No label defined (Q4521571)

From the Preface: One notes that ``this text presents the view that survey sampling properly proceeds from the same understanding as does statistical thought generally, namely that the data can be understood as realizations of random variables, and that models constructed to mirror the random process can be usefully employed for inference.'' In Chapter 1 (24 p.), Prediction Theory is introduced where the \(y\)-values \(y_1,y_2,\dots,y_N\) of a study variate on the units \(U_1,U_2, \dots, U_N\) of a finite population are treated as realized values of random variables \(Y_1,Y_2,\dots,Y_N\) and estimation of a function \(h(y_1,y_2, \dots, y_N)\) entails prediction of the unobserved \(Y\)'s after the sample has been observed using relationships among the random variables. These relationships are expressed in a model for their joint distribution using which predictions are made. The theory is formulated in Chapter 2 (24 p.) under a General Linear Model. The third Chapter deals with the question of bias-robust estimation under model failure (46 p.) by introducing balanced samples. The type of balance needed to reduce the error variance as much as possible under certain conditions is further discussed in the following Chapter (30 p.). Chapter 5 (42 p.) considers robust variance estimation in the earlier sense of model failure. The next Chapter (44 p.) extends the results hitherto derived to stratified sampling with emphasis on both statistical and administrative functions. In particular, special practical problems of allocation of sample size to strata and demarcation of strata are considered. Data on qualitative variables is extensively collected in many surveys in practice. In such situations, general regression theory may not be applicable in view of ill-conditioned matrices, unbalanced data, etc.. Chapter 7 deals with models where information on qualitative auxiliary variables is collected (35 p.). A standard intracluster correlation model of a clustered population is considered in the next Chapter (50 p.), starting with a common mean model and extending to a more general model. As in chapter 5, robust variance estimation is studied in Chapter 9 (27 p.) for two-stage cluster sampling designs as well. Estimation of variance of estimators of nonlinear functions is the theme of Chapter 10 (28 p.). The last chapter deals with a number of special topics and open problems such as estimation in the presence of outliers, nonlinear models, nonparametric estimation, estimation of distribution functions and quantiles and small area estimation (62 p.). There are three appendices describing basic tools, data sets and S-PLUS functions. A bibliography consisting of over 260 references is at the end. An author index and subject index are available. At the end of each chapter there are a number of exercises complementing the theory and a few of these have been solved in the text. The book is a welcome addition to the subject of survey sampling. Collecting the material published during the last three decades in various journals and putting it together in the form of a text in itself is a beginning which might fulfil the hope of the authors ``that the ideas embodied in this text will begin to assume their proper role, directly, in the sampling world''.