o u INFLUENCE RANKING IN THE UNITED STATES SENATE*" by Robert A. Dahl James G. March David Nasatir (Yale University) (Carnegie Institute of Technology) (Stanford University) * Paper to be read at the meetings of the American Political Science Asso elation, Washington D.C, September 6-8, 1956. The work on which the paper is based was done while the first two authors were Fellows at the Center for Advanced Study in the Behavioral Sciences. Acknowledgement is due Mr. Alex Clark for his help in assembling and processing some of T.he data.
INFLUENCE RANKING IN THE UNITED STATES SENATE I Whether the study of influence is at the very center of political science is no doubt debatable. But it is incontestable that the study of politics could not proceed very far without making seme assumptions about influence; and it is safe to say that no great political theorist from Socrates onward has found it possible to ignore the idea of influence. Political science is shot through with terms like influence and equivalent or closely related concepts such as power, control, and authority. It is difficult to imagine a political science that sought to exclude these concepts, or what is more, tried to avoid comparing different individuals, roles, organizations, behaviors, institutions, or states with respect to their relative influence, or with respect to their influence at different times. That political writers do, in fact, make numerous comparisons of relative influence is easily shown. One may choose examples from political writings almost at random. Thus Machiavelli in the Discourses: "Amongst the other indications by which the power of a republic may be recognized is the relation in which they live with their neighbors; if these are tributary to her by way of securing friendship and protection, then it is a sure sign that the republic is powerful;" or de Tocqueville: "Several particular circumstances combine to render the power of the majority in America not only preponderant, but irresistable; " or Bagehot: "The State of Delaware ie not equal in power and influence to the State of New York, and you cannot make it so by giving it an equal vote in the Upper Chamber;" or Woodrow Wilson: "The chairman of a great Committee... neither directs, nor can often influence,
2 those other chairmen who direct all the other important affairs of govern ment. " The question arises, therefore, whether one can measure relative influence in any meaningful way. Something like Wilson's statement is pretty widely echoed by observers of Congress; it is gospel that a committee chairman generally has more influence over legislation falling within the domain of his committee than do congressmen not on his committee. If such assertions are not entirely arbitrary, a comparison of influence among congressmen is clearly implied. Is it possible to determine the underlying logic of such comparisons and then to develop a definite measure that incorporates the logic? The general object of the present paper is to provide a partial answer to such questions and to a set of more specific empirical propositions concerned with influence in the United States Senate. II In a paper that is reproduced here as appendix A, March has attempted to discuss the general problem of influence measurement as It relates to the concepts of Influence inequalities among roles and among behaviors. The criteria discussed there involve identifying a set of behaviors, a set of roles, a set of possible outcomes, and a measure defined on that set of outcomes. A general knowledge of the concepts discussed in that paper will be assumed here. As is obvious from the nature of the concepts discussed in the appendix, the most elegant study of Influence ordering in the Senate would require a relatively large scale acquisition of original data. Nevertheless, it is possible to make a preliminary investigation on the basis of secondary sources
3 provided suitable caution in interpretation is exercised. This paper is essentially a report of such an investigation of some aspects of the influence structure among the 34 senators who held office continuously from early 1946 through late 1954. At the outset, therefore, we wish to specify the major problems specific to this application. For a discussion of more general problems, the reader is referred to the comments in appendices A and B We consider as roles the 34 senators, as behaviors being in favor of or opposed to the passage of a motion, as outcomes the probability of a bill being favored by a majority of the other senators, and a set-theoretic measure. Thus, between any two senators, S± and S2, we can construct a matrix of the form indicated in Table 3. in which the cell entries (p.,1) Table 1 about here refer to the probability of a majority of the remainder of the Senate voting in favor of a motion under the conditions indicated. Thus, the n J. are virij tually the probability of a motion pussing the Senate. Although we believe it represents the best translation of the concepts with which we wish to deal, the matri;. portrayed in Table 1 raises a number of problems that we wish to note explicitly and discuss briefly. (1) Impossibility of a complete analysis. With 34 senators there are, 34 of course, 2 different combinations of two behaviors. Consequently, for a complete analysis along the llnei; indicated in appendix A we would require estimates of outcomes (i.e., the bahavior of the remainder of the Senate) in 17,179,869,184 different situations, most of which rarely if ever occur. In view of this problem, the analysis presented here has been based on a large
4 S l Table 1: General form of influence matrix for comparing two senators
