Consider the problem of estimating a function T = T(z,y) of values of the variables z and y in a finite population, when values of y are known a priori for all units but values of z are not known for any. Data are then obtained from a sample of units, not in the form of values of z, however, but in the form of secondary variables x whose values depend stochastically on z. A framework for Bayesian analysis to T along the lines of Ericson (1969) is presented and illustrated with data from the Profile of American Youth (U.S. Department of Defense, 1982). (50pp.)
