The Jeffords Switch and Legislator Rolls in the U.S. Senate

Transcription

1 The Jeffords Switch and Legislator Rolls in the U.S. Senate Abstract On May 24, 2001 United States Senator James Jeffords announced that he was switching from Republican to independent and would vote with Democrats on organizational matters, effectively taking majority party control of the Senate from the Republicans and giving it to the Democrats. This created an unusually well controlled quasi-experimental opportunity for learning about the role of parties in the Senate it held most important variables constant while one variable, majority status, changed. We use roll call data to evaluate the probability of individual members of each party being rolled on Senate final passage votes, before and after the switch. We find that, contrary to conventional wisdom on the Senate, majority status is an important factor in Senate decision making. Our results show that Republicans were more likely to be rolled after the switch than they had been before, and that Democrats were less likely to be rolled than they had been before.

2 Congressional scholars sharply dispute the role of the majority party in shaping U.S. Senate decisions. On the one hand, the view that Senate parties exert little influence over the chamber runs throughout postwar congressional scholarship. The most influential studies, which focus on Senate procedures and practices such as holds, filibusters, and nearly unlimited amendment opportunities, portray the chamber as one in which power is broadly distributed across individual senators and the chamber as a whole is extremely difficult to manage (Matthews 1960; Ripley 1969; Sinclair 1989; Smith 1989). These studies are rich in detail and context, but tend to either ignore or explicitly downplay the importance of parties. 1 As partisanship has increased since the 1980s, parties have increasingly entered into analyses of Senate behavior but, rather than consensus, this has led to divergent views about parties effects. Some studies argue that partisanship exacerbates the gridlock and dysfunction caused by individualism (Binder 1997, 1999, 2003; Binder and Smith 1997; Oppenheimer and Hetherington 2008; Smith 2005, 2007). From this perspective, partisan and individual goals act as competing interests, putting enormous pressure on Senate leadership to deliver partisan advantage in a chamber set up to empower individual interests (Sinclair 2001). Other studies, however, point to Senate procedures and processes that work to the advantage of the majority party (Beth et al. 2009; Campbell 2001, 2004; Den Hartog and Monroe 2008, 2011; Evans and Oleszek 2001; Evans and Lipinski 2005; Koger 2010; Lee 2009; Schiller 1995, 2000, 2001; Wawro and Schickler 2006). Some of these 1 Other models of congressional decision making leave out parties altogether. See, for example, Krehbiel 1998; see also Brady and Volden 1998.

3 authors, as well as others, discern different types of pro-majority bias in Senate decisions (Bargen 2003; Campbell 2001, 2004; Campbell, Cox, and McCubbins 2002; Crespin and Finocchiaro 2008; Den Hartog and Monroe 2008; Gailmard and Jenkins 2007; Koger and Fowler 2006; Lee 2009). We test implications of these competing views by examining how legislative outcomes varied when Democrats suddenly gained majority status in the middle of the 107 th Senate, due to Senator James Jeffords s change in party affiliation. More specifically, we look at individual senators probability of being rolled that is, voting against a bill that passes on final passage votes in the periods immediately before and after the Jeffords switch (i.e., the change in majority status), and how these probabilities changed when Democrats gained and Republicans lost majority status. We come down on the side of the argument that majority status is an important factor in Senate decision-making. Our results show that Republicans were less successful after the switch than they had been before, and that Democrats were more successful after the switch than they had been before. 2 However, in some ways, the results also suggest limits on majority party power in the Senate, which we discuss later in the paper. The Jeffords switch provides a unique quasi-experimental opportunity for examining the majority party s role, because it allows us to isolate the treatment variable majority status while holding constant other variables that scholars identify as determinants of Senate decisions. It is the only instance in which majority status changed without any concurrent change in Senate membership every other change in 2 Also, Republicans were more successful than Democrats pre-switch, and Democrats were more successful than Republicans post-switch. 3

