¹ See Simon, Herbert A., “Causal Ordering and Identifiability,” and “Spurious Correlation: A Causal Interpretation,” reprinted in Models of Man (New York: Wiley, 1957)Google Scholar, chs. 1–2; and Lazarsfeld, Paul F., “Evidence and Inference in Social Research,” Daedalus, 87 (Fall, 1958), 99–130Google Scholar. Blalock's work is reported in Blalock, Hubert M. Jr., Causal Inferences in Nonexperimental Research (Chapel Hill: University of North Carolina Press, 1964)Google Scholar.

² Political science applications of causal modelling approaches include Miller, Warren E. and Stokes, Donald E., “Constituency Influence in Congress,” this Review, 57 (03, 1963), 45–56Google Scholar; Alker, Hayward R. Jr., “Causal Inferences and Political Analysis,” in Bernd, Joseph (ed.), Mathematical Applications in Political Science (Dallas: Southern Methodist University Press, 1966)Google Scholar; and Goldberg, Arthur S., “Discerning a Causal Pattern Among Data on Voting Behavior,” this Review, 60 (12, 1966), 913–922Google Scholar. Causal modelling ideas have also been used to clarify the study of power; for example, Herbert A. Simon, “Notes on the Observation and Measurement of Political Power,” op. cit., ch. 4; and Dahl, Robert A., “Power,” International Encyclopedia of the Social Sciences (Macmillan, 1968), vol. 12, pp. 405–415Google Scholar.

³ Cnudde, Charles F. and McCrone, Donald J., “The Linkage Between Constituency Attitudes and Congressional Voting Behavior: A Causal Model,” this Review, 60 (03, 1966), 66–72Google Scholar. More recently the same authors have published an analysis of the causes of democratic political development that suffers from the same faults as their earlier paper. See McCrone, Donald J. and Cnudde, Charles F., “Toward a Communications Theory of Democratic Political Development: A Causal Model,” this Review, 61 (03, 1967), 72–79Google Scholar.

⁴ Matthews, Donald R. and Prothro, James W., Negroes and the New Southern Politics (New York: Harcourt, Brace and World, 1966)Google Scholar, ch. 11.

⁵ These models are called “hierarchical” because they assume that all the variables in the analysis can be ordered a priori in a hierarchy of causes and effects. Variables in a model are said to form a causal hierarchy if they can be ranked so that those “higher” in the ranking appear in the equations of the model only as causes, and never as effects, of those variables which are “lower” in the ranking.

⁶ Moreover it is not true, in general, that we can “infer the most likely [model]” if we “resort to the use of regression coefficients”: Cnudde and Mc-Crone, op. cit., p. 68. A given set of data may be used to estimate the parameters of many different models. The basic logic of model building and testing is especially clearly set out in Stefan Valavanis, , Econometrics: An Introduction to Maximum Likelihood Methods (New York: McGraw-Hill, 1959)Google Scholar, ch. 1.

⁷ Ibid., pp. 71–72. Compare the second proposition with Lewis Anthony Dexter's description of the representative: “A congressman's conception of his district confirms itself, to a considerable extent, and may constitute a sort of self-fulfilling prophecy …. A congressman hears most often from those who agree with him …. Some men automatically interpret what they hear to support their own viewpoints.” See “The Representative and His District,” in Peabody, Robert L. and Polsby, Nelson W. (eds.), New Perspectives on the House of Representatives (Chicago: Rand McNally, 1963), pp. 9fGoogle Scholar. See also Bauer, Raymond A., Pool, Ithiel de Sola, and Dexter, Lewis Anthony, American Business and Public Policy (New York: Atherton, 1963)Google Scholar, part V. Donald R. Matthews makes the same point with reference to Senators: “Without the most stubborn and conscientious efforts, a senator is almost certain to see and talk mostly with friends and supporters on such a trip [to his constituency]. Since both categories are likely to be in general agreement with him, the image of constituency opinion he brings back to Washington is usually distorted in favor of his own views”: Matthews, Donald R., U.S. Senators and Their World (New York: Vintage Books, 1960), p. 229Google Scholar.

