The Significance of Roll Calls in Voting Bodies: A Model and Statistical Estimation

In the long h story of legislative roll call analyses, there continues to exist a particularly tro bling problem: There is no satisfactory method for measuring the relative im ortance or significance of individual roll calls. A measure of roll call significance 1 would be intersting in and of itself, but many have realized that it could also su stantially improve empirical research. The consequence of this situation is that undreds of researchers risk heteroskedastic disturbances (resulting in inefficient es imates and biased standard errors and test statistics), are unable to appropriate1 choose the roll calls most suited to their theory (resulting in analyses that ay not correctly test their theory), and often use methods that create more pr blems than they solve (resulting in selection bias, unrealistic weighting sche es, or relatively subjective measures). This article introduces a new method de i igned to meet these problems. Based on an application of Box- Tiao interventi analysis, the method extracts from observed voting participation scores the "rev t aled preferences" of legislators as a measure of roll call sidcance. Applying this dethod to roll calls from the U.S. Senate demonstrates the success of the method /and suggests its utility in applied research. @ 1986 Academic Press, Inc. In the numerou recent studies of voting behavior in legislatures, the question of deciding upon t d e relative significance of roll calls remains a persistent and unsolved methodological problem. (Riker, 1959: 377)

i GARY KING later, researchers continue to note that, "there is etermine the most important vote" (Edwards, 1985: e unusual quality and extent of the data--on the floor of the U.S. Congress, in some state legislatures, and s around the world-it is also an important methr social science research. nificance of roll calls could substantially improve example, Matthews and Stimson (1975) argued simulation "model applied only to a subset, the ions" (see also Chenyholmes and Shapiro, 1969). s the non-routine, often controversial, rancorous mary interest" (see also Kingdon, 1981). Most best to particular types of roll calls--either the t routine, or the most representative. However, ed method of estimating the significance of roll holars routinely use all roll calls for selected eir analyses. The consequence of not discriminating 1 calls can be serious. For example, since there is e (error) in predictions for roll calls that are opriate than for those that are appropriate, these are surely heteroskedastic (i.e., have nonconstant ient estimates and biased standard errors nce of roll calls were known and utilized, be avoided. Furthermore, this additional information existing substantive theory. Interaction terms ns could be used to model those independent variables be most powerful among less significant roll calls and 1 among more significant roll calls. Variations ss groups (party, regional, and ideological an added dimension to legislative analysis. nd justified method of assigning significance is by William H. Riker (1959). Professor 1 call significance score by combining the with the number of representatives who ught was that margin would measure "the ntest [a roll call] outcome" (1959: 379).
of "roll call significbnce..? I have adopted Riker's (1959: 378) "distinction between the sign$cance of a roll all, a judgment made at the time a roll call occurs, and the importance of a roll call, a judg ent made at some time after it occurs." Riker argues that the judgment of roll call importan e "both of the individual scholar and of any conceivable panel of outside experts is su pect. The only remaining persons available to evaluate the significance of roll calls are the members of the legislature themselves." Thus, instead of using a complex series of c ding rules to ascertain a priori importance, concentration is on the behavior of legislat rs to estimate realized importance, what is called here "roll call significance." 1 i I ESTIMATING ROLL CALL SIGNIFICANCE 137 as a measure of effort; the more significant ipants there would be. By assigning values els of participation and columns indicating n) he derives a unique score for each combination of ot of credit for being the first and still the most treatment of the topic. However, three relevant First, Yohe (1968) does not propose to change ument, but he does note several biases in, and corrections to, Riker's method. umes that a narrow margin is the mark of an this may sometimes be true, I would argue onsider five arguments: (a) A roll call with a not necessarily one which is significant. Imagine, for has neither meaning nor significance for any case, it is likely that votes would be cast erage, in a 50-50% tie and the narrowest of f very low significance, complete cue taking is likely to resylt in the margin reflecting the party division in the legry narrow margin. On much more significant roll calls, turn out to be small or large depending upon the ferences. Thus, margin seems not to be a very good 11 call significance. (b) Although heightened controversy elated to importance-since controversy implies that ant to at least some legislators-a close margin is ntroversy. In the case of an insignificant roll call , the issue should also be noncontroversial. It margin would also make a poor indirect measure other possibility for using margin is as an indirect e through visibility. It could be argued that since elected officials value publicity, a visible roll call would be a significant one. The proble here is that a close margin does not appear to guarantee visibility. A un 1 nimous censure of a colleague, for example, is likely to have very high visibility. Thus, the margin of victory is quite a poor direct and indifect measure of roll call significance. (d) Furthermore, margin and participation are empirically unrelated. For example, in the 200 roll calls anqyzed below, the bivariate correlation between participation and margin is pnly 0.015. This makes the idea that the two variables represent one lfnderlying dimension dubious at best. (e) Finally, since most legislative bodies are characterized by full information and strategic voting, the dic otomous outcome of the roll call (win/lose) is rob ably of more import ", nce than the margin. Votes over and above the margin many times reqect short-run constituency and other ~r e s s u r e s .~ ! GARY KING method seeks to "determine" rather than "estimate" This is a crucial distinction; for only with statistical formal models can we make reliable probabilistic based conclusions. The statistical model which on Riker's useful model but concentrates more on estimation remainder of this paper (1) builds a model the appropriate methods of measuring of this model, and (3) estimates of roll calls from the United States Senate.