4 majority status resulted from an election that produced a new majority, but that also produced simultaneous changes in factors such as ideological heterogeneity among members of the Senate, membership in the House of Representatives, and new expressions of constituent preferences. 3 Many elections also produced changes in control of the House, the White House, or issue salience. Across the period just before and just after the Jeffords switch, however, all these things remained constant. Others have used the tidy research design produced by the Jeffords switch to study aspects of American politics. For instance, Nicholson (2005) uses the switch to study public support for divided government, Jayachandran (2006) uses it to study asset prices, and Roberts (2007) uses it to illustrate difficulties with drawing inferences about legislators preferences from roll call votes. The work closest to ours is Den Hartog and Monroe (2008), which also uses the switch to study partisan influence on Senate outcomes. But their dependent variable stock returns for Democrat-supported and Republican-supported energy companies assesses the impact of the switch by examining outcomes external to Congress. Their research design relies on the premise that investors reactions to the switch accurately reflected shifting party power within Congress and the consequences such a shift would have on energy stocks. Put differently, to interpret their results as evidence of majority party bias in the Senate, one must believe that the market correctly anticipated the legislative consequences of the switch. We use a research design that is similar, but does not rely on such premises. Our dependent variable, a senator s probability of being rolled, is similar to those used by 3 There has never been a change in majority status in the House without a concurrent change in membership. 4

5 Carroll and Kim (2010), Carson, Monroe, and Robinson (2011), Cox and McCubbins (2005), Lawrence, Maltzman, and Smith (2005, 2006), and Smith (2007) each of which uses some measure of individual legislators legislative success and compares the success of majority party legislators to the success of minority party legislators to draw inferences about party power. A common element of these other studies, however, is that they examine time series that stretch across multiple congresses, with one observation for each member in each congress, and majority status changes only between congresses at the same time that changes occur in preferences and other important factors. The test in this paper has the novel twist that the period we study includes a change of majority status without a simultaneous change in other important variables. 4 We give a brief account of the Jeffords switch in the next section, followed by a section in which we sketch how the hypotheses that we test follow from relevant literature. In subsequent sections we detail our research design and empirical results, then discuss the implications of the results. The Jeffords switch Following the 2000 election, the Senate was evenly divided between Democrats and Republicans, leaving the Vice President as the tie-breaking vote and de facto determinant of which party enjoyed majority status. From January 3rd through the 20th, 4 A possible objection at this point is that Jeffords s policy preferences might have changed, and that such a change in preferences explains our results, rather than the change in majority status. We maintain the assumption that his preferences did not change; we discuss reasons for this assumption later in the paper, when we discuss our empirical results. 5

6 Al Gore remained Vice President and Democrats were the majority party although, knowing that they were about to lose majority status, they took little action. When Dick Cheney was sworn in as Vice President on January 20th, Republicans assumed majority status and, with it, unified control of the government. Realizing that they faced an extraordinary situation, Democratic leader Tom Daschle and Republican leader Trent Lott agreed at the outset to a so-called power-sharing agreement that, among other things, evenly divided seats on all committees between Republicans and Democrats, while making Republicans the chairs of each committee. This continued for the next few months, during which the Bush administration defied many predictions by pushing a conservative agenda, to the consternation of some moderate Republicans (Martinez 2001). On May 24, one of these moderates Senator James Jeffords of Vermont announced that he would switch from Republican to independent and would vote with Democrats on organizational matters. The effect was to give Democrats a advantage on organizational votes, thereby making them the majority party. 5 The Senate recessed on May 26, after passing Bush s tax bill; when the session resumed in June, Democrats assumed majority status; Daschle became majority leader, Democrats became chairs of each committee, and (after a period of bargaining and procedural wrangling) Democrats assumed one-seat majorities on each committee. 5 Jeffords s given reason for switching was that the increasingly conservative Republican caucus made it difficult for him to remain moderate; as part of the switch, Democratic leaders guaranteed that they would be tolerant of his moderate positions, and also gave him a committee chairmanship (Jeffords 2001, Daschle 2003). 6

