Analysis of longitudinal earning data indicates higher earnings growth for scientists of the same experience but more recent vintage. Theoretical justification for such a relation is suggested, and the implied biases in cross-section data are noted. Because of a basic identification problem, an alternative interpretation, time-experience interaction, is also considered. The general conclusion is that earning growth is not uniform or neutral. There is no simple mechanical method by which lifetime profiles can be inferred from single cross-section data.

IN THE ANALYSIS of economic data the investigator is frequently provided with only the simple sum of some elementary variables. The elementary variables are known to be perfect substitutes (i.e., to have the same effect) in some contexts, but not in all the relevant ones. Familiar examples are hours of work at possibly different jobs, hours of leisure spent at different activities, number of trips abroad, cigarette consumption (of possibly different brands). The question, naturally, is under what conditions can we expect that such aggregates will satisfy the same restrictions as those imposed by the theory on the elementary variables? As an answer we suggest the following variation of Hicks' [3, p. 33] composite good theorem: A simple sum of elementary goods can be treated as a single good if their (properly normalized) prices vary so as to keep all within group differences constant.2 We shall illustrate the proposition in the context of standard demand theory. Define the expenditure function: m (u, p) minm 1 l7pixi, subject to u(x1.. .xn) = u. Suppose that

Analysis of longitudinal earning data indicates higher earnings growth for scientists of the same experience but more recent vintage. Theoretical justification for such a relation is suggested, and the implied biases in cross-section data are noted. Because of a basic identification problem, an alternative interpretation, time-experience interaction, is also considered. The general conclusion is that earning growth is not uniform or neutral. There is no simple mechanical method by which lifetime profiles can be inferred from single cross-section data.