¹ Fry, Brian R. and Winters, Richard F., “The Politics of Redistribution,” American Political Science Review, 64 (June, 1970), 508–522 CrossRefGoogle Scholar.

² Fry and Winters, p. 508.

³ Fry and Winters, p. 508.

⁴ This literature starts with Dawson, Richard E. and Robinson, James A., “Inter-Party Competition, Economic Variables and Welfare Policies in the American States,” The Journal of Politics, 25 (May, 1963), 265–289 CrossRefGoogle Scholar. Fry and Winters provide a concise review of the literature. For a critique of the literature in this field see, Booms, Joan B. and Booms, Bernard H., “A Critique of Recent Public Policy Output Studies,” unpublished paper, The Pennsylvania State University, 1972 Google Scholar.

⁵ Fry and Winters, “The Politics of Redistribution,” p. 510. Others have also pointed to the interesting fact that public welfare is the one function for which outcomes are least well accounted. See, for example, Albin, Peter S. and Stein, Bruno, “The Determinants of Relief Policy at the Sub-Federal Level,” Southern Economic Journal, 37 (April, 1971), 446 CrossRefGoogle Scholar. Since public welfare is mainly a redistributive program, this finding suggests that redistributive elements of public policy are not well accounted for in previous studies.

⁶ Fry and Winters, p. 510.

⁷ See Musgrave, Richard A., The Theory of Public Finance (New York: McGraw-Hill, 1959)Google Scholar, for a discussion of the distinction between the allocative and distributive functions of government, and a definition of public outcomes based on the performance of the market.

⁸ Jacob, Herbert and Lipsky, Michael, “Outputs, Structure, and Power: An Assessment of Changes in the Study of State and Local Politics,” Journal of Politics (May, 1968), 515 Google Scholar, cited in Fry, and Winters, , “The Politics of Redistribution,” p. 510 Google Scholar.

⁹ John P. Crecine presents some evidence which tends to support these statements. His results seem to indicate that level of expenditures and revenues is, in the short run, relatively immune to political pressure. He concludes that political pressure and influence are more likely to affect whose street gets repaired, and how much is spent per pupil at a school, rather than the level of the total street repair or school district expenditures. See Crecine, John P., Government Problem Solving: A Computer Simulation of Municipal Budgeting (Chicago: Rand McNally, 1969)Google Scholar.

¹⁰ Fry, and Winters, , “The Politics of Redistribution,” p. 513 Google Scholar.

¹¹ Tax Foundation, Inc., Tax Burdens and Benefits of Government Expenditures By Income Class, 1961 and 1965 (New York: Tax Foundation, Inc., 1967)Google Scholar.

¹² Tax Foundation, p. 7.

¹³ U.S. Department of Labor, Bureau of Labor Statistics, Survey of Consumer Expenditures, 1960–1961, Report No. 237–93, February 1965 Google Scholar, and supplements.

¹⁴ For example, it was assumed that expenditures for automobile operations served as the basis of allocation for motor vehicle taxes. The average amount spent for automobile operations in each income class was multiplied by the number of families in that income class to determine the total amount spent by each income class for automobile operations. This amount was summed across income classes to determine the amount spent by all families in all income classes. The amount spent by each specific income class was then taken as a percentage of that total. If the families in the income class of $15,000 and over accounted for 10 per cent of all automobile operation expenditures, they were allocated 10 per cent of the burden for Motor Vehicle taxes. Similarly, on the expenditure side, it was assumed that benefits of government highway expenditures were allocated to each income class according to one-half its expenditures for automobile operation and one-half its expenditures for total current consumption. The same process was repeated to determine the percentage of the total amount spent for automobile operation and total current consumption by each income class. If the income class of $15,000 and over accounted for 12 per cent of expenditures for total current consumption and 8 per cent of expenditures for automobile operation, they would then be allocated 10 per cent of the benefits for government highway expenditures.

¹⁵ Fry and Winters, pp. 514–515.

¹⁶ An examination of 1960 Census data proves that the distribution of families by income class does vary significantly from state to state. As an illustration, the national percentage of families and unattached individuals in the income class of $3,000 to $3,999 is 11.0 per cent as reported by the BLS. According to 1960 Census data, the percentage of families who earn from $3,000 to $3,999 varies from 6.2 per cent in Connecticut to 14.3 per cent in Maine.

¹⁷ The error of assuming a uniform income distribution among states made it possible to use only two tables, one for burdens, the other for benefits. Our technique yields two such summary tables for each state so it is impossible to include these summary tables.

¹⁸ See comment by Noell, James J. and reply by Fry, and Winters, . American Political Science Review, 64 (12, 1970), 1249–1251. An unknown referee pointed out a problem that neither Noell nor we mention, viz., that caution should be used in choosing which income variables to include as independent variables, since the dependent variable is in part income-defined. Obviously it would be meaningless to regress a variable on its own constituent parts.CrossRefGoogle Scholar

¹⁹ For a description of this test, see Farrar, Donald E. and Glauber, Robert R., “Multicollinearity in Regression Analysis: The Problem Revisited,” Review of Economics and Statistics, 49 (February, 1967), 92–107 CrossRefGoogle Scholar.

²⁰ See Goldberger, Arthur S., Econometric Theory, John Wiley, New York, 1964, pp. 197–198 Google Scholar. A note of caution is in order here. The fact that beta coefficients are themselves statistics is often overlooked by researchers. Before any meaning can be attributed to a beta coefficient, one must test whether it is significantly different from zero. Likewise, in comparing the magnitudes of beta coefficients, one should test for a statistical significance difference between the two beta coefficients. See Hu, Teh-wei, “Beta Coefficients Used in Statistical Inference,” unpublished paper (The Pennsylvania State University, 1971)Google Scholar.

²¹ Note also that indications of statistical significances are given for the regression coefficients, making possible an evaluation of the reliability of these estimates.

²² Th e values of (C_h ) for the category of elementary and secondary education were not included in this formula. The basis of allocation for this category is the number of children under 18 years of age for each family. The values of (P_ih ) for this category were taken directly from the BLS study.

²³ U.S. Bureau of the Census, U.S. Census of the Population: 1960, Vol. 1, Characteristics of the Population (Washington, D.C.: U.S. Government Printing Office, 1963)Google Scholar, pts. 2–9, 11–52. A problem arises here since the income classes used by the BLS and those used by the Bureau of the Census do not coincide in all cases. To solve this problem, those income classes between $6,000 and $9,999 were lumped into one class. To combine the BLS data, weighted averages were taken (according to the number of families and single consumers in the income class to be combined) to determine the average amounts of income and expenditures for the income class of $6,000 to $9,999. For the Census data the classes were simply added together.

²⁴ U.S. Bureau of the Census (Washington, D.C.: U.S. Government Printing Office, 1962).