7 regained the. 6 The new alignment was still in place a few months later when the terrorist attacks of September 11 put a temporary end to politics-as-usual. Parties hypothesesized effects Despite the qualitative and quantitative richness of the Senate literature, theories of Senate parties effects are relatively scant. One approach is to transfer cartel (Cox and McCubbins 1993, 2005) or conditional party government theories (Aldrich and Rohde 2000, 2001; Rohde 1991) developed in the context of the House directly to the Senate (Campbell, Cox, and McCubbins 2002; Chiou and Rothenberg 2003; Rohde 1992). This approach relies on the assumption that Senate agenda setting is similar to House agenda setting. Another approach is to model Senate majority agenda setting as the result of a bargaining game in which proposal power and costs of moving bills through the legislative process interact to give the majority disproportionate influence over the agenda (Den Hartog and Monroe 2011). Regardless of the approach, however, these models all emphasize the majority party s ability to shape the Senate agenda, primarily via procedural tactics, 7 as the key to its influence over the chamber s legislative decisions. By moving forward bills favored 6 The Senate was in the midst of working on the tax bill at the time of Jeffords s announcement; part of his deal with Democrats was that the switch would not occur until after the Senate passed the bill (Jeffords 2001). 7 Much of the debate on parties influence centers on the extent to which Senate procedures either undermine or empower the majority party s ability to affect the agenda and, therefore, outcomes. See Chapter 2 of Den Hartog and Monroe (2011) for an overview. 7

8 by its caucus members, and by hindering bills opposed by its caucus members, the majority increases the likelihood that bills passed by the Senate will be to its liking and thereby decreases caucus members probability of being rolled. Thus, partisan theories predict members of the majority caucus are less likely to be rolled than members of the minority caucus, all else constant. 8 Combined with the Jeffords switch, this is the basis for the predictions that we test: Republican Change Hypothesis: Republican senators probability of being rolled on a final passage vote will be higher after the Jeffords switch than before it, c.p. Democratic Change Hypothesis: Democratic senators probability of being rolled on a final passage vote will be lower after the Jeffords switch than before it, c.p. Pre-Switch Hypothesis: In the period before the switch, Republican senators probability of being rolled on a final passage vote will be lower than Democratic senators probability of being rolled on a final passage vote, c.p. Post-Switch Hypothesis: In the period after the switch, Republican senators probability of being rolled on a final passage vote will be higher than Democratic senators probability of being rolled on a final passage vote, c.p. Note that the first two hypotheses serve as the key quasi-experimental tests that capture the effect of changes in the treatment majority status on roll probabilities, since only partisan theories predict changes across the switch, and non-partisan theories predict no changes across the switch. 8 Predictions of this type follow from a variety of formal models; see Carson, Robinson, and Monroe (2011) for a detailed discussion of the relationship between formal agenda control models and the likelihood of individual legislators being rolled. 8

9 The latter two hypotheses, taken individually, are weaker tests. They present opportunities to falsify the partisan hypotheses; but, under some conditions, non-partisan theories also predict majority senators having lower roll probabilities, so evidence consistent with the hypotheses does not allow us to differentiate partisan and non-partisan theories. Taken jointly, however, these two hypotheses constitute a stronger test, since that allow non-partisan models to predict lower roll probabilities for members of one caucus in the pre-switch period are unlikely to change in ways that allow non-partisan models to predict lower roll probabilities for members of the other caucus in the postswitch period. For example, in a given time interval, a non-partisan model might predict low roll probabilities for majority caucus members if most status quos are to the minority s liking. In such a case, minority members would unsuccessfully oppose proposals to move policy toward the floor median, thereby being rolled, while majority members would favor the shift and would not be rolled. In the context of the Jeffords switch, however, the status quos would have to suddenly shift from one side of the political spectrum to the other and at the same time that Jeffords switched for a nonpartisan model to predict that the last two hypotheses both be true at the same time. Test and data To compare senators probability of being rolled across the switch, we use a dataset with one observation for each senator in each period (i.e., one observation before the switch and one observation after it). The pre-switch period runs from the beginning of the 107 th Congress, in January 2001, through the switch; the post-switch period that we use runs from the switch through September 10, We thus compare two periods of roughly comparable length (3-4 months), and avoid the possible confounding influence of 9

10 September 11 th. Indeed, one of the advantages of using such a short time span is that it greatly reduces the possibility that an unobserved factor, such as changing political issues or alignments, is actually responsible for whatever changes we observe. For each observation, we calculate the number of final passage roll call votes on which the given senator voted in the given period, as well as the number of votes on which the senator was rolled. 9 We also code whether the observation is for the pre- or post-switch period, and whether the senator was a Republican or Democrat. Using this dataset, we estimate the following model using extended beta binomial regression: RollProbability it = α + β 1 Dem i + β 2 PostSwitch t + β 3 Dem i *PostSwitch t + β 4 Distance i + ε it, where: RollProbability it is the proportion of final passage votes on which senator i was rolled i.e., voted against a bill that passed in period t, 10 9 The set of votes used here is each Senate final passage vote on a Senate or House bill, a Senate or House joint resolution, an executive nomination, or a conference report. There were 26 votes in the pre-switch period and 18 votes in the post-switch period. Not all senators voted on every vote, although most voted on most of the votes. 10 The dependent variable in our hypotheses is a senator s probability of being rolled on a final passage vote, which suggests using a simple probit or logit model, with an observation for each member on each vote and a dummy dependent variable for whether or not senator i was rolled on that vote. To do so, however, would be to ignore the lack of independence among the votes that a legislator casts. Using the extended beta binomial (EBB) method (King 1989; Palmquist 1999) allows us to deal with this problem and produce better estimates of significance. EBB is somewhat akin to grouped logit but, 10

