Information about the people receiving OAA is available only from occasional studies in the Social Security Bulletin. Fisher (1946) described the characteristics of OAA recipients in twenty-one states in 1944.18 The median age of recipients was 74.1, compared to 70.3 for Social Security recipients. 53% of OAA recipients were women.19 OAA recipients, especially women, were less likely to be married than the rest of the elderly population. 9.2% of OAA recipients were nonwhite, though nonwhites accounted for 6.7% of the total elderly population.

OAA benefit levels were determined by subtracting an applicant’s income from what was called a monthly budget, or standard living requirement as a function of family size. This implies a 100% tax rate on benefits as earnings rose – a severe penalty against work. However, data on the maximum benefits were not regularly reported in the Social Security Bulletin. Program generosity can instead be inferred from a state’s average payout per recipient, reported monthly. Using this measure of average generosity will preclude decomposing estimates of the average treatment effect into income and substitution elasticities. A more serious problem might arise, though, because a state’s average benefit is potentially endogeneous. A state’s average benefit can be expressed as
where i indexes a state’s OAA recipients, N is the number of recipients, B™* is the state’s maximum benefit, YSi is the income of a recipient, and YS is the statewide average income of recipients. The problem is that differences in B across states will reflect not only statutory changes in Bmax, but also differences in recipients’ income, which is a function of individual behavior. This would not cause trouble if the regression of labor force participation on average benefits includes state and year fixed effects – unless income moved differently across states. For instance, low income people in one state might have become better off relative to another state -which raises Y and lowers B without a change in program rules. If changes in income distribution were also correlated with changes in labor force participation, then an inconsistent estimate of the program’s effect on participation would arise.