Do legislators’ votes reflect constituency preference? Several investigators have made impressive econometric attempts to answer this question in recent years (Kau and Rubin, 1982; Kalt and Zupan, 1984a, 1984b; Peltzman, 1984, 1985; Poole and Daniels, 1985; Nelson and Silberberg, 1987). The thrust of these studies is to discriminate between legislators’ votes driven by constituency preference—via the legislators’ self-interest in pleasing the constituency and thereby gaining reelection—and legislators’ votes driven by their own ideological preferences.

I have argued elsewhere (Higgs, 1987, 1989) that this distinction may be inadequate to capture how ideology actually operates, but for the point I want to make here I need not insist on any more subtle understanding of the distinction between self-interest and ideology than previous contributors, cited above, have maintained. Let us suppose then that a legislator may vote as he does either because he wants to express the constituency preference or because he wants to express his personal ideological preference.

For empirical testing, the immediate problem is that neither preference can be directly observed; hence, the investigator must resort to proxy variables. The typical procedure is to estimate a regression equation in which the legislator’s vote (yea or nay) is the dependent variable. A set of district attributes (e.g., percent of labor force unionized, percent of population urbanized, black, over 65, and so forth) measures what is putatively in the interest of the voters. An ideology index, itself derived from other votes cast by the legislator, measures the influence of personal ideology.

Of course, applied econometric studies always come equipped with a full set of actual and potential problems. Not surprisingly, the evolution of the literature on this issue has witnessed the typical progression from fairly simple tests toward increasingly complicated and incomprehensible procedures. If only the latest could really be trusted to provide more reliable conclusions than their predecessors.

Another Approach

Fortunately, there is a better way. At least, there is a very simple way to assess voting in the United States Senate. This procedure requires no econometics at all. It requires only two assumptions. It allows us to give a definite answer to the question of how often senators in their voting depart from the constituency preference of the states they represent. The answer is a lower bound; they may deviate more frequently, but the proposed test cannot tell us how much more frequently. My test exploits the unarguable fact that every state has two senators; both represent the same constituency. Whenever a state’s two senators cast votes that disagree, we know that one of them is voting against the constituency preference.

In pursuing this line of argument, we make two assumptions.

First, the constituency preference exists. Of course, for many votes cast in the Senate, almost no one in the state knows or has given much thought to how he would prefer that his senators vote. Although we might impute a preference to the voters on the basis of other information, the truth may be that a constituency preference, in any solid sense, just doesn’t exist; like “public opinion,” it would have to be created by lobbyists, media commentators, politicians, and others who happen to follow the issue (Bennett, 1980). Even though this possibility is real, its implications for voting studies are devastating. If there is no constituency preference, the question of whether a legislator’s vote reflects that preference is meaningless. To do this sort of analysis, one simply must assume that a constituency preference exists.

Second, one must assume that the legislators know what it is. If a constituency preference exists but the senators don’t know what it is, they must either guess and vote accordingly or disregard the constituents and vote to please themselves. Either way, their actions could hardly be employed to test how often their votes reflect the constituency preference. A reflection that is merely coincidental is not what democratic theory either asserts as fact or exalts as norm.

An extended argument, marshalling a variety of evidence, could be presented in support of my two assumptions, but I shall refrain from making that argument. All previous contributors to this literature, either expressly or implicitly, make these assumptions. For present purposes, I simply follow in the same track.

What does it mean when both senators cast the same vote, that is, both vote yea or both vote nay? It may mean that both votes are “constituency driven,” which is to say, both senators are following the known constituency preference. Or it may mean that both votes are “ideologically driven,” which is to say, both senators are ignoring the known constituency preference and voting instead in accordance with personal ideology, which happens to lead them both to the same decision and may or may not agree with constituency preference. Or the pair’s matching vote may be “mixed driven,” which is to say that one senator votes in accordance with the known constituency preference while the other senator votes the same way but because of ideological preference, not because of constituency preference.

Table 1. Possible conditions if the votes of a state’s two senators match or disagree,
given a known constituency preference for the affirmative.

