Theo K. Dijkstra – författare

This volume collects papers prepared for the workshop '~isspecification Ana- lysis" held in Groningen. The Netherlands. December 15th and 16th. 1983. The papers. which cover a wide range of problems. contain a number of interesting and fruitful ideas. A bird's-eye view of the contents of the papers is as follows: White describes in a very general setting properties of classical statistical procedures when models are misspecified. Verbeek provides a lucid analysis of the statistical complexity of the model selection process. Dijkstra indicates how to construct good estimators. taking the possibility of misspecification explicitly into account. Taking the view that measured variables are never continuous. De Leeuw dis- cusses in an illuminating way whether. and how. the discreteness of the data should and can be taken more seriously. Starting from the thesis that models are at best useful approximations of reality. Van ~ag and Koster derive properties of estimators determining best fitting hyperplanes. Within the context of the errors-in-variables problem.Bekker, Xapteyn and Wansbeek provide an elegant description of the variation in estimation re- sults due to part of the modelspecification varying, while estimators are .adjusted so as to keep them consistent. Bierens develops consistent tests of the hypothesis of parameter constancy against extremely diffuse alternatives.

In this book problems related to the choice of models in such diverse fields as regression, covariance structure, time series analysis and multinomial experiments are discussed. The emphasis is on the statistical implications for model assessment when the assessment is done with the same data that generated the model. This is a problem of long standing, notorious for its difficulty. Some contributors discuss this problem in an illuminating way. Others, and this is a truly novel feature, investigate systematically whether sample re-use methods like the bootstrap can be used to assess the quality of estimators or predictors in a reliable way given the initial model uncertainty. The book should prove to be valuable for advanced practitioners and statistical methodologists alike.

In this book problems related to the choice of models in such diverse fields as regression, covariance structure, time series analysis and multinomial experiments are discussed. The emphasis is on the statistical implications for model assessment when the assessment is done with the same data that generated the model. This is a problem of long standing, notorious for its difficulty. Some contributors discuss this problem in an illuminating way. Others, and this is a truly novel feature, investigate systematically whether sample re-use methods like the bootstrap can be used to assess the quality of estimators or predictors in a reliable way given the initial model uncertainty. The book should prove to be valuable for advanced practitioners and statistical methodologists alike.