( A MODEL OF ROLL CALL SIGNIFICANCE By dropping margin (of victory or defeat) from Riker's (1959) measure, only the numb r of members participating in a roll call remains. This is a plausible me sure of significance, consistent with Riker's, which allows the legislators' behavior to define the significance of roll calls (for an application, se King, 1986a). If a roll call is more important, it is assumed i that more will participate. If it is less significant, then the constant pressures of c nstituents and others will reduce participation. Since roll call significanc is unobservable, or is at least very difficult to measure (see Footnote ), we extract an indicator from a phenomenon which is easier to observ the participatory behavior of the legislators.

I of how important the legislators themselves see each roll call.
I These procedures dre not only irrelevant to their stated purpose, but they can result in variable which may cause problems in this case is the degree of limited procedures have been applied to or proposed for the decisions (see Ma and Stimpson, 1975). (4) Still others eliminate all unanimous or of this relatively subj ctive measure. (6) Some personally select a small number of signifcant roll calls based on in erviews or a close reading of the congressional process; this amounts to replicating the C ngressional Quarterly method, but perhaps with different or more systematic selection rules (see Kingdon, 1981 TI, an intercept term a parameter vector representing the influence of X,. a vector of observed procedural variables that influence the number of legislators participating in roll call t.
(1 -BIB -&B* -. .  , even if these succeeding roll calls are not ssigns q (the order of the moving average roll calls for which participation remains ng to the data used in estimating this model and q , a discussion of three unresolved parate roll call significance from random rrect assumption that there is no error. scholarly literature and the congressional nlikely to cause a problem. In one, s, rror-a series of random shocks with 1 estimation will produce an unbiased be distributed randomly and would er, is that physical time ime, and the latter can antly, alternative spece tried, but they added s may abstain in order uments help to refute The only reason ble way to explain such behavior is to say that the abstainers value the effect f their abstention more than the effect of their vote. And this is another way o saying that they do not value their vote on this issue as highly as the effects of bstention. One cannot evade the conclusion, therefore, that the vote in itself is n 1 t of the highest significance. (Riker, 1959: 381) Second, when in groups (such as the Americans for Democratic Action), constitu or party leaders maintain an attitude such that ith me is against me,' " legislators "are less likely , 1974: 79). Third, some interest groups and organgressional Quarterly do regular polls of abstainers their unstated preferences. These polls are expected swered by members of congress. They are then he U.S. Congress, abstention is quite a poor way ecision. Finally, abstention is not popular among n the basis of challengers' congressional compaigns. to be more unpopular when it occurs on significant even more likely to be an accurate measure of gnilicance model produces a "revealed preference" ent of the significance of each roll call is revealed gh their behavior. In economics and in other areas ars have shown that revealed preference measures to opinion polls, expert judgments, and any other to estimate expected inflation, economists prefer asury bills rather than to ask experts for their reases (Fama and Gibbons, 1984). In a similar oped here uses a realized or revealed preference f roll call significance.