5 number of comparisons between pairs of senators. Such an incomplete analysis raises two problems that are not present in the general concepts. First, paired comparisons may introduce some intransitivlties in the order ings. In fact, some modest Intransitivlties are recorded in our data. Second, paired comparisons may exaggerate the position of a "satellite," that is a senator who is highly correlated with a powerful colleague. Thus, the necessities of the data require us to weaken the general concepts (which are designed to avoid the correlation problem). We would require additional data to explore in any satisfactory detail the extent to which this is an empirical as well as a logical problem here, but we would not want to deny its importance. (2) Units of analysis. To estimate the probability of a motion securing majority support under a given circumstance, we have computed the proportion of motions that have gained a majority under such circumstances over the period studied. Such a procedure weights all motions equally, and we believe this to be the most defensible procedure in the present case. On the other hand, observers who wish to capture their conceptions of the "importance" of motions in the criterion can do so by specifying operational weighting rules. (3) Aggregation over time. A senator's influence probably is not stable over shorter periods and almost certainly changes over a nine-year span. Less obvious but probably also true is the proposition that relative influence among the 34 senators involved here changes somewhat over the period covered. By aggregating observations over the whole period, we eliminate the possibility of recording such shifts. Aggregation is made necessary by the desire to increase the sample sizes on which are based the estimates in the cells, but in the best of all possible worlds it would not be necessary. We present below some data indicating differences in the rankings associated with shifts in
6 party control of the Senafce; but a more detailed analysis would involve the use of extremely small frequencies or the stimulation of answers to hypothetical situations by means of field research. By means of the paired comparison technique, we have established rankings of the 34 senators in two different policy areas foreign affairs and tax and economic policy. The classification of motions to the two areas made by the Congressional Quarterly Almanac for each year has been accepted. The specific ranking procedures are indicated in appendix C. There was no attempt to make all 561 possible comparisons; rather, the object was to make enough comparisons to establish an unambiguous (or nearly unambiguous) order. In all, 158 paired comparisons were made for the foreign policy rankings over the whole period and 206 paired comparisons for the comparable tax and economic policy rankings. Thus, on the average, each senator was compared with about nine others in the first case and with about twelve others in the second. 11l In Tables 2 and 3 the rankings of the thirty-four senators according Tables 2 and 3 about here to the criteria mentioned above in the two areas of foreign policy and tax and economic policy are indicated. In order to avoid burdening the discussion with considerations of only moderate theoretical interest, we have presented the data using pseudonyms. We should be happy to provide a key to the results for persons having uncontrollable curiosity or an interest 1/ ln checking the rankings. As was indicated above, the procedures
7 Gunn (tie) Olds Court Xerxes** Foster** Pottle Graham** Hughes* (tie) York** Adams Irwin Law Ely* Upshaw* Murphy* Taylor Felton** (tie) Walsh Kelly Decker Howe** Eaton Sherman (tie) Roberts Dunstan Vernon Aster* Jensen Brown Baker Nash* Zimner Caspar* Quigley Table 2: Foreign Policy Influence Order among United States Senators (pseudonyms), 1946-1954.» ** member of Foreign Relations Committee five or more years * member of Foreign Relations Committee one-four years
Foster## Sherman## Eaton Brown## Walsh# Decker Quigley Zimmer Irwin## Pottle Jenean## (tie) Roberts Howe Duns tan Vernon Xerxes Murphy Adams Caspar Kelly Astor Gunn (tie) Court Baker## (tie) York (tie) Ely Nash (tie) Hughes (tie) Upshaw (tie) Olds Taylor (tie) Law (tie) Felton (tie) Graham Table 3: Tax and Economic Policy Influence Order among United States Senators (pseudonyms), 1946-1954. ## member of Finance Committee five or more years # member of Finance Committee one-four years
9 necessitated by our data resulted in a few cases of intransitivifcy but in general our results are consistent with the rankings indicated and reversals 2/ are rare. A comparison of the two rankings yields some striking results. First, it is clear from an inspection of Tables 2 and 3 that membership in a relevant Senate committee substantially increases the probability of a high ranking. Second, the rank order correlation (tau) between the two rankings is actually negative (-.20) but not significantly different from zero. This second result suggests the possibility of classifying the senators according to a two-dimensional influence classification. In Table 4 such Table 4 about here a classification is made by dividing both rankings into trichotomies (high, medium, low). We ask whether this nine-fold classification of senators is systematically related to other observable attributes of their senatorial role. Specifically, we have tested the relationship between influence and (a) party, (b) size of state from which elected, (c) region from which elected, and (d) seniority by comparing the "High-High", "Medium- High'i and "High-Medium" groups on the one hand with the "Low-Low", "Low-Medium", and "Medium-Low" groups on the other. We find no consistent relationship between seniority and influence position within thie group nor between either region of the country or party membership (see below for a reason) and influence. We find a slight (but not statistically significant) tendency for the more influential senators to be elected from the more populous states. One obvious factor that might well affect the influence ranking of a senator is party control of the senate. In general, we would predict that