Senator #1
Senator #2
Votes Match Y( c ) Y( c ) Both constituency driven
Y( c ) Y ( i ) Mixed driven
Y ( i ) Y ( c ) Mixed driven
Y ( i ) Y ( i ) Both ideologically driven
N ( i ) N ( i ) Both ideologically driven
Votes Disagree Y ( c ) N ( i ) One ideologically driven
Y ( i ) N ( i ) Both ideologically driven
N ( i ) Y ( c ) One ideologically driven
N ( i ) Y ( i ) Both ideologically driven

As Table 1 shows, five different possible conditions may underlie a matching vote. (Note that Table 1 is constructed for the case where the constituency preference is for the senators to vote yea.) In the table, the symbols are as follows: Y = yea, N = nay, ( c ) = constituency driven, ( i ) = ideologically driven, all as explained in the previous paragraph. The table also shows that when the votes of the two senators disagree, one or both of the votes must be ideologically driven. Notice that two possibilities are ruled out: because we assume that the senators know the constituency preference, it is not possible for either senator to be constituency driven to vote against a position the constituency favors.

Table 1 represents the situation where the constituency prefers the affirmative. A similar table can be constructed for the situation where the constituency favors a negative vote, but nothing would be added by constructing this table. The point is the same.

In sum, the logic could hardly be simpler. When a state’s two senators vote differently, one of them is necessarily voting against the constituency preference; one or both are ideologically driven. Although a matching vote may be consistent with constituency driven voting, it may also be purely ideologically driven or driven by mixed motives for the two senators.

Table 2. Voting by the U.S. senator state pairs on 10 defense-related motions, 1987

Vote numbers
1 2 3 4 5 6 7 8 9 10 Totals
Pairs of same party (max=28)
Pairs differing 4 2 3 6 5 10 8 6 4 3 51
Pairs voting 20 26 23 28 28 28 27 27 28 27 262
% differing 20 8 13 21 18 36 30 22 14 11 19
Pairs of different parties (max=22)
Pairs differing 14 15 11 10 13 10 9 11 14 15 122
Pairs voting 19 21 15 22 21 21 21 21 21 21 203
% differing 74 71 73 45 62 48 43 52 67 71 60
All pairs (max=50)
Pairs differing 18 17 14 16 18 20 17 17 18 18 173
Pairs voting 39 47 38 50 49 49 48 48 49 48 465
% differing 46 36 37 32 37 41 35 35 37 38 37
Differing pairs of different parties as % of all differing pairs 78 88 79 63 72 50 53 65 78 83 71

An Application

To apply the method proposed above, I compiled information on Senate voting on 10 defense-related motions in 1987. These measures were selected by the compilers of a reference work, Congress & Defense 1988, to establish the extent to which each member of the Senate supported more defense spending. The votes were “selected solely on the basis of whether they were relatively clear indications” of a senator’s “opinion for or against higher levels of military spending on a well-defined program or group of programs” (Hull, 1988: vii). The ten measures, summarized in my appendix, do have implications for defense spending. They are also, in my judgment, all related to sensitive arms-control issues. I do not think that this attribute makes them any less suitable for analysis here, but readers are warned so that they can apply any discount they consider justified by the character of the votes. In any event, the data set was generated independently of my analysis.

Each of the votes considered was fairly close. No vote resulted in more than 61 yeas; each resulted in 32 or more nays. (See the appendix for the outcomes of each vote.) It is important to consider fairly close votes, because a search for mismatching votes by senatorial state pairs would be doomed to fail on very lopsided votes. If a vote is unanimous, then obviously every state pair must match. To the extent that the total vote approaches unanimity, the potential for mismatching state pairs diminishes.

For each of the 10 votes, I counted the number of senatorial state pairs voting on the motion. Whenever one or both of a state’s senators did not vote (votes included so-called “paired” votes, which are of course something different from the state pairs being studied here) or did not announce a position, that state’s pair was dropped from the count. I then counted the number of differing state pairs, that is, the number of voting pairs for which one senator voted differently from the other senator of the same state. Differing pairs never amounted to less than 32 percent of all voting pairs, and they reached as high as 46 percent. On average for the 10 votes, differing pairs made up 37 percent of all state voting pairs. (See Table 2 for detail.)

To determine how much of the state pair voting mismatch was associated with mismatch of political party, I compiled the data separately for states with both senators of the same party and states with senators of different parties. Overall, party-mismatched state pairs accounted for 71 percent of the voting mismatches. For the individual votes, the extent of this association ranged from 50 percent to 88 percent (See Table 2 for detail.)