gislators participating, n,, is measured with all 200 ed States Senate occurring during the first 6 months o compelling reason to select this period rather than s no reason to believe that the relationships reported rest of the series. I believe that the results her voting bodies around the world, but this verdict ication of this model to these other legislatures. del, only the precise definition of the procedural mber of roll calls for which a shock to significance need to be specified. les, X I , are measured in a variety of ways and ed as 1 if roll call t occurred on Friday through through Thursday. To test the assumption of ntervals, and to control for unequal intervals s coded as the number of days between the 011 call at time t -1.4 For testing purposes, feat), whether the president takes a position ea of the roll call, and measures of intraparty ates on the previous random shocks (the moving y plausibly indicate that a change in significance 1 call would have about 58% reverberation during during the next, 26% in the next, and 13% in the errors are all very small, indicating very precise arginal probabilities (in the last column) indicate sample of roll calls were taken, all of the parameters main different from zero and retain the same sign. coefficients, the estimate of the significance of bers of the U.S. senate was generated. Although 11s in this paper is not feasible, Tables 2 and 3 cance scores (j,), Congressional Quarterly roll e), and substantive descriptions for the 10 most calls-as determined by the legislators' actions d by the roll call significance model, and as g the most highly significant to the senators ortant issues (Table 2). All were of relatively ors and high visibility to the public. In general, t important (Table 3) were considerably more the tables. Reduce authorizations from $13.7 million in fiscal 1980 and $25 million in fiscal 1981 to $9.5 and $9.8 million, respectively. There was substantial debate and public attention. This was the first of 10 votes attempting to amend the bill. The final roll call was pro forma, since the real decision had been made on this vote (p. 305). Adopted, n, = 80, 6/2/80. H J Res 521 Draft Registration Funding. Motion to invoke cloture (thus limiting debate) on the joint resolution to transfer $13.3 million from the Airforce System in order to provide enough funds to register 19-and 20-year-old males, beginning in 1980 in case a military draft became necessary in the future. President Carter, changing his position in response to the Soviet invasion of Afghanistan, already had the authority and needed only these funds for implementation. The plan was very controversial: "all of Carter's opponents in the 1980 presidential election came out against the proposal." Motion agreed to, n, = 94, 6/10/80. S 10 Rights of the Institutionalized. Amendment allowing Justice Department suits only when the Attorney General has substantial evidence that a person is being subjected to egregious or flagrant conditions. This was fourth of eight consecutive roll calls on the same topic, but all were on different points. The language in this amendment ("egregious or flagrant conditions") struck at the heart of this plan. Rejected, n, = 90, 2/28/80 (see HR 10, below  Adoption of the conference report on the joint resolution to provide a supplemental appropriation of a little less than 0 asked for by the Senate previously for 1980. It directs the Agriculture Department to proceed on the assumption that * P total spending for the year would be $9.191 billion and to prohibit more spending. The day before there had been a by adoption of the authorization measure (S 13091, both chambers adopted the conference report" (p. 415). Adopted, Z 0 n, =  Passage of bill to authorize $2.2 billion in fiscal 1981-1983 to assist low-income persons to "weatherize" their homes to conserve energy. It was perceived as an energy issue rather than social welfare and was not the focus of much attention. Later, "the legislation got bogged down in a committee jurisdiction dispute in the House and never cleared" (p. 462). Passed, n, = 80, 2/28/80. HR 7428 Oil Import Fee/Debt Limit. Passage, over the president's veto the day before, of the existing debt limit of $879 billion and to disapprove the $4.62 fee per barrel of imported oil that President Carter imposed effective March 15. Required and received two-thirds vote. This was a high-visibility vote which was never taken seriously by most members of congress. "From the beginning the opposition movement attracted members from opposite ends of the political spectrum." Although Carter, June 4, had appealed to legislators to "stand up and take the political heat by making tough decisions," "earlier votes left no doubt that the veto would be overridden" (p. 273). This was probably important to Carter, but not to the Senate. Passed, n, = 78, 6/6/80.

V1
Motion to table (kill) the Thurmond, R-SC, amendment to cut $800 million in budget authority and $400 million in