10 Tax and Economic Policy Table 4: Two-Dimensional Influence Classl.J cation of United States Senators (pseudonyms;, 1946-1954. ** member of Foreign Relations Committee five or more year * member of Foreign Relations Committee one-four years M member of Finance Committee five or more years # member of Finance Committee one-four years
11 Democratic senators would be more influential with other Democrats than with Republicans and vice-versa and that, therefore, the more Democrats there were in the Senate the more influential would be any given Democrat. In addition, party control of the formal senatorial machinery almost certainly increases the opportunities for influence over and above the increase due to simple numbers. To test such propositions, we have computed the foreign policy rankings for those years in which the Democrats controlled the Senate (1946, 1949, 1950, 1951, 1952) and compared them with similar rankings based on the years in which the Republicans controlled (1947, 1948, 1953, 1954). That Democratic senators are more influential in a Democratic-controlled Senate than in a Republican-controlled Senate and Republican senators are more influential in a Republican-controlled Senate than in a Democratic-controlled Senate is confirmed by the results indicated in Table 5. Table 5 about here The effect of the party-control mechanism is to increase systematically the influence of majority senators. We would expect, however, that it would have significantly less effect on the influence order within a given party. Our data are consistent with such an hypothesis. The rank correlation coefficient between the two foreign policy rankings taken as a whole (.19) is not significantly different from zero. On the other hand, if the two parties are treated separately, both show correlations (.38 and.41) significant at the.05 level despite the reduction in sample size involved in splitting the total sample. Finally, we have attempted to extend our analysis to an investigation of the relationship between roles in which different individuals have acted over the nine year period. In particular, we have examined the influence ordering
12 Table 5: Relationship between Party Membership and Influence Rankings in Democratic and Republi.ean-c ntrolled Senates, 1946-1954. Note N equals 31 because one senator (Law) had the same rank in both rankings, one senator (Nash) could not be effectively ranked in one case, and one senator (Taylor) could not be classified unambiguously according Co political party.
among the majority floor leader, the minority floor leader, the chairman of the Senate Committee on Foreign Relations, and the chairman of the Senate Committee on Finance. Over the nine year period there were six majority floor leaders, five minority floor leaders, three chairmen of the Foreign Relations Committee, and two chairmen of the Finance Committee; and we aggregate without respect to the personalities involved in an attempt to eliminate through averaging the Idiosyncratic elements of the role-players. This analysis yields the following orders. Foreign Policy Majority floor leader Finance committee chairman Foreign relations committee chairman Minority floor leader Tax and Economic Policy Finance committee chairman Foreign relations committee chairman Majority floor leader Minority floor leader The results are hardly conducive to simple explanations. We will simply note first that aggregation for the role of Finance committee chairman involved just two individuals, one of whom (Foster) ranked very high in both policy areas. Consequently, the fact that the "role" of Finance committee chairman appears more influential than the "role" of Foreign relations committee chairman in foreign policy can be interpreted as due in large part to the unique position of this senator. On the other hand, the difference between the majority leader's apparent influence in the two areas may be of greater theoretical importance. In particular, we can raise the question (without providing an answer here) of the extent to which the difference is related to comparable differences in Presidential power and the character of the usual relationship between the President and the majority leader.
14 IV Can the general technique outlined here and in appendix A be extended successfully to the study of other institutions? Or to other role comparisons in Congress? If one is limited to secondary sources, studies of state legislatures, multiple judge courts, and other institutions in which individual positions are recorded is possible. In addition, studies of the position of formal roles comparable to the present analysis of party floor leaders and committee chairmen can be fruitfully pursued. However, if finer nuances of relationships and particularly if an extension to more types of behavior is desired, original field research and the exploration of hypothetical situations seems indispensible. In this regard, we can report that one of us (Dahi) has found the concepts useful in planning a study of community decision-making in a middle-sized eastern city and another of us (March) has used them in developing research on the political institutions of an American Indian tribe.
.. t» FOOTNOTES 1 Such a key will be provided at the meetings. In addition, the code itself is easily broken by anyone sufficiently motivated to do so. 2 Specifically, only 1.4% of the 364 pairings involved are inconsistent: with the rankings presented. 3 See the remarks in appendix B concerning the possible need to extend the system beyond the confines of the Senate.