商:Carfax Publishing Company    

年代:1987

数据来源: Taylor

摘要:

This paper reports the results of an empirical study of applying seven commonly used methods for identifying outliers and influential observations in linear regression models to six well-known data sets in the literature. The results show that the methods do not always agree. Observations which are detected as ‘significantly influential’ may only moderately affect the parameter estimates when they are removed from the regression. Combinations of procedures should be judiciously used in empirical studies.

ISSN:0266-4763    

DOI:10.1080/02664768700000022

出版商:Carfax Publishing Company    

年代:1987

数据来源: Taylor

ISSN:0266-4763    

DOI:10.1080/02664768700000023

出版商:Carfax Publishing Company    

年代:1987

数据来源: Taylor