Legislative voting and the economic theory of politics.

• Author: Coates, Dennis

1. Introduction

   Empirical studies of legislative voting have multiplied since the mid-1970s. Early work [13] distinguished two modes of representation: "Delegates," who vote as voters want, and "Trustees," who use their own judgment. More recently, researchers have considered only delegates, claiming that legislators who use their own "ideology" are shirking. Following principal-agent theory, representatives are conceived as conduits through which citizens transmit and express interests. Empirical work on this question, however, has found that the legislator's ideology explains more variation in vote choices than is explained by measures of constituent interest. The seeming contradiction has led some [9; 10; 17] to reject the "economic theory" of legislator behavior as erroneous. The truth is that no such theory has ever been fairly stated, much less tested. Estimating equations have been ad hoc, and the usual approach to measuring "ideology" has imposed correlation between the equation error and an explanatory variable [8]. In both cases, the result is biased coefficient estimates.

   Empirical models should be derived, not assumed. We show that the theory of constrained choice leads to a theoretically defensible, and empirically rich, estimation procedure. For the sake of easy comparison with well-known results, the approach is tested using the data analyzed by Kalt and Zupan [9]. The next section reviews previous work on legislative voting and argues tests of shirking based on ideological ratings are flawed. Section III presents the theoretical model, and section IV describes the data and results. Conclusions are offered in the fifth and final section.

2. Previous Work

   Empirical research into the voting patterns of legislators is often based on the "theory of economic regulation" [19]. The literature on political "shirking" [9; 10; 11; 12; 16] has regressed Congressional vote choices on constituent characteristics, and on "ideology" (measured as an index of votes).(1) A generic estimating equation representative of this research could be written:

   [Mathematical Expression Omitted]

   where [Y.sub.ik] is legislator i's vote on bill k, [C.sub.ij] is the jth characteristic of i's district, the first [J.sup.c] of which have a similar or "common" influence on all bills, and the next [J.sup.k] - 1 - ([J.sup.c] + 1) are relevant only for bill k. [I.sub.i] is i's "ideology," [[Mu].sub.ik] is unobservable and [Beta] are parameters to be estimated.(2) Kau and Rubin [11] and Kalt and Zupan [10] find that ideology has a statistically significant effect on vote choice; each paper claims this significance captures mismeasured constituency interests. The ideology variable is then in turn modeled as being explained by constituency characteristics from equation (1), and some additional constituency variables. The fitted ideology variable and the fitted error are used as regressors in equation (1). If Stigler's model of politics is correct, these authors contend, then the unexplained portion of the ideology variable will be insignificant. But the null hypothesis is always rejected; ideology matters.

   Such tests for shirking are seriously flawed, both theoretically and econometrically, because any rigorous test for shirking must be tied to a model of behavior. One place to start is the principal-agent model, where performance is found to be positively related to levels of monitoring and enforcement. If Congressmen value incumbency, and monitoring their actions is costly, then those in close races will shirk less, all else equal, than others whose seats are uncontested. Consequently, the electoral pressures on the legislator should condition the effects of constituent interest variables. Otherwise, community interests are restricted to be equal whether the legislator is safe or marginal.

   Some attempts to account for the effect of electoral pressures have recognized this conditioning influence.(3) None of these approaches, however, allows for the simple expedient of including an interaction among constituent preferences, legislator preferences, and political support. Such an approach improves existing work in at least three ways. First, if shirking is a "good" to be purchased, then higher cost reduces quantity demanded. Expected political support is a measure of the marginal cost of shirking that is specific to the legislator in question, so that our approach is more direct than more ad hoc specifications. Second, the use of the concept of "ideology" is misleading if the fundamental hypotheses involve shirking on votes, but the only measures of ideology are themselves vote indices. The necessary assumptions for such analysis to be valid, a unidimensional ideology and a noninterior ideal point for the group calculating the ranking, are rarely satisfied.(4) Further, such a notion of ideology is quite different from the psychological definition of a shared belief system [5].

   Finally, the use of voting indices as regressors creates correlation between the equation error and an explanatory variable. Jackson and Kingdon [8] show that the use of vote indices (such as those calculated by Americans for Democratic Action or the U.S. Chamber of commerce), over-estimates the impact of "ideology" and underestimates the influence of economic considerations. They claim that the unobservable influences on a legislator to vote "yes" on one bill are correlated with the unobservable influences on that legislator to support, or oppose, other bills. Jackson and Kingdon conclude that use of these indices produces biased and inconsistent coefficient estimates. On the other hand, omission of a measure of ideology similarly biases the estimates. Since these are the only two alternatives, Jackson and Kingdon claim voting models cannot be estimated meaningfully.

   To grasp these points, it is useful to consider an arbitrary vote index:

   [Mathematical Expression Omitted].

   [I*.sub.i] is the proportion of the time that legislator i votes the position of some interest group, [Y.sub.in] is again i's vote on bill n, [[Beta]*.sub.j] is the average effect of regressor j, and [[Mu].sub.i] is the average unobservable over the N votes. Because [I.sub.i] is unobservable [I*.sub.i] is substituted in estimation of equation (1). Equation (3) is the result of this substitution.

   [Mathematical Expression Omitted]

   Three implications follow from the argument made in Jackson and Kingdon [8]. First, the influence of constituent characteristics that affect all votes, the [[Beta].sub.j] (which enter (3) in two places), is likely to be small. The [[Beta].sub.j] enter (3) first as they affect voting on average (the [[Beta].sub.j] from j = 1 to j = [J.sup.c]) and second as they affect this specific vote (the [[Beta].sub.j] from j = [J.sup.c] + 1 to j = [J.sup.k] - 1). For the characteristics that influence all votes to be significantly different from zero they must have an impact on this vote different from their influence on the average vote. Second, influences on this vote alone, the [Mathematical Expression Omitted], enter the equation only once, and are more likely to be significant than are the general influences.

   Third, the vote index will be correlated with the equation error, so that coefficient estimates are biased and inconsistent. Two-stage methods are inappropriate for obtaining consistent estimates of the impact of "ideology," however, because the vote index (used directly or fitted in a preliminary regression) is a poor...