# The influence of ideology on congressional voting.

I. INTRODUCTION

The question of what determines congressional voting on legislation has inspired a number of recent empirical analyses. All are in agreement that congressional voting on specific bills is correlated with the economic self-interest of the representative's constituents and a measure of the representative's ideology as indicated by his broader voting record. However, the interpretation of the empirical results regarding the influence of ideology has generated more debate than consensus in the current literature.

Kau and Rubin [1979] find evidence that a congressman's ideology influences his voting, but also find evidence for an alternative, but not mutually exclusive, hypothesis that a congressman's rating by ideological groups reflects membership in a logrolling coalition. Kalt and Zupan [1984] argue that although a representative's votes are likely to reflect the economic interests and the ideology of his constituents, the weak incentive for the constituent (principal) to monitor the voting behavior of his representative (agent) gives the representative the opportunity to "shirk" or vote according to his personal ideology rather than his constituents'. Both Nelson and Silberberg [1987] and McArthur and Marks [1988] find that the degree of shirking increases as the opportunity cost of shirking decreases.

In contrast to the above findings, Peltzman [1984] argues that ideological voting by congressmen need not necessarily reflect shirking, but instead may be a way for the representative to cheaply provide his constituents with information about his general position on issues. What appears to be shirking allowed by an imperfect principal-agent relationship may instead be signaling designed to at least partially overcome the imperfect nature of the relationship. Lott [1987] extends Peltzman's signaling argument by hypothesizing that if politicians are viewed by constituents as "search" goods rather than "experience" goods, then they will have an incentive to establish a "political brand name" by establishing a consistent ideological voting record. Dougan and Munger's [1989] approach is similar to Lott's. They argue that politicians with strongly held views that match those of their constituents make ideal representatives because they are less likely to vote against their constituents' interests when pressured by special interest groups. An incumbent therefore has an incentive to invest in "reputational capital" by maintaining the consistency over time of his broader voting record because this reputational capital generates quasi-rents in electoral competition with candidates who possess less of it. Neither Lott nor Dougan and Munger find evidence of shirking in their analyses of congressmen's broader voting records as measured by ideological groups' ratings.

Dougan and Munger also develop a "variable intensity of commitment" version of the reputational capital hypothesis. This version contends that voters with unambiguously liberal or conservative ideologies may prefer a candidate whose voting record indicates that his views are more extreme than their own. Such a zealous candidate, voters believe, is even less likely to be influenced by special interest groups. A congressman serving an unambiguously liberal or conservative constituency therefore has an incentive to establish and maintain an extremely ideological overall voting record. In contrast, a congressman serving a constituency with centrist ideological preferences has an incentive to establish an overall voting record that matches his constituency's ideological leanings. The systematic deviation of a congressman's voting record from his constituents' preferences, when near either extreme of the ideological spectrum, is not inconsistent with a faithful principal-agent relationship.

One cannot help but be struck by the observation that those researchers finding evidence consistent with shirking studied voting on specific legislation, whereas those researchers finding no evidence of shirking studied representatives' broader voting records as measured by ideological ratings. When analyzing voting on a particular bill, how does a researcher determine whether his empirical findings reflect shirking or reputational capital? After all, reputational capital is established by consistent ideological voting on a great number of bills. This does not, however, rule out the possibility that shirking did occur for voting on any particular bill. (1) This paper investigates more completely the influence of ideology on congressional voting. It presents a more general conceptual framework for analyzing voting than has been utilized heretofore. A representative is assumed to vote on legislation so as to maximize the expected value of his office. Constituent ideology and representative ideology, as well as the direct economic interests of constituents and representatives, influence voting by inducing the representative to make the standard trade-offs at the margin. The empirical work takes advantage of two congressional votes to conduct a natural experiment on the influence of ideology on voting. Analyses of voting on two floor amendments to the U.S. House Administration Committee's 1974 Federal Election Campaign Act bill indicate that ideological considerations, whether of congressmen or their constituents, are ignored when the direct self-interest of the congressman to do so is sufficiently large.

II. CONCEPTUAL FRAMEWORK

Assume that a congressman's objective is to maximize the expected value of his office, which is equal to the product of the value of his office and his probability of reelection. Let the value of the office to a congressman be an increasing function of the total compensation of the office and the exercise of his personal ideology through his voting. The inclusion of the congressman's personal ideology implicitly assumes that he receives utility from attempting to influence legislation in what he deems to be the "correct" direction.

Assume that probability of reelection varies directly with the degree to which a congressman's voting satisfies the economic interests and ideological preferences of his constituents, as well as directly with his election campaign expenditures. Larger campaign expenditure presumably enables a congressman to more effectively present his record to the voters in a favorable light. Other factors that likely influence probability of reelection are the fraction of voters registered in the incumbent's party and the incumbent's party affiliation. The former indicates the fraction of the electorate likely to hold the same position on the issues that they expect the incumbent to hold. The latter reflects any possible national swings of preference toward one party or the other. These last two variables are parameters.

If a congressman's personal ideology does not coincide with the ideology and economic interests of his constituents, then voting on legislation according to his personal ideology and against his constituent's interests and ideology will lower his probability of reelection. However, the failure to act in accordance with his personal ideology will lower the value of holding office. A congressman will vote so as to satisfy the standard marginal conditions for maximizing the expected value of his office. This is consistent with the argument by Dougan and Munger [1989] that a congressman whose personal ideology is identical to his constituents' ideology is in a stronger long-run competitive position than a congressman whose ideology is different than his constituents', because the former can satisfy his constituents at a lower cost.

Campaign contributions from special interest groups also may influence a congressman's voting. To the extent that these groups' objectives conflict with the economic or ideological interests of his constituents or with his own ideological interests, then voting to accommodate these groups will lower the expected value of the congressman's office by either decreasing his reelection probability or decreasing the value of his office. However, the increased campaign expenditure made possible by additional special interest contributions will increase his reelection probability.

I define shirking by a congressman as his failure to vote according to the economic interests or ideology of his constituents. Shirking can be induced by either the offer of contributions from special interest groups (2) or a divergence between the ideologies of the congressman and his constituents. More generally, any factor that can marginally increase the expected value of a congressman's office while influencing his vote can potentially induce that congressman to shirk. Indeed, a congressman may be induced to vote not only against the ideology of his constituents but against his own ideology as well. Shirking therefore can occur even if the congressman's and constituents' ideologies are identical.

For my empirical analysis of the influence of ideology on congressional voting, I elected to analyze the voting on two floor amendments to the U.S. House Administration Committee's 1974 Federal Election Campaign Act bill. For these two amendments, only the direct congressman-specific benefits and costs generated by each amendment and ideology should affect the voting. Since the amendments differ greatly in terms of the magnitude of their direct benefits and costs, analyses of voting on the amendments is equivalent to conducting a natural experiment to shed light on the influence of ideology on congressional voting.

III. BACKGROUND AND DESCRIPTION OF THE FLOOR AMENDMENTS

The U.S. Congress passed the 1974 Federal Election Campaign Act in October of that year. In brief, the 1974 campaign act provided for partial public funding of presidential election campaigns, ceilings on individual campaign contributions, and ceilings on campaign expenditures by candidates in presidential, Senate, and House elections starting in 1976.

On 7 August 1974, the House Administration Committee's bill was brought to the floor of the House for consideration under the rules of an adopted motion (Congressional Quarterly notation 325) that restricted the scope of potential floor amendments. (For notational convenience and consistency, the House Administration Committee's bill (HR 16090) will be referred to as bill 325.) (3) As in the eventual 1974 campaign act, bill 325 provided for partial public financing of presidential campaigns and ceilings on campaign contributions by individuals and campaign expenditures by candidates for federal office starting in 1976. The campaign expenditure ceiling for U.S. House candidates was $93,750. (4) No provision for any type of public funding of congressional campaigns appeared in 325.

On 8 August 1974 a floor amendment (329) that lowered the expenditure ceiling for House candidates from $93,750 under bill 325 to $75,000 (5) was adopted by a vote of 240-175 (130-102 by Democrats and 110-73 by Republicans). Later that day a floor amendment (333) that provided for partial public funding of congressional campaigns was defeated by a vote of 187-228 (114-118 by Democrats and 73-110 by Republicans). Specifically, amendment 333 would have granted matching federal funds to congressional candidates for all individual contributions of $50 or less up to one-third of the expenditure ceiling. To qualify for these matching funds, the candidate would have to raise 10 percent of his maximum allowable matching funds in contributions of $50 or less.

Since campaign expenditures produce electoral support, both 329 and 333 could have the effect of changing each congressman's probability of reelection. This direct, congressman-specific benefit or cost is the only impact of these amendments. The direct economic interests of the congressman's constituents are not affected, (6) and it is unlikely that there would be an incentive for economic special interest groups to influence a congressman's votes on these amendments. (7) However, passage of each amendment was an important liberal cause. (8) Consequently, congressional voting on both amendments should have been determined solely by the congressman-specific direct benefits or costs and by ideology.

