Design, data, and analysis
Bibliographic details
Additional authors
Daniel, Cuthbert. Mallows, Colin L.
Summary
This paper discusses issues and methods for selecting a subset of initially observed variables in three different statistical data analysis context. the first is multiple regression analysis,where the problem is to choose a subset of the independent variables that will do well in predicting the dependent variable in the regression relationship. the second situation is discriminant analysis, where one would like to choose a subset of initially observed variables which would maximally separate prespecified groups from which the observations are known to arise. the third context is that of clustering, wherein the grouping of the observations themselves emerge from the data analysis, and the problem is to select a subset of initially observed dependent variables that would account for cluster structure present in the data. the paper illustrates the uses of the methods for all three contexts in analyzing a real data set.