Party affiliation hardly accounts for all the state pair vote splitting. Individual detail (not shown here), indicates that state pairs of the same party sometimes differed frequently in their votes on the 10 measures considered. For example, the senatorial pairs of the same party from Georgia, Minnesota, Missouri, Ohio, Oregon, and West Virginia differed on at least five of the 10 votes. On the other hand, belonging to a different party by no means assured voting differently. Senatorial pairs of differing parties from Connecticut, Maine, Mississippi, Oklahoma, Rhode Island, South Carolina, Vermont, Washington, and Nebraska cast matching votes on at least five of the 10 votes. Rhode Island and South Carolina’s senatorial pairs, mismatched by party, cast matching votes on all 10 motions.


U.S. senators frequently vote against the preference of their constituency, assuming that such a preference exists. Both of a state’s senators represent the same constituency. Whenever they split their votes, one or the other is necessarily going against the constituency preference. For the sample of defense-related votes analyzed above, “misrepresentation”—either observable vote splitting or unobservable vote matches that go against the constituency preference—occurred at least 37 percent of the time, at least 46 percent on one vote. Although party differences accounted for more than two-thirds of the vote splitting, a substantial number of splits remained. Besides, a party difference for a state’s senatorial pair is itself problematical.

The method employed here can be applied easily to any data whatever on senatorial voting. Its application will show that, quite often, many senators depart from constituency preference. This finding refutes the hypothesis, popular in certain circles, that ours is a more or less “perfect political market” with little or no scope for ideologically driven voting by legislators.


Bennett, W. L. (1980). Public Opinion in American Politics. New York: Harcourt Brace Jovanovich.

Higgs, R. (1987). Crisis and Leviathan: Critical Episodes in the Growth of American Government. New York: Oxford University Press.

Higgs, R. (1989). Organization, Ideology and the Free-rider Problem: Comment. Journal of Institutional and Theoretical Economics 145 (March).

Hull, S., (ed.) (1988). Congress and Defense 1988. Palo Alto, CA: EW Communications.

Kalt, J. P. and Zupan, M. A. (1984a). Capture and Ideology in the Economic Theory of Politics. American Economic Review 74 (June): 279-200.

Kalt, J. P. and Zupan, M. A. (1984b). The Ideological Behavior of Legislators: Rational On-the-job Consumption or Just a Residual? Harvard Institute for Economic Research, Discussion Paper Number 1043 (March).

Kau, J. B. and Rubin, P. H. (1982). Congressmen, Constituents, and Contributors: Determinants of Roll-call Voting in the House of Representatives. Boston: Nijhoff.

Nelson, D. and Silberberg, E. (1987). Ideology and Legislator Shirking. Economic Inquiry 25 (January): 15-25.

Peltzman, S. (1984). Constituent Interest and Congressional Voting. Journal of Law and Economics 27 (April): 181-210.

Peltzman, S. (1985). An Economic Interpretation of the History of Congressional Voting in the Twentieth Century. American Economic Review 75 (September): 656-675.

Poole, K. T. and Daniels, R. S. (1985). Ideology, Party, and Voting in the U.S. Congress, 1959-1980. American Political Science Review 79 (June): 373-399.


Votes used for the test

Vote #1. Motion to table an amendment allocating $5 million to replace 100 Minuteman II ICBMs with Minuteman III ICBMs. (Replacing two-warhead Minuteman IIs with three-warhead Minuteman IIIs would exceed limits of unratified SALT II agreement.) Approved 57-32.

Vote #2. Motion to table an amendment lifting restrictions on development and testing of space-based and other mobile ABM systems. Approved 58-38.

Vote #3. Motion to table an amendment requiring replacement of 50 Minuteman II ICBMs with single warheads by Minuteman IIIs with three warheads. Approved 51-33.

Vote #4. Motion to table an amendment authorizing $3.7 billion for SDI in FY88. (Administration sought $5.7 billion.) Approved 51-50.

Vote #5. Motion to table an amendment barring ASAT tests against targets in space. Approved 51-47.

Vote #6. Motion to table an amendment placing conditions on nuclear tests with explosive yields greater than one kiloton in 1988-89. Approved 61-36.

Vote #7. Motion to table an amendment delaying assembly of binary chemical weapons until October 1, 1988. Approved 53-44.

Vote #8. Motion to table an amendment eliminating funds for production of Bigeye nerve-gas bomb. Approved 49-48.

Vote #9. Motion to table an amendment limiting the number of certain nuclear weapons deployed to ceilings set by unratified SALT II agreement. Rejected 44-55.

Vote #10. Motion to maintain SALT II deployment ceilings for certain nuclear weapons. Approved 57-41.

Source: Hull (1988: 156-157).