IV. CALCULATION OF CHANGES IN REELECTION PROBABILITIES

In order to calculate changes in reelection probabilities under each amendment, it is necessary to estimate the impact of changes in campaign expenditures on reelection probabilities. This requires an estimated vote-share regression equation having campaign expenditures as independent variables and the ability to calculate the changes in campaign expenditures under each amendment. Calculating such changes is possible only if we know what the unconstrained expenditures would have been in the 1976 elections. In January 1976, the Supreme Court fortuitously declared that the section (608[c]) of the 1974 campaign act setting expenditure limits for congressional election campaigns was unconstitutional (Buckley v. Valeo [46 LEd 2d 659]). Therefore, passage of 329 did not affect campaign expenditures in the 1976 congressional elections. Obviously, neither did 333, since it was defeated.

The vote-share regression equation to be estimated is the one used in Bender [1988]. (9) That regression equation is

(1) INCUMBENT VOTE SHARE = [b.sub.0] + [b.sub.1] (INCUMBENT PARTY STRENGTH) + [b.sub.2] (CHALLENGER EXPENDITURE) + [b.sub.3] (RELATIVE CHALLENGER EXPENDITURE) + [b.sub.4] (DEMOCRATIC INCUMBENT) + e

where INCUMBENT VOTE SHARE is the fraction of total votes cast for the Democratic and Republican candidates in the congressional election received by the incumbent, DEMOCRATIC INCUMBENT is a dummy variable for the incumbent being a Democrat, INCUMBENT PARTY STRENGTH is the fraction of registered Democratic and Republican voters in the congressional district registered in the incumbent's party, CHALLENGER EXPENDITURE is the challenger's campaign expenditure in thousands of dollars, RELATIVE CHALLENGER EXPENDITURE is the challenger's campaign expenditure as a fraction of the sum of expenditures by the challenger and the incumbent, (10) and e is the error term. The inclusion of CHALLENGER EXPENDITURE and RELATIVE CHALLENGER EXPENDITURE presupposes that both the level of spending by the challenger and relative spending by the challenger influence vote share. This specification allows for the marginal products of challenger spending and incumbent spending in the production of vote share each to be a function of the levels of both challenger and incumbent spending. It can be shown that [b.sub.2] < 0 and [b.sub.3] < 0 is a sufficient condition for the incumbent's marginal product to be positive but decreasing with the incumbent's expenditure, and for the challenger's marginal product to be negative but increasing with the challenger's expenditure.

Regression equation (1) was estimated for a sample of seventy-two major party U.S. House of Representatives elections involving incumbents in 1976. (11) The ordinary least squares estimate of (1) is presented below (t-statistics in parentheses):

(2) INCUMBENT VOTE SHARE = 0.6250 + 0.1915 (INCUMBENT PARTY (20.30) (4.15) STRENGTH) - 0.0014 (CHALLENGER (5.36) EXPENDITURE) - 0.0833 (RELATIVE CHALLENGER (1.71) EXPENDITURE

[R.sup.2] = .62, n = 72

Estimated coefficients are m accordance with expectations. The incumbent's vote share varies directly with the strength of the incumbent's party in the district and inversely with the challenger's level of spending and relative spending. The positive constant term can be attributed to incumbency. DEMOCRATIC INCUMBENT had an insignificant estimated coefficient (t-statistic of 0.19) and was therefore dropped from the regression. This indicates that the mood of the national electorate favored neither party in 1976. All fitted values are in the interval [.456,.799], and the standard error of the regression is only 0.065. Finally, at the mean values of incumbent and challenger spending (72.607 and 32.207), challenger spending has a marginal product (-.00193) that is 7.9 times larger in absolute value than the marginal product (.00024) of incumbent spending.

I attempted to measure the impact on the incumbent's vote share, and consequently his probability of reelection, of the divergence of his voting record, as measured by his 1974 rating by Americans for Democratic Action, from his constituents' desired value of that rating. Estimation of this divergence followed the procedure of Kalt and Zupan [1984]. Specifically, I regressed a log-of-the-odds transformation of the congressman's ideological rating (ADA) (12) on the fraction of the congressional district that voted for George McGovern in the 1972 presidential election (MCGOVERN VOTESHARE) and DEMOCRATIC INCUMBENT. The ordinary least squares estimate of the regression is

(3) ADA = -4.4002 + 9.8607 (MCGOVERN (7.88) (6.27) VOTESHARE)

+ 0.9803 (DEMOCRATIC INCUMBENT) (2.86)

[R.sup.2] = .54, n = 72

The fitted value of ADA for each congressman can be interpreted as his constituents' desired value. The absolute value of the residual therefore can be interpreted as the divergence between the congressman's ideological voting record as measured by ADA and the ideological voting record that his constituents desire him to have.

The absolute value of the residual of (3) was added as an independent variable in regression (2), which was then reestimated. Since shirking is expected to reduce vote share, the estimated coefficient is expected to be negative. It was in fact insignificantly positive (t-statistic of 0.95). There are two, not necessarily mutually exclusive, possible explanations for this result. One is that the electoral advantages of incumbency make possible some degree of shirking and that this potential freedom to shirk varies directly with the electoral strength of the incumbent. For example, voters may be more willing to tolerate ideological shirking by an incumbent because of his greater ability to satisfy the economic interests of his constituents in ways other than simply voting (e.g., by obtaining pork barrel benefits). If this should be the case, then the positive, although insignificant, coefficient of the absolute value of the ideology residual may reflect a simultaneity problem. A second possible explanation is that the residuals of congressmen with unambiguously liberal or conservative constituencies may reflect their investments in reputational capital. If so, then these incumbents will not be punished at the polls, and consequently the coefficient of the absolute value of the residual is less likely to be negative or significant.

I now present the methodology used to infer reelection probabilities. Consider the incumbent's actual vote-share outcome in 1976 to be a random draw from an assumed normal probability distribution of outcomes generated by the estimated vote-share regression (2). Let this distribution for each incumbent be described by an expected value equal to the fitted value (INCUMBENT VOTE SHARE *) and by a standard deviation equal to the standard error of the regression (STANDARD ERROR). Probability of reelection for the i-th incumbent can be inferred from

(4) [Z.sub.i] = (INCUMBENT VOTE [SHARE.sup.*.sub.i] -.5) /(STANDARD ERROR)

where [Z.sub.i] is the number of standard deviations that the i-th incumbent's expected vote share lies from the minimum vote share of .5 necessary to win reelection. Given the estimated coefficients of (2), fitted values of INCUMBENT VOTE SHARE, and therefore probabilities of reelection, can be simulated for the expenditure ceiling of $93,750 in the House Administration Committee's bill (325), for the expenditure ceiling of $75,000 under floor amendment 329, and for the partial public financing provisions of floor amendment 333. It should be kept in mind that the maximum allowable public funding of campaigns under 333, as described above, is a function of the expenditure ceiling. Since 329 was passed prior to the consideration of 333, the $75,000 ceiling under 329 is the relevant ceiling to be used for the calculation of public funding under 333. (13)

The direct benefit or cost to the individual congressman of each amendment is the increase or decrease in reelection probability faced by that congressman. The change in reelection probability under amendment 329 is the change in reelection probability caused by lowering the expenditure ceiling from $93,750 under 325 to $75,000 under 329 and is denoted by PROBABILITY CHANGE UNDER 329. The change in reelection probability under 333 is the change in reelection probability caused by the partial public funding provisions of 333 in the presence of the $75,000 ceiling under 329 and is denoted by PROBABILITY CHANGE UNDER 333.

Summary statistics for these changes in reelection probabilities are presented in Tables I and II. Since there was one abstention for the vote on 329 and 333, the full sample size for each vote is seventy-one observations, of which the votes on the two amendments have seventy in common. Although the full sample summary statistics in Tables I and II appear to suggest that 329 and 333 had similar trivial effects on reelection probabilities, this is not in fact the case. While the public financing provisions of 333 did little to help or hurt congressmen, the lower expenditure ceiling of 329 considerably helped a number of congressmen. Amendment 329 increased the reelection probabilities of eleven of the congressmen by an average of .0796 or almost 8 percentage points. Also note that the relatively small standard deviation of PROBABILITY CHANGE UNDER 329 of .0263 for these congressmen indicates that the mean value of .0796 was not driven just by a couple of outliers with large increases in reelection probabilities.

That amendment 329 would have a greater impact on reelection probabilities than 333 is not surprising. By design the 329 expenditure ceiling would be binding only in elections in which one or both candidates would otherwise spend more than the ceiling. Such elections are typically the closer elections (i.e., expected vote shares closer to .5). A given change in expected vote share induced by limiting expenditures will tend to cause a larger change in reelection probability the closer the expected vote share is to .5. (14) This, along with the challenger's marginal product of expenditure in the production of vote share substantially exceeding the incumbent's, explains why those congressmen who benefited from an increase in reelection probability under 329 did so by an average of 8 percentage points. In contrast to 329, the matching public funds provisions of 333 offer no matching funds for a candidate already at the expenditure ceiling under 329. Therefore, 333 has a trivial impact on reelection probability, because it provides only limited public funds to candidates in elections in which the expected vote share is not likely to be close to .5.

