1. We should caution the reader that ONE and FIVE can only provide a partial view of the distribution of income within a state. Broader measures available after World War II (see the references in footnote 12 above) rank some western and southern states relatively high in inequality.
2. Delaware and North Dakota are omitted from our analysis. For Delaware, 29 percent of 1929 net income was in only two returns and the Commerce Department frequently mentions difficulties in estimating and using the state's income statistics. North Dakota has been excluded because of errors in the 1929 published tax data, pointed out in Kuznets, Shares, p. 249.
3. The desired denominator is gross income payments to residents by state rather than by location of the payer. The difference between the two concepts is probably greatest for wages and salaries. Estimates were based on establishment data, and wages of non-residents would therefore be incorrectly allocated. The Commerce Department adjusted totals for New York, New Jersey, Maryland, and Virginia to reduce this discrepancy. Proprietor and property (capital) payments are based on individual censuses (including the tax returns) and are affected to a lesser degree. A different conceptual problem is that the functional distribution of income by state is mismeasured when property income generated in a state is not listed in that state's production because it is received by a non-resident. For comments on these problems and the wage and salary adjustments see, Creamer and Merwin, ‘State Income Payments,’ pp. 19, 26.
4. Kuznets, Shares, Table III, pp. 516–18. The number of persons per return (PPR) can be calculated across states or income classes but not jointly across both. Variation in PPR by class is small except for lower income groups ($3,000 and under); state PPRs range from 1.8 to 2.3. State variations result from differences in the distribution of returns by income class, the number of dependents per joint return, and the ratio of joint to single returns. We assumed that the first factor is the crucial one and accordingly used the national PPRs by income class for all states rather than the state PPR for every income class in the state. We examined the potential bias from this assumption by comparing the estimated population covered by all returns using this method with the calculated return population for the states. For states with high PPRs (e.g., southern states) our method underestimated the population, consequently underestiating the number of returns needed to find one percent of the population as well as the income for the group. The opposite holds for the below average PPR states. The maximum potential error appears to be about 0.5 percent for ONE. Only a few states, however, are affected to this degree.
5. Kuznets, Shares, p. 35. For a number of states with higher income levels (e.g., New York, Illinois, and Massachusetts) it is possible to estimate shares further into the income distribution.