¹ Dawson, Richard E. and Robinson, James A., “Interparty Competition, Economic Variables, and Welfare Policies in the American States,” Journal of Politics, 25 (05, 1963), 265–289 CrossRefGoogle Scholar; Dye, Thomas R.; Politics, Economics, and the Public: Policy Outcomes in the American States (Chicago: Rand McNally, 1966)Google Scholar; Hofferbert, Richard I., “The Relation Between Public Policy and Some Structural and Environmental Variables in the American States,” this Review, 60 (03, 1966), 73–82 Google Scholar. One of the present authors has employed a different conception of “political” characteristics and his results qualify those of the authors listed above. See Sharkansky, Ira, Spending in the American States (Chicago: Rand MacNally, 1968)Google Scholar.

² See Sharkansky, op. cit., plus Wilson, James A., City Politics and Public Policy (New York: Wiley, 1968), Chapter 1Google Scholar; and Lockard, Duane, “State Party Systems and Policy Outputs,” in Garceau, Oliver (ed.), Political Research and Political Theory (Cambridge: Harvard University Press, 1968), pp. 190–220 Google Scholar.

³ For a lucid critique of single indicator analysis and a discussion of the methodological justification for multiple indicators see Webb, Eugene J., Campbell, Donald T., Schwartz, Richard D., and Sechrest, Lee, Unobtntsive Measures: Nonreactive Research in the Social Sciences (Chicago: Rand McNally, 1966), pp. 3ffGoogle Scholar.

⁴ For an example of such an index, see Hofferbert, “The Relation …,” op. cit.

⁵ The technique of factor analysis starts from the basic assumption that interrelations among separate variables signal the existance of underlying traits–or “factors”–that they share in common. A factor analysis manipulates a collection of variables in order to discern the various patterns of relationships among them. The groups of variables that relate closely to one another, but only loosely (or not at all) to variables in other groups are extracted as the principal factors. The individual variables that show the strongest relationships with other members of their factor have, in the language of factor analysis, the highest loadings. A variable's loading is, in effect, the coefficient of correlation between that variable and the underlying factor. By knowing the variables with the highest loadings on each factor, it is possible to infer something about the underlying traits that the factor represents. The variables with the highest loadings come closest to representing the underlying trait, although it is unlikely that any single variable represents that trait perfectly. The factoring technique used with the political and service variables employs orthogonal rotation. The device examines the correlation matrix of the individual variables and extracts factors whose members show maximum correlations among themselves and minimum correlations with the members of other factors. The program used is “Image,” by Walter Stoly, Mass Communications Center, University of Wisconsin, Madison, version of July 2, 1968. For an explanation of techniques used in the socioeconomic factor, see Hofferbert, “Socioeconomic Dimensions …,” op. cit. Although Hofferbert employed oblique rotation in the extraction of his factors, the comparability to our political and service factors is not appreciably affected. The intercorrelation of the two socio-economic factors for 1960 is .023.

⁶ Armstrong, J. Scott, “Derivation of Theory of Means of Factor Analysis or Tom Swift and His Electric Factor Analysis Machine,” American Statistician (12, 1967), 17–21 Google Scholar.

⁷ See, for example, Mueller, John E., “Some Comments on Russett's ‘Discovering Voting Groups in the United Nations,’” this Review, 61 (03, 1967), 146–148 Google Scholar.

⁸ But we have been sensitive to the possibility of biased results. Certain initial steps have been taken to reduce this bias. Factor analysis cannot be relied upon to define the dimensions of a complex phenomenon (e.g., the state political system), particularly if the collection of variables subject to factor analysis is pre-loaded with many measures of the same phenomenon, each of which is highly correlated with others. Thus our first task was to insure that we did not “stack” our deck of variables with highly intercorrelated measures (r > .70) of the same phenomenon. This problem was particularly acute in the case of measures of party competition. Similar screening of the policy variables we examined did not reveal comparable redundancies.

⁹ To maintain prima facie relevance, we use factor analysis in “stages” first to select from among a large collection those variables that load highly on principal factors and then to define the loadings of these variables on factors that contain only those highly-loaded components. In this way, the factor technique produces relatively “pure” factors, devoid of large numbers of variables that contribute only weakly to the principal factors. In both the political and policy analyses, a large number of variables load low on two principal factors, and several variables load about equally on each principal factor. In order to simplify these factors for the purposes of clarity and further analysis, we eliminated all variables loading below .5 on both principal factors and those loading at least .5 on one but above .4 on the other factor in each sector. (The variables surviving this elimination are noted as such on Tables 1 and 3.) We then made separate factor analyses of the variables that remained after this culling procedure. (The results in Tables 2 and 4.) The values derived from this final set of analyses constitute the bases for constructing indices of the political and policy sectors of our model.