V. ANALYSIS OF VOTING ON THE AMENDMENTS

Congressional voting on each amendment should be a function of only the implied change in a congressman's reelection probability and ideology. Define VOTE ON 329 as a qualitative variable taking the value 1 if a congressman voted in favor of amendment 329 and the value 0 if he voted against 329. Analogously define VOTE ON 333 as the qualitative variable for congressional voting on amendment 333. (15) I then estimate a logit regression of the qualitative voting variable on the relevant change in reelection probability and the previously defined ideology variable ADA. The estimated logit regressions for congressional voting on 329 and 333 for the full sample of congressmen are presented below (t-statistics in parentheses):

(5) VOTE ON 329 = - 0.1710 + 22.0598 (PROBABILITY (0.67) (2.04) CHANGE UNDER 329)

[chi square] = 5.86, n = 71

(6) VOTE ON 329 = - 0.1973 + 25.1159 (PROBABILITY (0.76) (2.13) CHANGE UNDER 329)

- 0.1999 (ADA) (1.39)

[chi square] = 7.88, n=71

(7) VOTE ON 333 = 0.3983 + 6.3601 (PROBABILITY (1.54) (0.90) CHANGE UNDER 333)

[chi square] = 0.84, n = 71

(8) VOTE ON 333 = 0.6204 + 4.3984 (PROBABILITY (1.94) (0.53) CHANGE UNDER 333) + 0.8032 (ADA) (3.68)

[chi square] = 21.80, n = 71

The contrasts between the estimates of regressions (5) and (6) and of regressions (7) and (8) are quite sharp. Voting on amendment 329 is significantly, positively related to the implied change in reelection probability under 329; whereas voting on amendment 333 is not at all significantly related to the implied change in reelection probability under 333. On the other hand, voting on 333 is significantly, positively related to ADA; whereas voting on 329 is at best marginally negatively related to ADA. In short, voting on 329 appears driven by the direct personal benefits or costs in terms of implied changes in reelection probability, and voting on 333 appears driven by ideological considerations.

As noted in the previous section, changes in implied reelection probability under 329 were nontrivial for a considerable number of congressmen, while changes in implied reelection probability under 333 were generally negligible. Since both the expenditure ceiling and the partial public funding of campaigns were liberal issues, then a reasonable interpretation of the regression estimates is that ideological considerations were ignored when the opportunity cost of ideological voting was sufficiently high. Furthermore, with the defeat (by a 187-228 vote) of 333, there was no other opportunity for voting on partial public funding of campaigns. In contrast, a congressman who voted against 329 was aware that he could always be on record as voting for an expenditure ceiling by voting in favor of the final House bill (passed by a vote of 355-48), which would necessarily contain the expenditure ceiling provision of either 329 or 325. There was, therefore, an additional incentive for congressmen to pay less attention to ideological concerns when voting on 329 than on 333.

This interpretation of regressions (5) through (8) becomes even more credible in the light of regressions (9) through (12). These are the same as (5) through (8) except that the VOTE ON 329 regressions are run on a subsample consisting of the twenty-nine congressmen who had a non-zero change in reelection probability under 329 and the VOTE ON 333 regressions are run on the subsample of those forty-three congressmen who had a non-zero change in reelection probability under 333. The estimated regressions are

(9) VOTE ON 329 = - 0.5416 + 25.9934 (PROBABILITY (1.16) (2.24) CHANGE UNDER 329)

[chi square] =6.80, n=29

(10) VOTE ON 329 = - 0.5850 + 28.1071 (PROBABILITY (1.24) (2.24) CHANGE UNDER 329) - 0.1321 (ADA) (0.64)

[chi square] = 7.22, n = 29

(11) VOTE ON 333 = 0.3679 + 6.0622 (PROBABILITY (1.05) (0.82) CHANGE UNDER 333)

[chi square] = 0.69, n = 43

(12) VOTE ON 333 = 1.0227 + 7.5560 (PROBABILITY (2.11) (0.80) CHANGE UNDER 333)

+ 0.9747 (ADA) (3.12)

[chi square] = 15.37, n = 43

The differences between the full sample VOTE ON 333 regressions (7) and (8) and the subsample VOTE ON 333 regressions (11) and (12) are slight indeed. This is to be expected if only generally negligible changes in implied reelection probabilities are generated under 333. Ideology is clearly the dominant influence on the amendment 333 voting.

In contrast, there are noticeable differences between the full sample VOTE ON 329 regressions (5) and (6) and the subsample VOTE ON 329 regressions (9) and (10). First, the magnitudes and t-statistics of the PROBABILITY CHANGE UNDER 329 coefficients are somewhat larger in (9) and (10). Second, and more strikingly, the magnitude (absolute value) and t-statistic of the coefficient of ADA are considerably smaller in (10) than in (6). Both of these changes are to be expected if nontrivial changes in implied reelection probabilities are generated under 329. The opportunity cost of voting ideologically is higher for the affected subsample than for the full sample. (16)

Let us look more closely at the coefficient of ADA in (6) and (10). Both the magnitude (absolute value) and t-statistic are larger in the full sample regression (6) than in the subsample regression (10). More importantly, the sign of the coefficient is, surprisingly, negative. This appears to imply that the more liberal (conservative) congressmen are more willing to vote against (for) a liberal issue. In fact it does not.

The insignificant coefficient of ADA (t-statistic of 0.64) in (10) indicates that ideological considerations did not affect the voting of those congressmen whose reelection probability would have changed. But why does the magnitude (absolute value) of the coefficient of ADA increase, and why does it now become significant at the .10 level on a one-tailed test (t-statistic of 1.39) in the full sample regression (6)? One would think that the addition to the sample of those forty-two congressmen unaffected by amendment 329 would result in voting more consistent with ideology and a consequent move towards a positive or less negative coefficient of ADA. An answer is that those congressmen unaffected by 329 in 1976 appear to be considering the impact that 329 may have on their future reelection probabilities. For the subsample of those forty-two congressmen receiving neither an increase nor a decrease in reelection probability, the correlation coefficients of ADA and probability of reelection under the expenditure ceiling imposed by 325, challenger expenditure, and incumbent expenditure are .32, -.23, and -.44. Noting that the measure of conservative ideology varies inversely with ADA, these correlation coefficients indicate that the relatively conservative congressmen tend to have a lower probability of reelection under 325, tend to face higher levels of campaign expenditures by their challengers, and tend to have higher levels of campaign expenditures than do the relatively liberal congressmen. This indicates that the relatively conservative congressmen in this subsample are facing tougher electoral challenges. (17) Regressions (5) and (6) along with the summary statistics of Table I indicate that these conservative congressmen likely find that it is more in their interest to vote for the lower expenditure ceiling of 329 as a means of weakening their electoral challengers than do the liberal congressmen. The fact that all the congressmen in this subsample have a value of PROBABILITY CHANGE UNDER 329 of zero does not necessarily mean that they all have equal stakes in the passage of amendment 329.

The estimated regressions (5) through (12) are consistent with the hypothesis that a congressman is a maximizer of the expected value of his office with respect to his voting behavior. A congressman is willing to ignore ideology when the opportunity cost of not doing so is sufficiently high. Does this mean that congressmen shirk?

The answer depends upon how narrowly or broadly one defines shirking. If one narrowly defines shirking as a congressman's voting according to his own ideology rather than according to his constituents', then regressions (5) through (12) provide no evidence that congressmen shirk. Comparison of the VOTE ON 329 and VOTE ON 333 logit regressions indicates that when the cost of voting ideologically is sufficiently high, as in the case of amendment 329, ideology will be ignored. Since both amendments were unambiguously liberal ideological issues, the sharp contrasts between the voting regressions for 329 and 333, run for virtually identical samples of congressmen, suggest that both personal and constituent ideology probably were ignored in the 329 voting. Indeed, this interpretation must hold if the ideologies of constituents and their representatives are identical. The fortuitous circumstances generating the negative coefficient of the ideology variable in the 329 regressions merely strengthen this interpretation.

If, however, one broadly defines shirking as a congressman's voting against his constituents' interest, whether economic or ideological, in order to obtain some personal benefit, then a comparison of the voting regressions for amendments 329 and 333 clearly indicates that some congressmen shirked when voting on 329. This conclusion could only be avoided if we made the counterintuitive assumption that a congressman's broader voting record, as measured by his Americans for Democratic Action rating, has at best only a weak positive correlation with his constituents' ideological preferences.

The broader definition of shirking clearly is the more appealing one. The narrow definition of shirking as the sacrificing of the constituents' ideology for the representative's ideology is, quite simply, too narrow. There is no reason to presume that the constituent-principal cares why his representative-agent has sacrificed his economic or ideological interests. All that matters is whether or not the sacrifice took place.

VI. SUMMARY AND CONCLUSIONS

This paper has investigated the influence of ideology on congressional voting within a conceptual framework built on the assumption that a congressman acts so as to maximize the expected value of his office. Voting on specific bills is a function of not only the constituents' economic and ideological interests but also the representative's ideological interests, campaign contributions from special interest groups, and, in general, any factors related to the voting that change either his reelection probability or the value of his office. Maximization of the expected value of office requires that voting satisfies the standard optimality conditions.

