Abstract: Three classes of kernel density estimates are proposed which are appropriate in the analysis of complex survey data. The three classes of estimates pertain to use of the whole data file, to use of binned data and to smoothing binned data. In each class a model-based asymptotic integrated mean square error is obtained under the complex sampling design. The parallel design-based asymptotic integrated mean square errors are obtained for the binned data and the smoothed binned data only. Quantile estimates from the smoothed binned data are proposed. The methodology is applied to data from the Ontario Health Survey of 1990.
Key words and phrases: Histograms, integrated mean square error, kernel density estimation, quantile estimation, smoothing.