The first proposition seems inconsistent with the frequent emphasis on the “localism” of Congressmen; see Huntington, Samuel P., “Congressional Responses to the Twentieth Century,” in Truman, David B. (ed.), The Congress and America's Future (Englewood Cliffs, N.J.: Prentice-Hall, 1965), pp. 5–31Google Scholar, especially Table II, p. 13; also Truman, David B., “Federalism and the Party System,” in Macmahon, Arthur W. (ed.), Federalism: Mature and Emergent (Garden City, N.Y.: Doubleday, 1955), pp. 115–136Google Scholar. Even though restricted to the civil rights issue area (in 1958), Cnudde and McCrone's findings, if valid, would be of considerable substantive importance in view of the above literature.

⁸ This seems to be the approach of Cnudde and McCrone; see the reasoning leading to their model III (p. 69) and the comment about parsimony (p. 72).

⁹ Model 2(c) is actually somewhat plausible; it implies that the representative's perceptions of constituent attitudes have an impact only when mediated by voting in accordance with district opinion, that is, playing the role of agent of the constituency.

¹⁰ In addition, the absence of a direct link between district opinion and Congressmen's attitudes in a single issue area would not seem to be an adequate basis for inferences about “elite recruitment.”

¹¹ The three models just discussed are hierarchical models; they force the investigator to decide at the outset, for example, whether attitudes cause perceptions or the other way around. Yet both these alternatives seem excessively strong in the light of research revealing the significant interaction between attitudes and perceptions. Miller and Stokes (op. cit., p. 51) observe: “Out of respect for the processes by which the human actor achieves cognitive congruence we have also drawn arrows between the two intervening factors, since the Congressman probably tends to see his district as having the same opinion as his own and also tends, over time, to bring his own opinion into line with the district's.” Lacking adequate theory to justify temporal or psychological Prediction priority of one variable over another, one will be unable to select between a number of hierarchical models that fit the data.

The assumptions made in a reciprocal model, in contrast to a hierarchical model, allow some variables to be both the cause and effect of each other. Such a model would, in theory at least, help disentangle attitudes and perceptions. Estimation of links in such models is a difficult empirical matter, however. On the problems and requirements in estimating the parameters of reciprocal models, see Valavanis, op. cit., chs. 4 and 6; and Johnston, J., Econometric Methods (New York: McGraw Hill, 1963)Google Scholar, ch. 9.

¹² Alker has drawn attention to the importance of this point in inter-nation comparisons. See Alker, Hayward R. Jr., “Regionalism Versus Universalism in Comparing Nations,” in Russett, Bruceet al., World Handbook of Political and Social Indicators (New Haven: Yale University Press, 1964), pp. 322–340Google Scholar; and also Blalock, Hubert M. Jr., “Theory Building and the Statistical Concept of Interaction,” American Sociological Review, 30 (06, 1965), 374–380CrossRefGoogle Scholar. In many cases it may be useful to transform the variables to eliminate nonadditive effects. See Kruskal, Joseph B., “Transformations of Data,” International Encyclopedia of the Social Sciences (Macmillan, 1968), vol. 16, pp. 182–193Google Scholar.

¹³ Miller, Warren E., “Majority Rule and the Representative System of Government,” in Allardt, Erik and Littunen, Yrjo (eds.), Cleavages, Ideologies, and Party Systems: Contributions to Comparative Political Sociology (Helsinki: Proceedings of the Westermarck Society, 1964), pp. 343–376Google Scholar.

¹⁴ Cnudde and McCrone, op. cit., p. 72.

¹⁵ Although Cnudde and McCorne restrict themselves to the civil rights issue area, their hypotheses would, if true, have more general significance. We tested their model for welfare and foreign policy issues in competitive and non-competitive districts (based on the data in Miller, op. cit.). The model failed to fit in all four of these tests.

¹⁶ Matthews and Prothro, op. cit., pp. 323–324, emphasis in the original.

¹⁷ See the reasoning about “direction of causation,” ibid., pp. 321–323.

¹⁸ The beta weights included in the analysis (ibid., p. 321) also run contrary to the model.

¹⁹ Blalock, Hubert M. Jr., “Correlated Independent Variables: The Problem of Multicollinearity,” Social Forces, 62 (12, 1963). p. 233CrossRefGoogle Scholar. For more detailed discussions of the problem, see Farrar, Donald E. and Glauber, Robert R., “Multicollinearity in Regression Analysis: The Problem Revisited,” Review of Economics and Statistics, 49 (02, 1967), 92–107CrossRefGoogle Scholar; and Johnston, op. cit., pp. 201–207.