In order to explicitly focus on the influence of ideology, I selected congressional voting on two floor amendments (one proposing a lower election campaign expenditure ceiling and the other proposing a partial public funding of congressional campaigns) to the House Administration Committee's 1974 Federal Election Campaign Act bill for analysis. Both amendments represented liberal ideological issues, and their only impacts were the generation of congressman-specific benefits or costs in the form of increases or decreases in reelection probabilities. Voting on each amendment was therefore influenced solely by the congressman-specific changes in reelection probability and by ideology. Since the expenditure ceiling amendment generated nontrivial changes in reelection probabilities and the partial public funding amendment generated generally negligible changes, a comparative analysis of the congressional voting on the two amendments allowed the influence of ideology on the voting to be isolated.

Logit regressions of congressional voting on the two amendments strongly indicate that ideology is ignored when the opportunity cost, in terms of changes in reelection probability, of not ignoring ideology is sufficiently high. The sharp contrast between the regressions suggests that both constituent ideology and congressman-specific ideology were ignored in the voting on the expenditure ceiling amendment. These regressions are consistent with the hypothesis that a congressman acts so as to maximize the expected value of his office when casting his votes. If congressional shirking is broadly defined, then the voting on the campaign expenditure ceiling amendment is consistent with shirking.

REFERENCES

Abrams, Burton A. and Russell F. Settle. "The Economic Theory of Regulation and Public Financing of Presidential Elections." Journal of Political Economy, April 1978, 245-57.

Barone, Michael, Grant Ujifusa, and Douglas Matthews. The Almanac of American Politics. New York: Dutton, 1974 and 1978.

Bender, Bruce. "An Analysis of Congressional Voting on Legislation Limiting Congressional Campaign Expenditures." Journal of Political Economy, October 1988, 1005-21.

Dougan, William R. and Michael C. Munger. "The Rationality of Ideology." Journal of Law and Economics, April 1989, 119-42.

Kalt, Joseph P. and Mark A. Zupan. "Capture and Ideology in the Economic Theory of Politics." American Economic Review, June 1984, 279-300.

Kau, James B. and Paul H. Rubin. "Self-Interest, Ideology, and Logrolling in Congressional Voting." Journal of Law and Economics, October 1979, 365-84.

Lott, John R. "Political Cheating." Public Choice 52(2), 1987, 169-86.

McArthur, John and Stephen V. Marks. "Constituent Interest vs. Legislator Ideology: The Role of Political Opportunity Cost." Economic Inquiry, July 1988, 461-70.

Nelson, Douglas, and Eugene Silberberg. "Ideology and Legislator Shirking." Economic Inquiry, January 1987, 15-25.

Peltzman, Sam. "Constituent Interest and Congressional Voting." Journal of Law and Economics, April 1984, 181-210.

(1.) Both Peltzman [1984] and Dougan and Munger [1989] recognize this.

(2.) A representative who votes according to the desires of a special interest group in exchange for campaign contributions from that group is typically referred to as being "captured" by that group. However, if the desires of the special interest group differ from those of the congressman's constituents, then whether the representative was captured or is shirking is a semantic distinction from the viewpoint of his constituents.

(3.) Congressional Quarterly Weekly Report assigns numbers to all votes on bills, amendments, and motions. Although a vote of the full House is not necessary to bring a bill out of committee to the floor, for notational convenience and consistency the House Administration Committee's bill HR 16090 will be referred to by the Congressional Quarterly Weekly Report number, 325, that was assigned to the vote on the motion restricting the scope of floor amendments to the bill.

(4.) The ceiling of $93,750 represents a ceiling of $75,000 for campaign expenditures plus an additional 25 percent expenditure ($18,750) for fund-raising expenses.

(5.) The $75,000 ceiling represents $60,000 for campaign expenditures plus an additional 25 percent ($15,000) for fund-raising expenses.

(6.) One might argue that constituents in general would be harmed by the public funding of congressional election campaigns in their role as taxpayers. However, the relatively small amount of funds required would have resulted in a trivial cost per taxpayer, and the method of financing would have been a voluntary check-off on the federal income tax form. Furthermore, to whatever small degree constituents in general would be harmed financially, there is no reason whatsoever to believe that the constituents in any one congressional district would be harmed more or less than the constituents in any other district.

(7.) First, it is not clear whether individual special interest groups will be benefited or harmed by the amendments since many special interest groups compete against each other to influence congressional actions. Furthermore, even if we suppose that all special interest groups are either benefited or harmed by the amendments, the existence of a large number of such groups would likely result in a formidable free-rider problem. Of course it is not impossible that mechanisms could be developed to at least mitigate the free-rider problem.

(8.) It may not at first be completely clear that both the campaign expenditure ceiling and partial public financing of campaigns were liberal issues. While both reduce the role of money in politics, both also go against freedom of speech and the spirit of the first amendment. The ceiling limits a candidate's ability to make his positions known, and public financing requires the public to subsidize the speech of selected individuals.

However, it is not at all obvious that a liberal, as so labeled in American politics, can be accurately characterized on the basis of adherence to a consistent philosophy regarding individual freedom. A central tenet of liberal ideology is that government intervention in the market, economic or political, is desirable if it promotes the public interest. In this light, an expenditure ceiling and partial public financing can be viewed as promoting the public interest by preventing public office from being "bought." Anecdotal evidence from the 1970s indicates that the liberal position was in favor of an expenditure ceiling and public financing on precisely these public interest grounds of limiting the role of money in politics. In their description of the public interest argument for an expenditure ceiling and public financing Abrams and Settle [1978, 246-47] quote the liberal politicians John Anderson and Walter Mondale making just such an argument. Also, on the day before the U.S. House passed the 1971 Federal Election Campaign Act, which provided for a ceiling on media expenditures by presidential and congressional candidates, the New York Times (29 November 1971, p. 38) editorial page bemoaned the evils of money in politics and supported both public financing and comprehensive expenditure limits.

(9.) Bender [1988] provides a much more complete discussion of the vote-share regression equation (1).

(10.) The data sources for these variables are Barone, Ujifusa, and Matthews [1974; 1978].

(11.) This data set is the same one used by Bender [1988]. The reasons for limiting the sample size to seventy-two observations are presented there. Also, the estimated vote-share regression (2) and the estimated logit regression (9) below appear in Bender [1988].

(12.) Of the seventy-two observations, four congressmen had ratings of 100. These were reduced to 99 in order to allow the log-of-the-odds computation.

(13.) One problem with calculating the amount of public funding a candidate would receive under 333 is that the amount is also a function of campaign contributions of $50 or less. Unfortunately, the finest breakdown of campaign contributions published by the Federal Election Commission is the category $100 or less. (Source: FEC Disclosure Series No. 9: 1976 House of Representatives Campaigns Receipts and Expenditures. Washington, D.C.: Federal Election Commission, 1977.) Therefore public funding under 333 was calculated under three different assumptions: all contributions in the $100 or less category were contributions of $50 or less; one-half were contributions of $50 or less; and one-quarter were contributions of $50 or less. The differences in impacts on probabilities of reelection under the three assumptions were negligible. On the premise that contributors would easily find ways of turning a $100 contribution into $50 contributions by two different people, I calculated probabilities of reelection under 333 on the assumption that all contributions in the $100 or less category were contributions of $50 or less.

(14.) This follows directly from equation (4) and the assumption that [(INCUMBENT VOTESHARE).sub.i] is normally distributed.

(15.) The data source for the construction of VOTE ON 329 and VOTE ON 333 is Congressional Quarterly Weekly Report (17 August 1974).

(16.) Examination of the voting on amendment 329 by those eleven congressmen whose implied reelection probability increased sizably (mean increase of .0796) under 329 provides additional evidence supporting the conclusion that ideology was ignored when the opportunity cost of not ignoring ideology was sufficiently high. Of these eleven congressmen, nine voted in favor of 329. The mean untransformed ADA rating of these eleven congressmen was 52.5 (standard deviation of 32.7) as compared to the mean rating of 47.7 (standard deviation of 29.6) for the full sample of seventy-one congressmen. Of the full sample, thirty-six of the seventy-one voted in favor of 329.

(17.) Furthermore, at the time of the vote on amendment 329 in August 1974, the Watergate scandal was dominating the political scene. Republican congressmen were indeed worried about facing serious electoral competition, and Republican congressmen tend to be on average more conservative than Democratic congressmen. For this subsample of forty-two congressmen, the twelve Republicans had a mean untransformed ADA rating of 34.2 (standard deviation of 25.6) and the thirty Democrats had a mean rating of 54.6 (standard deviation of 25.7).

BRUCE BENDER, Associate Professor, School of Business Administration, University of Wisconsin-Milwaukee. I benefited from comments on previous drafts by Bill Dougan, George Tolley, participants of the Economics and Policy Workshop of the University of Chicago, and two anonymous referees.

The question of what determines congressional voting on legislation has inspired a number of recent empirical analyses. All are in agreement that congressional voting on specific bills is correlated with the economic self-interest of the representative's constituents and a measure of the representative's ideology as indicated by his broader voting record. However, the interpretation of the empirical results regarding the influence of ideology has generated more debate than consensus in the current literature.

Kau and Rubin [1979] find evidence that a congressman's ideology influences his voting, but also find evidence for an alternative, but not mutually exclusive, hypothesis that a congressman's rating by ideological groups reflects membership in a logrolling coalition. Kalt and Zupan [1984] argue that although a representative's votes are likely to reflect the economic interests and the ideology of his constituents, the weak incentive for the constituent (principal) to monitor the voting behavior of his representative (agent) gives the representative the opportunity to "shirk" or vote according to his personal ideology rather than his constituents'. Both Nelson and Silberberg [1987] and McArthur and Marks [1988] find that the degree of shirking increases as the opportunity cost of shirking decreases.

