This paper is concerned with the problem of combining a data set that identifies the conditional distribution P (y|x) with one that identifies the conditional distribution P (z|x), in order to identify the regressions E (y|x, middot) identical with [E (y|x, z = j), j element of Z] when the...