11 Dem i is a dummy, coded one if senator i is a Democrat, PostSwitch t is a dummy, coded one for the post-switch period, and Distance i is the absolute value of the difference between i s ideal point and the floor median s ideal point, using the first dimension common-space NOMINATE scores (Poole 1998; Poole, McCarty, and Rosenthal 1997; Poole and Rosenthal 1997). The dummy variable PostSwitch t captures the change in Republicans probability of being rolled and is thus the key variable for testing the Republican Change Hypothesis, which predicts a positive coefficient for this variable (indicating that Republicans probability of being rolled increased after the switch). The PostSwitch t and Dem i *PostSwitch t variables jointly capture the change in Democrats probability of being rolled and are thus the key variables for testing the Democratic Change Hypothesis, which predicts that the sum of the coefficients for these variables will be negative (indicating that Democrats probability of being rolled decreased after the switch). The unlike the latter method, it explicitly models and accounts for the fact that each binary trial (i.e., each separate final passage vote) is not independent of other trials for a given individual; it entails specifying the number of positive trials and the number of total trials. As with simple probit or logit, EBB is a maximum likelihood method that produces non-linear coefficients that are not intuitive, but can be translated into a variable s effect on the probability that a trial has a positive outcome (i.e., the probability that a senator is rolled on a final passage vote). Hence, our labeling of our dependent variable as the probability of senator i being rolled and our description of it as the proportion of votes on which senator i is rolled. 11

12 variable Dem i captures how Democrats pre-switch probability of being rolled compares to Republicans pre-switch probability of being rolled; the Pre-Switch Hypothesis predicts it will be positive. Finally, the Dem i and Dem i *PostSwitch t variables jointly capture how Democrats post-switch probability of being rolled compares to republicans post-switch probability of being rolled; the Post-Switch Hypothesis predicts it will be negative. In addition, we add the Distance i variable as a control to account for the ideological extremeness of each senator. In prominent models of legislative agenda setting (Cox and McCubbins 2005; Krehbiel 1998), a legislator s probability of being rolled is (weakly) greater as the member s ideal point is farther from the floor median s ideal point. 11 Thus, while most possible confounding factors are unlikely to vary across the short period of time we examine, we include this control to account for ideologydriven heterogeneity in senators probability of being rolled. 11 More accurately, a legislator s probability of being rolled is greater as the member s ideal point is farther outside the protected interval, making standard assumptions about the distribution of status quo points. Because different assumptions about the agendasetter and floor veto players lead to different protected intervals, the appropriate control for ideological extremeness differs from one model to the next. Including all different possible permutations of this control would increase the length and complexity of the text and results with, we believe, little payoff. In results not reported here, we have experimented with different permutations, and in no case has it made a difference in the inferences that we draw about the hypotheses. We thus report results using only the crude Distance variable defined above. 12