In contrast to the above findings, Peltzman [1984] argues that ideological voting by congressmen need not necessarily reflect shirking, but instead may be a way for the representative to cheaply provide his constituents with information about his general position on issues. What appears to be shirking allowed by an imperfect principal-agent relationship may instead be signaling designed to at least partially overcome the imperfect nature of the relationship. Lott [1987] extends Peltzman's signaling argument by hypothesizing that if politicians are viewed by constituents as "search" goods rather than "experience" goods, then they will have an incentive to establish a "political brand name" by establishing a consistent ideological voting record. Dougan and Munger's [1989] approach is similar to Lott's. They argue that politicians with strongly held views that match those of their constituents make ideal representatives because they are less likely to vote against their constituents' interests when pressured by special interest groups. An incumbent therefore has an incentive to invest in "reputational capital" by maintaining the consistency over time of his broader voting record because this reputational capital generates quasi-rents in electoral competition with candidates who possess less of it. Neither Lott nor Dougan and Munger find evidence of shirking in their analyses of congressmen's broader voting records as measured by ideological groups' ratings.

Dougan and Munger also develop a "variable intensity of commitment" version of the reputational capital hypothesis. This version contends that voters with unambiguously liberal or conservative ideologies may prefer a candidate whose voting record indicates that his views are more extreme than their own. Such a zealous candidate, voters believe, is even less likely to be influenced by special interest groups. A congressman serving an unambiguously liberal or conservative constituency therefore has an incentive to establish and maintain an extremely ideological overall voting record. In contrast, a congressman serving a constituency with centrist ideological preferences has an incentive to establish an overall voting record that matches his constituency's ideological leanings. The systematic deviation of a congressman's voting record from his constituents' preferences, when near either extreme of the ideological spectrum, is not inconsistent with a faithful principal-agent relationship.

One cannot help but be struck by the observation that those researchers finding evidence consistent with shirking studied voting on specific legislation, whereas those researchers finding no evidence of shirking studied representatives' broader voting records as measured by ideological ratings. When analyzing voting on a particular bill, how does a researcher determine whether his empirical findings reflect shirking or reputational capital? After all, reputational capital is established by consistent ideological voting on a great number of bills. This does not, however, rule out the possibility that shirking did occur for voting on any particular bill. (1) This paper investigates more completely the influence of ideology on congressional voting. It presents a more general conceptual framework for analyzing voting than has been utilized heretofore. A representative is assumed to vote on legislation so as to maximize the expected value of his office. Constituent ideology and representative ideology, as well as the direct economic interests of constituents and representatives, influence voting by inducing the representative to make the standard trade-offs at the margin. The empirical work takes advantage of two congressional votes to conduct a natural experiment on the influence of ideology on voting. Analyses of voting on two floor amendments to the U.S. House Administration Committee's 1974 Federal Election Campaign Act bill indicate that ideological considerations, whether of congressmen or their constituents, are ignored when the direct self-interest of the congressman to do so is sufficiently large.

II. CONCEPTUAL FRAMEWORK

Assume that a congressman's objective is to maximize the expected value of his office, which is equal to the product of the value of his office and his probability of reelection. Let the value of the office to a congressman be an increasing function of the total compensation of the office and the exercise of his personal ideology through his voting. The inclusion of the congressman's personal ideology implicitly assumes that he receives utility from attempting to influence legislation in what he deems to be the "correct" direction.

Assume that probability of reelection varies directly with the degree to which a congressman's voting satisfies the economic interests and ideological preferences of his constituents, as well as directly with his election campaign expenditures. Larger campaign expenditure presumably enables a congressman to more effectively present his record to the voters in a favorable light. Other factors that likely influence probability of reelection are the fraction of voters registered in the incumbent's party and the incumbent's party affiliation. The former indicates the fraction of the electorate likely to hold the same position on the issues that they expect the incumbent to hold. The latter reflects any possible national swings of preference toward one party or the other. These last two variables are parameters.

If a congressman's personal ideology does not coincide with the ideology and economic interests of his constituents, then voting on legislation according to his personal ideology and against his constituent's interests and ideology will lower his probability of reelection. However, the failure to act in accordance with his personal ideology will lower the value of holding office. A congressman will vote so as to satisfy the standard marginal conditions for maximizing the expected value of his office. This is consistent with the argument by Dougan and Munger [1989] that a congressman whose personal ideology is identical to his constituents' ideology is in a stronger long-run competitive position than a congressman whose ideology is different than his constituents', because the former can satisfy his constituents at a lower cost.

Campaign contributions from special interest groups also may influence a congressman's voting. To the extent that these groups' objectives conflict with the economic or ideological interests of his constituents or with his own ideological interests, then voting to accommodate these groups will lower the expected value of the congressman's office by either decreasing his reelection probability or decreasing the value of his office. However, the increased campaign expenditure made possible by additional special interest contributions will increase his reelection probability.

I define shirking by a congressman as his failure to vote according to the economic interests or ideology of his constituents. Shirking can be induced by either the offer of contributions from special interest groups (2) or a divergence between the ideologies of the congressman and his constituents. More generally, any factor that can marginally increase the expected value of a congressman's office while influencing his vote can potentially induce that congressman to shirk. Indeed, a congressman may be induced to vote not only against the ideology of his constituents but against his own ideology as well. Shirking therefore can occur even if the congressman's and constituents' ideologies are identical.

For my empirical analysis of the influence of ideology on congressional voting, I elected to analyze the voting on two floor amendments to the U.S. House Administration Committee's 1974 Federal Election Campaign Act bill. For these two amendments, only the direct congressman-specific benefits and costs generated by each amendment and ideology should affect the voting. Since the amendments differ greatly in terms of the magnitude of their direct benefits and costs, analyses of voting on the amendments is equivalent to conducting a natural experiment to shed light on the influence of ideology on congressional voting.

III. BACKGROUND AND DESCRIPTION OF THE FLOOR AMENDMENTS

The U.S. Congress passed the 1974 Federal Election Campaign Act in October of that year. In brief, the 1974 campaign act provided for partial public funding of presidential election campaigns, ceilings on individual campaign contributions, and ceilings on campaign expenditures by candidates in presidential, Senate, and House elections starting in 1976.

On 7 August 1974, the House Administration Committee's bill was brought to the floor of the House for consideration under the rules of an adopted motion (Congressional Quarterly notation 325) that restricted the scope of potential floor amendments. (For notational convenience and consistency, the House Administration Committee's bill (HR 16090) will be referred to as bill 325.) (3) As in the eventual 1974 campaign act, bill 325 provided for partial public financing of presidential campaigns and ceilings on campaign contributions by individuals and campaign expenditures by candidates for federal office starting in 1976. The campaign expenditure ceiling for U.S. House candidates was $93,750. (4) No provision for any type of public funding of congressional campaigns appeared in 325.

On 8 August 1974 a floor amendment (329) that lowered the expenditure ceiling for House candidates from $93,750 under bill 325 to $75,000 (5) was adopted by a vote of 240-175 (130-102 by Democrats and 110-73 by Republicans). Later that day a floor amendment (333) that provided for partial public funding of congressional campaigns was defeated by a vote of 187-228 (114-118 by Democrats and 73-110 by Republicans). Specifically, amendment 333 would have granted matching federal funds to congressional candidates for all individual contributions of $50 or less up to one-third of the expenditure ceiling. To qualify for these matching funds, the candidate would have to raise 10 percent of his maximum allowable matching funds in contributions of $50 or less.

Since campaign expenditures produce electoral support, both 329 and 333 could have the effect of changing each congressman's probability of reelection. This direct, congressman-specific benefit or cost is the only impact of these amendments. The direct economic interests of the congressman's constituents are not affected, (6) and it is unlikely that there would be an incentive for economic special interest groups to influence a congressman's votes on these amendments. (7) However, passage of each amendment was an important liberal cause. (8) Consequently, congressional voting on both amendments should have been determined solely by the congressman-specific direct benefits or costs and by ideology.

IV. CALCULATION OF CHANGES IN REELECTION PROBABILITIES

In order to calculate changes in reelection probabilities under each amendment, it is necessary to estimate the impact of changes in campaign expenditures on reelection probabilities. This requires an estimated vote-share regression equation having campaign expenditures as independent variables and the ability to calculate the changes in campaign expenditures under each amendment. Calculating such changes is possible only if we know what the unconstrained expenditures would have been in the 1976 elections. In January 1976, the Supreme Court fortuitously declared that the section (608[c]) of the 1974 campaign act setting expenditure limits for congressional election campaigns was unconstitutional (Buckley v. Valeo [46 LEd 2d 659]). Therefore, passage of 329 did not affect campaign expenditures in the 1976 congressional elections. Obviously, neither did 333, since it was defeated.