¹⁰ Hofferbert, , “Socioeconomic Dimensions of the American States: 1890–1960,” Midwest Journal of Political Science, 12 (08, 1968), 401–418 CrossRefGoogle Scholar. The factor scoring program used for the political and policy factors if “FACTSCR2” by Keith R. Billingsley, Department of Political Science, University of Wisconsin, Madison, version of July 20, 1968. The scoring technique is described in Kaiser, Henry F., “Formulas for Component Scores,” Psychometrika, 27 (03, 1962), 83–87 CrossRefGoogle Scholar.

¹¹ Dye, op. cit., and Grumm, John G., “Structure and Policy in the Legislature,” a paper presented at the Southwestern Social Science Association annual meeting, Dallas, 03, 1967 Google Scholar.

¹² Schlesinger, Joseph A., “The Politics of the Executive,” in Jacob, Herbert and Vines, Kenneth N. (eds.), Politics in the American States (Boston: Little, Brown, 1965), pp. 207–238 Google Scholar.

¹³ Froman, Lewis A. Jr., “Some Effects of Interest Group Strength in State Politics,” this Review, 60 (12, 1966), 952–962 Google Scholar.

¹⁴ See Bishop, George A., “The Tax Burden by Income Class, 1958,” National Tax Journal, 14 (03, 1961), 41–58 CrossRefGoogle Scholar.

¹⁵ In interpreting the results of the factor analysis, a loading of .700 will be considered to be high.

¹⁶ Key, V. O., Southern Politics (New York: Alfred A. Knopf, 1949)Google Scholar; Key, , American State Politics: An Introduction (New York: Alfred A. Knopf, 1956)Google Scholar; Lockard, Duane, New England State Polilics (Princeton: Princeton University Press, 1959)CrossRefGoogle Scholar; Fenton, John H., Politics in the Border States (New Orleans: Hauser Press, 1957)Google Scholar; and Fenton, , Midwest Politics (New York: Holt, Rinehart and Winston, 1966)Google Scholar.

¹⁷ We have described above our justifications for including measures pertaining to government personnel, structure, and revenue among our political variables. In brief, they represent features of revenue inputs and governmental structures that may affect response to citizens' needs and demands. Our policy measures are more narrow in representing the benefits which governments actually provide to their citizens within the prominent categories of service.

¹⁸ Although some research has relied almost exclusively on expenditures as indicators of public policy, a recent study shows the risk in assuming an expenditure-service relationship. The current expenditures of state and local governments may not be uniformly reliable indicators of the service that is actually provided to the population. But on the other hand spending may represent the effort that state and local governments are making in order to improve or maintain their services. Therefore it deserves a place on our list of policy variables. See Sharkansky, , “Government Expenditures and Public Services in the American States,” this Review, 61 (12, 1967), 1066–1077 Google Scholar.

¹⁹ See Burch, Philip H. Jr., Highway Revenue and Expenditure Policy in the United States (New Brunswick: Rutgers University Press, 1962), p. 23 Google Scholar.

²⁰ In collecting the data for the measures of outputs and services, the assumption was made that recorded information is a reasonably accurate reflection of fact. The authors recognize considerable controversy about the reliability of the data. Yet the data chosen appear to be the best available, and they enjoy wide use among social scientists. For comments about each item the reader is referred to the sources. The sources for the political variables include: Census of Governments, 1962: The Book of the States 1964–1965; Statistical Abstract of the United States, 1964; the index of gubernatorial power comes from Schlesinger's article cited in note #12 ; the index of suffrage regulations comes from Lester Milbrath, “Political Participation in the States,” in Jacob and Vines, op. cit., pp. 25–60; the indices of legislative apportionment comes from David, Paul T. and Eisenberg, Ralph, Devaluation of the Urban and Suburban Vote (Charlottesville: Bureau of Public Administration, University of Virginia, 1961), p. 5, 15 Google Scholar, and Shubert, Glendon and Press, Charles, “Measuring Malapportionment,” this Review, 58 (1964), 969 Google Scholar. The sources for policy variables include: National Education Association, Rankings of the States, 1963; U.S. Bureau of Public Road, Annual Report 1963; Social Security Administration, Social Security Bulletin: Annual Statistical Supplement 1963; The Book of the States, 1964–1965; Statistical Abstract of the United States, 1964.

