The level of measurement of the dependent variable (nominal, ordinal, interval) crucially affects the selection of statistical techniques. A comparison is made of three powerful statistical techniques appropriate for each of the levels of measurement, & sharing a similar set of assumptions. These techniques are: (1) ordinary least squares regression for intervally measured, (2) probit for ordinally measured, & (3) discriminant analysis for nominally measured dependent variables. The assumptions & uses of each technique are reviewed, & an example of the use of each in political research is presented. 10 Figures, 4 Tables, Appendix. Modified HA.
Recent comparative, quantitative studies of state politics conclude that party competition and other political variables have little or no impact on important state policies such as per pupil expenditure, old age assistance, unemployment compensation, and aid to dependent children. These are rather unexpected and disturbing conclusions for they disconfirm relationships predicted by some of the most important theoretical formulations concerning democratic politics. Before re-examining the bases for these conclusions, a review of theory is in order.V. O. Key, Jr., set the context for examining the effects of political variables on state policies. He stressed the importance of two-party competition, or bi-factionalism in one-party states, as a determinant of policy. Key sees the degree of party competition as crucial because it reflects the extent to which politics is organized or unorganized. Party competition by producing some semblance of an organized politics lessens the difficulty of lower status groups in sorting out political actors and issues, thereby enabling them to promote their own interests more effectively. Since state social welfare policies are undoubtedly relevant to the interests of "have-nots," we can utilize state expenditures in this area as a measure of the success that these groups have enjoyed. Key's formulation, then, would lead to a simple two variable model:P = e1S = k1P + ezWhere P is inter-party competition, S is expenditure on social welfare, and e1 and ez represent error or variables left out of the system, and k1 is a constant.
The construction of an empirical theory of democratic political development is dependent on the formulation of causal propositions which are generalizations of the developmental process. To date, several essential steps in the process of constructing such a theory have been taken. First, concept formation and clarification by students of political development has led to an emphasis upon political democracy as one of the dependent variables for the field. Second, the gathering and publication of quantitative indicators of social, economic, cultural, and political phenomena provide a firm basis for subsequent empirical inquiry. Finally, correlational analysis has identified numerous variables which are closely associated with the development of democratic political institutions.The next major task is the formulation and testing of empirical models of democratic political development which provide a basis for inferring causal relationships by distinguishing between spurious correlations and indirect and direct effects. The accomplishment of this task would enable us to derive explanatory propositions concerning the process of democratic political development.
Warren E. Miller and Donald E. Stokes' publication in 1963 of a preliminary report on the Survey Research Center's representation study is an important landmark in the development of empirical political theory. That report addressed itself to the crucial theoretical question of the linkage between mass political opinions and governmental policy-making. More specifically, the report found considerable policy agreement between Congressional roll call votes and the attitudes of the individual Congressman's constituency. This policy agreement was then interpreted through several causal paths and the Congressman's perception of his constituency's attitudes was found to be the main path by which the local district ultimately influenced Congressional outputs.The main body of the report dealt with the broad civil rights issue dimension, and, by specifying the perceptual path by which constituency influence is brought to bear, documented the effect of political issues despite the generally low level of political information held at the mass level. Thus, the Congressmen, through their broad cognitive evaluations, were aware of how far they could proceed in determining their civil rights roll call votes on the basis of their own attitudes before risking the displeasure of their constituents.Beyond such major substantive contributions the representation study introduced to political science a variance-apportioning technique similar to that developed by Sewall Wright, in 1921. Through this variance-apportioning technique, the importance of the perceptual link was isolated and evaluated. This study, then, symbolizes the growing recognition in political science of the importance of more sophisticated methodological tools in the process of theory building.