The vote-share regression equation to be estimated is the one used in Bender [1988]. (9) That regression equation is

(1) INCUMBENT VOTE SHARE = [b.sub.0] + [b.sub.1] (INCUMBENT PARTY STRENGTH) + [b.sub.2] (CHALLENGER EXPENDITURE) + [b.sub.3] (RELATIVE CHALLENGER EXPENDITURE) + [b.sub.4] (DEMOCRATIC INCUMBENT) + e

where INCUMBENT VOTE SHARE is the fraction of total votes cast for the Democratic and Republican candidates in the congressional election received by the incumbent, DEMOCRATIC INCUMBENT is a dummy variable for the incumbent being a Democrat, INCUMBENT PARTY STRENGTH is the fraction of registered Democratic and Republican voters in the congressional district registered in the incumbent's party, CHALLENGER EXPENDITURE is the challenger's campaign expenditure in thousands of dollars, RELATIVE CHALLENGER EXPENDITURE is the challenger's campaign expenditure as a fraction of the sum of expenditures by the challenger and the incumbent, (10) and e is the error term. The inclusion of CHALLENGER EXPENDITURE and RELATIVE CHALLENGER EXPENDITURE presupposes that both the level of spending by the challenger and relative spending by the challenger influence vote share. This specification allows for the marginal products of challenger spending and incumbent spending in the production of vote share each to be a function of the levels of both challenger and incumbent spending. It can be shown that [b.sub.2] < 0 and [b.sub.3] < 0 is a sufficient condition for the incumbent's marginal product to be positive but decreasing with the incumbent's expenditure, and for the challenger's marginal product to be negative but increasing with the challenger's expenditure.

Regression equation (1) was estimated for a sample of seventy-two major party U.S. House of Representatives elections involving incumbents in 1976. (11) The ordinary least squares estimate of (1) is presented below (t-statistics in parentheses):

(2) INCUMBENT VOTE SHARE = 0.6250 + 0.1915 (INCUMBENT PARTY (20.30) (4.15) STRENGTH) - 0.0014 (CHALLENGER (5.36) EXPENDITURE) - 0.0833 (RELATIVE CHALLENGER (1.71) EXPENDITURE

[R.sup.2] = .62, n = 72

Estimated coefficients are m accordance with expectations. The incumbent's vote share varies directly with the strength of the incumbent's party in the district and inversely with the challenger's level of spending and relative spending. The positive constant term can be attributed to incumbency. DEMOCRATIC INCUMBENT had an insignificant estimated coefficient (t-statistic of 0.19) and was therefore dropped from the regression. This indicates that the mood of the national electorate favored neither party in 1976. All fitted values are in the interval [.456,.799], and the standard error of the regression is only 0.065. Finally, at the mean values of incumbent and challenger spending (72.607 and 32.207), challenger spending has a marginal product (-.00193) that is 7.9 times larger in absolute value than the marginal product (.00024) of incumbent spending.

I attempted to measure the impact on the incumbent's vote share, and consequently his probability of reelection, of the divergence of his voting record, as measured by his 1974 rating by Americans for Democratic Action, from his constituents' desired value of that rating. Estimation of this divergence followed the procedure of Kalt and Zupan [1984]. Specifically, I regressed a log-of-the-odds transformation of the congressman's ideological rating (ADA) (12) on the fraction of the congressional district that voted for George McGovern in the 1972 presidential election (MCGOVERN VOTESHARE) and DEMOCRATIC INCUMBENT. The ordinary least squares estimate of the regression is

(3) ADA = -4.4002 + 9.8607 (MCGOVERN (7.88) (6.27) VOTESHARE)

+ 0.9803 (DEMOCRATIC INCUMBENT) (2.86)

[R.sup.2] = .54, n = 72

The fitted value of ADA for each congressman can be interpreted as his constituents' desired value. The absolute value of the residual therefore can be interpreted as the divergence between the congressman's ideological voting record as measured by ADA and the ideological voting record that his constituents desire him to have.

The absolute value of the residual of (3) was added as an independent variable in regression (2), which was then reestimated. Since shirking is expected to reduce vote share, the estimated coefficient is expected to be negative. It was in fact insignificantly positive (t-statistic of 0.95). There are two, not necessarily mutually exclusive, possible explanations for this result. One is that the electoral advantages of incumbency make possible some degree of shirking and that this potential freedom to shirk varies directly with the electoral strength of the incumbent. For example, voters may be more willing to tolerate ideological shirking by an incumbent because of his greater ability to satisfy the economic interests of his constituents in ways other than simply voting (e.g., by obtaining pork barrel benefits). If this should be the case, then the positive, although insignificant, coefficient of the absolute value of the ideology residual may reflect a simultaneity problem. A second possible explanation is that the residuals of congressmen with unambiguously liberal or conservative constituencies may reflect their investments in reputational capital. If so, then these incumbents will not be punished at the polls, and consequently the coefficient of the absolute value of the residual is less likely to be negative or significant.

I now present the methodology used to infer reelection probabilities. Consider the incumbent's actual vote-share outcome in 1976 to be a random draw from an assumed normal probability distribution of outcomes generated by the estimated vote-share regression (2). Let this distribution for each incumbent be described by an expected value equal to the fitted value (INCUMBENT VOTE SHARE *) and by a standard deviation equal to the standard error of the regression (STANDARD ERROR). Probability of reelection for the i-th incumbent can be inferred from

(4) [Z.sub.i] = (INCUMBENT VOTE [SHARE.sup.*.sub.i] -.5) /(STANDARD ERROR)

where [Z.sub.i] is the number of standard deviations that the i-th incumbent's expected vote share lies from the minimum vote share of .5 necessary to win reelection. Given the estimated coefficients of (2), fitted values of INCUMBENT VOTE SHARE, and therefore probabilities of reelection, can be simulated for the expenditure ceiling of $93,750 in the House Administration Committee's bill (325), for the expenditure ceiling of $75,000 under floor amendment 329, and for the partial public financing provisions of floor amendment 333. It should be kept in mind that the maximum allowable public funding of campaigns under 333, as described above, is a function of the expenditure ceiling. Since 329 was passed prior to the consideration of 333, the $75,000 ceiling under 329 is the relevant ceiling to be used for the calculation of public funding under 333. (13)

The direct benefit or cost to the individual congressman of each amendment is the increase or decrease in reelection probability faced by that congressman. The change in reelection probability under amendment 329 is the change in reelection probability caused by lowering the expenditure ceiling from $93,750 under 325 to $75,000 under 329 and is denoted by PROBABILITY CHANGE UNDER 329. The change in reelection probability under 333 is the change in reelection probability caused by the partial public funding provisions of 333 in the presence of the $75,000 ceiling under 329 and is denoted by PROBABILITY CHANGE UNDER 333.

Summary statistics for these changes in reelection probabilities are presented in Tables I and II. Since there was one abstention for the vote on 329 and 333, the full sample size for each vote is seventy-one observations, of which the votes on the two amendments have seventy in common. Although the full sample summary statistics in Tables I and II appear to suggest that 329 and 333 had similar trivial effects on reelection probabilities, this is not in fact the case. While the public financing provisions of 333 did little to help or hurt congressmen, the lower expenditure ceiling of 329 considerably helped a number of congressmen. Amendment 329 increased the reelection probabilities of eleven of the congressmen by an average of .0796 or almost 8 percentage points. Also note that the relatively small standard deviation of PROBABILITY CHANGE UNDER 329 of .0263 for these congressmen indicates that the mean value of .0796 was not driven just by a couple of outliers with large increases in reelection probabilities.

TABLE I Summary Statistics for Changes in Reelection Probabilities Generated by a Lowering of the Campaign Expenditure Ceiling From $93,750 to $75,000 Under Amendment 329 Standard Sample Mean Deviation Observations Full Sample .0112 .0313 71 (PROBABILITY CHANGE UNDER .0275 .0445 29 329) [not equal to] 0 (PROBABILITY CHANGE UNDER .0796 .0263 11 329) > 0 (PROBABILITY CHANGE UNDER -.0044 .0051 18 329) < 0 TABLE II Summary Statistics for Changes in Reelection Probabilities Generated by the Partial Public Financing Provisions of Amendment 333 in the Presence of the Campaign Expenditure Ceiling of Amendment 329 Standard Sample Mean Deviation Observations Full Sample -.0133 .0350 71 (PROBABILITY CHANGE UNDER -.0220 .0429 43 333) [not equal to] 0 (PROBABILITY CHANGE UNDER .0041 .0051 9 333) > 0 (PROBABILITY CHANGE UNDER -.0289 .0459 34 333) < 0

That amendment 329 would have a greater impact on reelection probabilities than 333 is not surprising. By design the 329 expenditure ceiling would be binding only in elections in which one or both candidates would otherwise spend more than the ceiling. Such elections are typically the closer elections (i.e., expected vote shares closer to .5). A given change in expected vote share induced by limiting expenditures will tend to cause a larger change in reelection probability the closer the expected vote share is to .5. (14) This, along with the challenger's marginal product of expenditure in the production of vote share substantially exceeding the incumbent's, explains why those congressmen who benefited from an increase in reelection probability under 329 did so by an average of 8 percentage points. In contrast to 329, the matching public funds provisions of 333 offer no matching funds for a candidate already at the expenditure ceiling under 329. Therefore, 333 has a trivial impact on reelection probability, because it provides only limited public funds to candidates in elections in which the expected vote share is not likely to be close to .5.