²¹ For a discussion of the politics of old age assistance in Alabama, see Sharkansky, , Spending in Ike American Stales, pp. 138–141 Google Scholar.

²² Hofferbert, “Socioeconomic Dimensions …,” op. cit.

²³ Ibid. The label given to one of the socioeconomic factors is new with this article. In the original article “Affluence” was labelled “Cultural Enrichment,” the present term seems more appropriate as a description of the contents of this factor.

²⁴ See, for example, Dye, op. cit., p. 125 and p. 162.

²⁵ See Sharkansky, Spending in the American States., Chapters 4 and 7.

²⁶ The strength of relationships between the political and economic factors is stronger than that customarily obtained with individual variables. The correlations that we have described are .73 and .66. Only two out of the forty-eight simple correlations between economic and political variables that Dye reports in his study reach the level of .66 and none reach .73 (see his Chapter 3). The underlying dimensions that are tapped by factor analysis appear to be more salient representations of the economic-political nexus than are apparent in studies using single variables.

²⁷ The political and economic factors are more successful in accounting for interstate variance in the policy factors than are most individual measures of these phenomena. Table 8 shows simple, partial, and multiple correlation coefficients between policy factors as dependent variables and the political and economic factors as independent variables. It shows that both political factors together account for 61 and 36 percent of the variance in the policy factors. In contrast, the multiple correlation coefficients that Dye reports show that his collection of individual political variables account for 36 percent of the variance in only fifteeen out of fifty-four dependent variables, and they account for 61 percent of the variance in only two out of the fifty-four instances. (See Dye, op. cit., pp. 286–287.) Both of our socioeconomic factors account for 59 and 68 percent of the variance in the policy factors. In contrast, Dye's collection of individual social and economic variables account for 59 percent of the variance in only fourteen of fifty-four dependent variables, and they account for 68 percent of the variance in only eight of fifty-four dependent variables. A similar finding appears when we compare with his data the success of our economic and political factors together in explaining variance in our policy factors. We explain 69 and 70 percent of the variance in the policy factors by means of all political and socioeconomic variables: variables Dye explains 69 percent of the variance for only thirteen of fifty-four dependent variables,

²⁸ See their works cited in note #16, supra.

²⁹ Cnudde, Charles F. and McCrone, Donald J., “Party Competition and Welfare Policies in the American States, this Review, 63 (09, 1969)Google Scholar.

³⁰ One limitation of the data employed throughout this article is their lack of historical perspective. A separate study of the two socioeconomic factors and their relationship with individual policy and electoral variables has shown that the strength of relationship is, in some cases, quite fluid over time. (See Hofferbert, “Socioeconomic Dimensions …,” op. cit.) This fluidity, however, may be a result of simple changes in the value of isolated dependent variables and not an accurate picture of the more analytically interesting dimensions of policy and political life which we have been considering here. Hofferbert found that the relative contribution of individual variables maintained a high degree of consistency in their loadings on the principal factors from one decade to the next, although individual variable-by-variable. correlations shifted considerably. If it were possible to analyze the full component of political and service variables over extended time, the same type of configuration could probably be expected. Consequently, the strength of relationship of factors could be fairly constant over time even though the individual variable correlations changed considerably. Because of limited availability of the political and policy data for previous years, however, this must remain a matter for speculation.

For a useful discussion of the methodological problems of theory construction employing measures standardized for a specific population and time, such as the correlation coefficients used here, versus unstandardized regression coefficients see Blalock, Hubert “Causal Inferences, Closed Populations, and Measures of Association,” this Review, 61 (03, 1967), 130–136 Google Scholar. Blalock argues that in building theoretical statements of more general application, the regression is the most use since it is not as sensitive as the correlation coefficient to specific population values at a given time. However, in an investigation such as ours, where possibility of longitudinal analysis is limited due to the availability of relevant data, it seemed advisable to retain the greater precision of standardized measures. Furthermore, in the case of the states, the relative magnitudes of most of the variables is not particularly fluid as might be the case with some other types of indicators, such as the congressional roll call votes discovered by Blalock.

³¹ Although the figures in Table 2 report that the political factors account for 66 percent of the variance, this pertains to the variance in the eleven variables that were subject to this factor analysis. These factors do not account for this much of the variance in the total collection of fifty-three variables that are presented in Table 1.