Robert Moffitt argues that trends in U.S. public assistance expenditures from the late 1960s to the mid-1990s have been driven by “demographic” factors, particularly the rise in the fraction of households headed by single women. The evidence for this conclusion derives from a simple accounting identity that expresses mean welfare benefits per capita for a demographic group as the product of an annual welfare participation rate and a conditional mean level of benefits among participants. For the country as a whole, mean welfare benefits per capita are a weighted average of per capita benefits among each group, leading to Moffitt's equation (1). Over time, one can therefore decompose the change in per capita benefits into a component due to shifting weights, another due to shifting participation rates, and a third due to changes in the mean levels of benefits received by participants. As is well known from the wage discrimination literature (e.g., Oaxaca, 1974), there is no unique way to perform such a decomposition. Moffitt presents six variants, which give similar answers with respect to the trend in AFDC/TANF spending per capita (Table 9.4). Had benefit levels per participant remained constant, the changing shares of different household types (married, female nevermarried head, female divorced/separated head, male head) would have led to a substantially larger increase in benefits than actually occurred. This rise was offset by reductions in average benefits per participant, presumably driven by the secular decline in AFDC/TANF benefit rates.