⁵The methods for construcing this figure draw on a literature dating at least to Browning, Deaton, and Irish 1985 and with recent contributions by Attanasio and Weber 1995. These authors have shown how to construct ‘synthetic panels’ from a series of cross-section surveys like the four SCFs used in this paper. That literature has noted that age, time, and cohort effects cannot be independently distinguished using such data, because age, time, and cohort are linearly related. The assumptions I made to identify age effects were, first, that cohort effects can be captured by a single term reflecting the lifetime level of permanent income of each cohort, (which I assume increased on average by 1.5 percent per annum for the cohorts in question, if anything an underestimate of the relevant average productivity growth rate and therefore a source of downward bias in the slope of the estimated age profile); and, second, that the time effects averaged to zero over the four SCF surveys.