Chi-Squared Tests with Complex Survey Data Subject to Misclassification Error推荐.pdfVIP

CHAPTER 31 CHI-SQUARED TESTS WITH COMPLEX SURVEY DATA SUBJECT TO MISCLASSIFICATION ERROR J . N. K. Rao and D. Roland Thomas Carleton University 31.1 INTRODUCTION Standard methods for analyzing cross-classified categorical data are based on some form of chi-squared test procedure under loglinear models and simple random sampling with replacement . Unfortunately, many of the surveys used in practice have complex designs that involve stratifica- tion, unequal probability sampling, clustering and several stages of unit selection, and so yield data that do not satisfy the distributional assumptions underlying the standard analyses . Furthermore , it is often the case that the data items themselves are subject to classification errors that will bias the estimated cell probabilities and distort the properties of the hypothesis test s in various ways . There is now an extensive literature on methods for analyzing categorical data from complex surveys in the absence of misclassification, and on methods for dealing with classification errors under simple random sampling with replace- ment. The aim of thi s chapter is to unify the two approaches and provide methods that can deal with the practical reality of complex categorical survey data subject to misclassification. Both streams of research will be briefly reviewed, and then the strategy for a unified analysis will be described. Three different analysis strategies have been proposed for incorpor- ating the effect of the survey design in the analysis of categorical data in The author s both wish t o acknowledge research grant support from the Natural Science s and Engineering Research Council of Canada. Measurement Errors in Surveys. Edited by Bieme