Selection using a dichotomized auxiliary variate
Date
Authors
Major Professor
Advisor
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Altmetrics
Authors
Research Projects
Organizational Units
Journal Issue
Is Version Of
Versions
Series
Department
Abstract
Consider a random sample of n objects from which we wish to select the k best. Quality will be defined in terms of measurements Y(,i) (i = 1, ..., n); the larger the value of Y(,i), the more valuable the object. The obvious solution--measure Y(,1), ..., Y(,n), and select those objects with the k largest Y(,i) values--may be impractical, if the Y(,i)'s are difficult to observe. An alternative is to select objects on the basis of an auxiliary variate X(,i) (i = 1, ..., n), related to Y(,i), but less elusive;Under the assumption that X(,i) and Y(,i) are positively associated, a reasonable procedure is to choose those objects with X(,i) values exceeding some fixed point x(,c). Denote the probability that the sub- set of items thus selected does indeed include the k best by (,n)(PI)(,k). Then, x(,c) may be chosen so that (,n)(PI)(,k) (GREATERTHEQ) P*, where P* (0 ) (INFIN), is examined when k is either constant or a function of n (k = k(n) = f(.)n or n('f) (0 ) 1; when k is proportional to n, (,n)(PI)(,k) (--->) 0; and when k = n('f), the limit is distribution dependent (see text for details);Yeo and David (1984) consider a fixed-size subset selection procedure--choosing those objects with the s largest X(,i) values--and calculate (,n)(PI)(,s:k), the probability that this subset includes the k best;objects. Our approach is contrasted with Yeo and David's, and an alternative derivation of (,n)(PI)(,s:k) is provided; Reference Yeo, W. B., and David, H. A. 1984. Selection through an associated; characteristic, with applications to the random effects model. JASA 79, 399-405.