V. ANALYSIS OF VOTING ON THE AMENDMENTS

Congressional voting on each amendment should be a function of only the implied change in a congressman's reelection probability and ideology. Define VOTE ON 329 as a qualitative variable taking the value 1 if a congressman voted in favor of amendment 329 and the value 0 if he voted against 329. Analogously define VOTE ON 333 as the qualitative variable for congressional voting on amendment 333. (15) I then estimate a logit regression of the qualitative voting variable on the relevant change in reelection probability and the previously defined ideology variable ADA. The estimated logit regressions for congressional voting on 329 and 333 for the full sample of congressmen are presented below (t-statistics in parentheses):

(5) VOTE ON 329 = - 0.1710 + 22.0598 (PROBABILITY (0.67) (2.04) CHANGE UNDER 329)

[chi square] = 5.86, n = 71

(6) VOTE ON 329 = - 0.1973 + 25.1159 (PROBABILITY (0.76) (2.13) CHANGE UNDER 329)

- 0.1999 (ADA) (1.39)

[chi square] = 7.88, n=71

(7) VOTE ON 333 = 0.3983 + 6.3601 (PROBABILITY (1.54) (0.90) CHANGE UNDER 333)

[chi square] = 0.84, n = 71

(8) VOTE ON 333 = 0.6204 + 4.3984 (PROBABILITY (1.94) (0.53) CHANGE UNDER 333) + 0.8032 (ADA) (3.68)

[chi square] = 21.80, n = 71

The contrasts between the estimates of regressions (5) and (6) and of regressions (7) and (8) are quite sharp. Voting on amendment 329 is significantly, positively related to the implied change in reelection probability under 329; whereas voting on amendment 333 is not at all significantly related to the implied change in reelection probability under 333. On the other hand, voting on 333 is significantly, positively related to ADA; whereas voting on 329 is at best marginally negatively related to ADA. In short, voting on 329 appears driven by the direct personal benefits or costs in terms of implied changes in reelection probability, and voting on 333 appears driven by ideological considerations.

As noted in the previous section, changes in implied reelection probability under 329 were nontrivial for a considerable number of congressmen, while changes in implied reelection probability under 333 were generally negligible. Since both the expenditure ceiling and the partial public funding of campaigns were liberal issues, then a reasonable interpretation of the regression estimates is that ideological considerations were ignored when the opportunity cost of ideological voting was sufficiently high. Furthermore, with the defeat (by a 187-228 vote) of 333, there was no other opportunity for voting on partial public funding of campaigns. In contrast, a congressman who voted against 329 was aware that he could always be on record as voting for an expenditure ceiling by voting in favor of the final House bill (passed by a vote of 355-48), which would necessarily contain the expenditure ceiling provision of either 329 or 325. There was, therefore, an additional incentive for congressmen to pay less attention to ideological concerns when voting on 329 than on 333.

This interpretation of regressions (5) through (8) becomes even more credible in the light of regressions (9) through (12). These are the same as (5) through (8) except that the VOTE ON 329 regressions are run on a subsample consisting of the twenty-nine congressmen who had a non-zero change in reelection probability under 329 and the VOTE ON 333 regressions are run on the subsample of those forty-three congressmen who had a non-zero change in reelection probability under 333. The estimated regressions are

(9) VOTE ON 329 = - 0.5416 + 25.9934 (PROBABILITY (1.16) (2.24) CHANGE UNDER 329)

[chi square] =6.80, n=29

(10) VOTE ON 329 = - 0.5850 + 28.1071 (PROBABILITY (1.24) (2.24) CHANGE UNDER 329) - 0.1321 (ADA) (0.64)

[chi square] = 7.22, n = 29

(11) VOTE ON 333 = 0.3679 + 6.0622 (PROBABILITY (1.05) (0.82) CHANGE UNDER 333)

[chi square] = 0.69, n = 43

(12) VOTE ON 333 = 1.0227 + 7.5560 (PROBABILITY (2.11) (0.80) CHANGE UNDER 333)

+ 0.9747 (ADA) (3.12)

[chi square] = 15.37, n = 43

The differences between the full sample VOTE ON 333 regressions (7) and (8) and the subsample VOTE ON 333 regressions (11) and (12) are slight indeed. This is to be expected if only generally negligible changes in implied reelection probabilities are generated under 333. Ideology is clearly the dominant influence on the amendment 333 voting.

In contrast, there are noticeable differences between the full sample VOTE ON 329 regressions (5) and (6) and the subsample VOTE ON 329 regressions (9) and (10). First, the magnitudes and t-statistics of the PROBABILITY CHANGE UNDER 329 coefficients are somewhat larger in (9) and (10). Second, and more strikingly, the magnitude (absolute value) and t-statistic of the coefficient of ADA are considerably smaller in (10) than in (6). Both of these changes are to be expected if nontrivial changes in implied reelection probabilities are generated under 329. The opportunity cost of voting ideologically is higher for the affected subsample than for the full sample. (16)

Let us look more closely at the coefficient of ADA in (6) and (10). Both the magnitude (absolute value) and t-statistic are larger in the full sample regression (6) than in the subsample regression (10). More importantly, the sign of the coefficient is, surprisingly, negative. This appears to imply that the more liberal (conservative) congressmen are more willing to vote against (for) a liberal issue. In fact it does not.

The insignificant coefficient of ADA (t-statistic of 0.64) in (10) indicates that ideological considerations did not affect the voting of those congressmen whose reelection probability would have changed. But why does the magnitude (absolute value) of the coefficient of ADA increase, and why does it now become significant at the .10 level on a one-tailed test (t-statistic of 1.39) in the full sample regression (6)? One would think that the addition to the sample of those forty-two congressmen unaffected by amendment 329 would result in voting more consistent with ideology and a consequent move towards a positive or less negative coefficient of ADA. An answer is that those congressmen unaffected by 329 in 1976 appear to be considering the impact that 329 may have on their future reelection probabilities. For the subsample of those forty-two congressmen receiving neither an increase nor a decrease in reelection probability, the correlation coefficients of ADA and probability of reelection under the expenditure ceiling imposed by 325, challenger expenditure, and incumbent expenditure are .32, -.23, and -.44. Noting that the measure of conservative ideology varies inversely with ADA, these correlation coefficients indicate that the relatively conservative congressmen tend to have a lower probability of reelection under 325, tend to face higher levels of campaign expenditures by their challengers, and tend to have higher levels of campaign expenditures than do the relatively liberal congressmen. This indicates that the relatively conservative congressmen in this subsample are facing tougher electoral challenges. (17) Regressions (5) and (6) along with the summary statistics of Table I indicate that these conservative congressmen likely find that it is more in their interest to vote for the lower expenditure ceiling of 329 as a means of weakening their electoral challengers than do the liberal congressmen. The fact that all the congressmen in this subsample have a value of PROBABILITY CHANGE UNDER 329 of zero does not necessarily mean that they all have equal stakes in the passage of amendment 329.

The estimated regressions (5) through (12) are consistent with the hypothesis that a congressman is a maximizer of the expected value of his office with respect to his voting behavior. A congressman is willing to ignore ideology when the opportunity cost of not doing so is sufficiently high. Does this mean that congressmen shirk?

The answer depends upon how narrowly or broadly one defines shirking. If one narrowly defines shirking as a congressman's voting according to his own ideology rather than according to his constituents', then regressions (5) through (12) provide no evidence that congressmen shirk. Comparison of the VOTE ON 329 and VOTE ON 333 logit regressions indicates that when the cost of voting ideologically is sufficiently high, as in the case of amendment 329, ideology will be ignored. Since both amendments were unambiguously liberal ideological issues, the sharp contrasts between the voting regressions for 329 and 333, run for virtually identical samples of congressmen, suggest that both personal and constituent ideology probably were ignored in the 329 voting. Indeed, this interpretation must hold if the ideologies of constituents and their representatives are identical. The fortuitous circumstances generating the negative coefficient of the ideology variable in the 329 regressions merely strengthen this interpretation.

If, however, one broadly defines shirking as a congressman's voting against his constituents' interest, whether economic or ideological, in order to obtain some personal benefit, then a comparison of the voting regressions for amendments 329 and 333 clearly indicates that some congressmen shirked when voting on 329. This conclusion could only be avoided if we made the counterintuitive assumption that a congressman's broader voting record, as measured by his Americans for Democratic Action rating, has at best only a weak positive correlation with his constituents' ideological preferences.

The broader definition of shirking clearly is the more appealing one. The narrow definition of shirking as the sacrificing of the constituents' ideology for the representative's ideology is, quite simply, too narrow. There is no reason to presume that the constituent-principal cares why his representative-agent has sacrificed his economic or ideological interests. All that matters is whether or not the sacrifice took place.

VI. SUMMARY AND CONCLUSIONS

This paper has investigated the influence of ideology on congressional voting within a conceptual framework built on the assumption that a congressman acts so as to maximize the expected value of his office. Voting on specific bills is a function of not only the constituents' economic and ideological interests but also the representative's ideological interests, campaign contributions from special interest groups, and, in general, any factors related to the voting that change either his reelection probability or the value of his office. Maximization of the expected value of office requires that voting satisfies the standard optimality conditions.

In order to explicitly focus on the influence of ideology, I selected congressional voting on two floor amendments (one proposing a lower election campaign expenditure ceiling and the other proposing a partial public funding of congressional campaigns) to the House Administration Committee's 1974 Federal Election Campaign Act bill for analysis. Both amendments represented liberal ideological issues, and their only impacts were the generation of congressman-specific benefits or costs in the form of increases or decreases in reelection probabilities. Voting on each amendment was therefore influenced solely by the congressman-specific changes in reelection probability and by ideology. Since the expenditure ceiling amendment generated nontrivial changes in reelection probabilities and the partial public funding amendment generated generally negligible changes, a comparative analysis of the congressional voting on the two amendments allowed the influence of ideology on the voting to be isolated.

Logit regressions of congressional voting on the two amendments strongly indicate that ideology is ignored when the opportunity cost, in terms of changes in reelection probability, of not ignoring ideology is sufficiently high. The sharp contrast between the regressions suggests that both constituent ideology and congressman-specific ideology were ignored in the voting on the expenditure ceiling amendment. These regressions are consistent with the hypothesis that a congressman acts so as to maximize the expected value of his office when casting his votes. If congressional shirking is broadly defined, then the voting on the campaign expenditure ceiling amendment is consistent with shirking.

REFERENCES

Abrams, Burton A. and Russell F. Settle. "The Economic Theory of Regulation and Public Financing of Presidential Elections." Journal of Political Economy, April 1978, 245-57.

Barone, Michael, Grant Ujifusa, and Douglas Matthews. The Almanac of American Politics. New York: Dutton, 1974 and 1978.

Bender, Bruce. "An Analysis of Congressional Voting on Legislation Limiting Congressional Campaign Expenditures." Journal of Political Economy, October 1988, 1005-21.

Dougan, William R. and Michael C. Munger. "The Rationality of Ideology." Journal of Law and Economics, April 1989, 119-42.

Kalt, Joseph P. and Mark A. Zupan. "Capture and Ideology in the Economic Theory of Politics." American Economic Review, June 1984, 279-300.

Kau, James B. and Paul H. Rubin. "Self-Interest, Ideology, and Logrolling in Congressional Voting." Journal of Law and Economics, October 1979, 365-84.

Lott, John R. "Political Cheating." Public Choice 52(2), 1987, 169-86.

McArthur, John and Stephen V. Marks. "Constituent Interest vs. Legislator Ideology: The Role of Political Opportunity Cost." Economic Inquiry, July 1988, 461-70.

Nelson, Douglas, and Eugene Silberberg. "Ideology and Legislator Shirking." Economic Inquiry, January 1987, 15-25.

Peltzman, Sam. "Constituent Interest and Congressional Voting." Journal of Law and Economics, April 1984, 181-210.

(1.) Both Peltzman [1984] and Dougan and Munger [1989] recognize this.

(2.) A representative who votes according to the desires of a special interest group in exchange for campaign contributions from that group is typically referred to as being "captured" by that group. However, if the desires of the special interest group differ from those of the congressman's constituents, then whether the representative was captured or is shirking is a semantic distinction from the viewpoint of his constituents.

(3.) Congressional Quarterly Weekly Report assigns numbers to all votes on bills, amendments, and motions. Although a vote of the full House is not necessary to bring a bill out of committee to the floor, for notational convenience and consistency the House Administration Committee's bill HR 16090 will be referred to by the Congressional Quarterly Weekly Report number, 325, that was assigned to the vote on the motion restricting the scope of floor amendments to the bill.

(4.) The ceiling of $93,750 represents a ceiling of $75,000 for campaign expenditures plus an additional 25 percent expenditure ($18,750) for fund-raising expenses.

(5.) The $75,000 ceiling represents $60,000 for campaign expenditures plus an additional 25 percent ($15,000) for fund-raising expenses.

(6.) One might argue that constituents in general would be harmed by the public funding of congressional election campaigns in their role as taxpayers. However, the relatively small amount of funds required would have resulted in a trivial cost per taxpayer, and the method of financing would have been a voluntary check-off on the federal income tax form. Furthermore, to whatever small degree constituents in general would be harmed financially, there is no reason whatsoever to believe that the constituents in any one congressional district would be harmed more or less than the constituents in any other district.

(7.) First, it is not clear whether individual special interest groups will be benefited or harmed by the amendments since many special interest groups compete against each other to influence congressional actions. Furthermore, even if we suppose that all special interest groups are either benefited or harmed by the amendments, the existence of a large number of such groups would likely result in a formidable free-rider problem. Of course it is not impossible that mechanisms could be developed to at least mitigate the free-rider problem.

(8.) It may not at first be completely clear that both the campaign expenditure ceiling and partial public financing of campaigns were liberal issues. While both reduce the role of money in politics, both also go against freedom of speech and the spirit of the first amendment. The ceiling limits a candidate's ability to make his positions known, and public financing requires the public to subsidize the speech of selected individuals.

However, it is not at all obvious that a liberal, as so labeled in American politics, can be accurately characterized on the basis of adherence to a consistent philosophy regarding individual freedom. A central tenet of liberal ideology is that government intervention in the market, economic or political, is desirable if it promotes the public interest. In this light, an expenditure ceiling and partial public financing can be viewed as promoting the public interest by preventing public office from being "bought." Anecdotal evidence from the 1970s indicates that the liberal position was in favor of an expenditure ceiling and public financing on precisely these public interest grounds of limiting the role of money in politics. In their description of the public interest argument for an expenditure ceiling and public financing Abrams and Settle [1978, 246-47] quote the liberal politicians John Anderson and Walter Mondale making just such an argument. Also, on the day before the U.S. House passed the 1971 Federal Election Campaign Act, which provided for a ceiling on media expenditures by presidential and congressional candidates, the New York Times (29 November 1971, p. 38) editorial page bemoaned the evils of money in politics and supported both public financing and comprehensive expenditure limits.

(9.) Bender [1988] provides a much more complete discussion of the vote-share regression equation (1).

(10.) The data sources for these variables are Barone, Ujifusa, and Matthews [1974; 1978].

(11.) This data set is the same one used by Bender [1988]. The reasons for limiting the sample size to seventy-two observations are presented there. Also, the estimated vote-share regression (2) and the estimated logit regression (9) below appear in Bender [1988].

(12.) Of the seventy-two observations, four congressmen had ratings of 100. These were reduced to 99 in order to allow the log-of-the-odds computation.

(13.) One problem with calculating the amount of public funding a candidate would receive under 333 is that the amount is also a function of campaign contributions of $50 or less. Unfortunately, the finest breakdown of campaign contributions published by the Federal Election Commission is the category $100 or less. (Source: FEC Disclosure Series No. 9: 1976 House of Representatives Campaigns Receipts and Expenditures. Washington, D.C.: Federal Election Commission, 1977.) Therefore public funding under 333 was calculated under three different assumptions: all contributions in the $100 or less category were contributions of $50 or less; one-half were contributions of $50 or less; and one-quarter were contributions of $50 or less. The differences in impacts on probabilities of reelection under the three assumptions were negligible. On the premise that contributors would easily find ways of turning a $100 contribution into $50 contributions by two different people, I calculated probabilities of reelection under 333 on the assumption that all contributions in the $100 or less category were contributions of $50 or less.

(14.) This follows directly from equation (4) and the assumption that [(INCUMBENT VOTESHARE).sub.i] is normally distributed.

(15.) The data source for the construction of VOTE ON 329 and VOTE ON 333 is Congressional Quarterly Weekly Report (17 August 1974).

(16.) Examination of the voting on amendment 329 by those eleven congressmen whose implied reelection probability increased sizably (mean increase of .0796) under 329 provides additional evidence supporting the conclusion that ideology was ignored when the opportunity cost of not ignoring ideology was sufficiently high. Of these eleven congressmen, nine voted in favor of 329. The mean untransformed ADA rating of these eleven congressmen was 52.5 (standard deviation of 32.7) as compared to the mean rating of 47.7 (standard deviation of 29.6) for the full sample of seventy-one congressmen. Of the full sample, thirty-six of the seventy-one voted in favor of 329.

(17.) Furthermore, at the time of the vote on amendment 329 in August 1974, the Watergate scandal was dominating the political scene. Republican congressmen were indeed worried about facing serious electoral competition, and Republican congressmen tend to be on average more conservative than Democratic congressmen. For this subsample of forty-two congressmen, the twelve Republicans had a mean untransformed ADA rating of 34.2 (standard deviation of 25.6) and the thirty Democrats had a mean rating of 54.6 (standard deviation of 25.7).

BRUCE BENDER, Associate Professor, School of Business Administration, University of Wisconsin-Milwaukee. I benefited from comments on previous drafts by Bill Dougan, George Tolley, participants of the Economics and Policy Workshop of the University of Chicago, and two anonymous referees.

Printer friendly Cite/link Email Feedback | |

Author: | Bender, Bruce |
---|---|

Publication: | Economic Inquiry |

Date: | Jul 1, 1991 |

Words: | 6940 |

Previous Article: | Why doesn't society minimize central bank secrecy? |

Next Article: | Is the budget deficit "too large?" |

Topics: |