13 Results We now turn to the data. Before discussing the regression results, we first look at the pattern of change by examining how each senator s roll rate (i.e., the proportion of votes on which i was rolled) for the post-switch period compares to his or her roll rate for the pre-switch period, in Figure 1. Each plotted point represents the difference between the post-switch roll rate and the pre-switch roll rate for a given senator. We plot this difference (on the y-axis) against the senator s first-dimension common-space NOMINATE score (on the x-axis), with triangles denoting Republicans and squares denoting Democrats. Note the horizontal line across the middle of the figure, which marks a zero roll rate change; individuals below this line (i.e., with a negative change) had higher roll rates before the switch than after, while individuals above this line (i.e., with a positive change) had higher roll rates after the switch than before. Thus, if our hypotheses are correct, we should see Democrats tend to be below the line, and Republicans tend to be above the line. Clearly, this is the pattern we observe. Of the 51 Democrats (counting Jeffords as a Democrat, which we have done in Figure 1), only two Russ Feingold, at the far left of Figure 1, and Jeffords himself, who is at.26 on the x-axis are above the line, while one more Zell Miller, in the middle of the figure had a change of zero (his roll rate was zero in each period). The other 48 Democrats all had lower roll rates after the switch than before. In sharp contrast, 40 of the 49 Republicans are above the line, indicating that their roll rates were higher after the switch than before. Of the other nine, four are below the line and five are on the line, indicating no change. For Republicans, the mean change in roll rate is.054, while for Democrats the mean change is -.158; in each case, a single-sample test allows us to reject the null hypothesis of no change with p-values smaller than Thus, it clearly seems 13

14 that Democrats did better after the change than before, and that the reverse is true for (most) Republicans. Figure 1 here We now turn to the regression results, shown in Table 1. The coefficient for PostSwitch, which captures the change in Republicans probability of being rolled and is predicted by the Republican Change Hypothesis to be positive, is significantly positive (the coefficient is with a standard error of.157). 12 This indicates that Republicans probability of being rolled did indeed increase after the switch. Table 1 here The results also support the Democratic Change Hypothesis, which predicts that the sum of the coefficients for PostSwitch and Dem*PostSwitch will be negative. A linear combination test shows that this sum is significantly negative (the sum of the coefficients is , with a standard error of.121; the coefficient for Dem*PostSwitch is , with a standard error of.198). This indicates that Democrats probability of being rolled significantly declined after the switch. As mentioned, we can also look at which party s members did better in each period. The coefficient for Dem captures the difference between Democrats and Republicans probability of being rolled in the pre-switch period. The positive coefficient 12 We drop subscripts from variable names in the rest of the discussion. The Pseudo R 2 is.1168, the Log-likelihood is , and the number of observations is 198 because Jeffords is excluded from the analysis, since he was in the majority in each period. Including Jeffords makes no significant difference in the results. 14

15 (2.558, with a standard error of.139) indicates that, as expected, Democrats were more likely than Republicans to be rolled before the switch. In the post-switch period, the difference between Democrats and Republicans probability of being rolled is the sum of the coefficients for Dem and Dem*PostSwitch. This sum is negative (-.268, with a standard error of.142) as predicted, 13 indicating that Republicans probability of being rolled was higher than Democrats after the switch, all else constant. 14 Table 2 shows the substantive content of these results, casting them in terms of the predicted probability of being rolled. Each cell shows the estimated probability that a final passage vote will result in a roll for an individual senator, differentiated by party and period. 15 The probability for Republicans goes from.026 before the switch to.098 after the switch, while for Democrats it goes from.214 to.061. In addition, we see that Democrats probability (.214) was higher than Republicans (.026) pre-switch, but that Democrats probability was lower than Republicans post-switch. All of these results are consistent with our hypotheses. Table 2 here 13 In a one-tailed test, the significance level is 97.1%; in all other cases discussed here, the significance level is over %. 14 At the suggestion of an anonymous reviewer, we have also estimated this model without the Distance variable. The results are very similar, with two small exceptions: the Pseudo-R 2 drops from.1168 to.096, and the signifcance level of the Dem-plus- Dem*PostSwitch sum drops from 97.1% to 94.7%. 15 Distance is held constant at its median value for members of each party. 15

16 Could a change in Jeffords s preferences explain our results? A potential problem with the research design used in this paper is the possibility that the results presented here are a function of a change in Jeffords s preferences, rather than of the change in majority party. In other words, changing preferences might be an important omitted variable. In this section, we address and reject this possibility. Obviously, we cannot actually observe or perfectly measure Jeffords s preferences, but we can estimate them using roll call based measures such as NOMINATE (Poole and Rosenthal 1997) and, at first blush, some evidence seems to suggest that Jeffords s preferences did change. Using various approaches to estimating preferences, Clinton, Jackman, and Rivers (2004), Nokken and Poole (2004), and Poole and Rosenthal s NOMINATE data all indicate that Jeffords s voting became distinctly more liberal after he switched. Nonetheless, there are multiple reasons to believe that a significant change in Jeffords s preference does not explain our results. First, the results are based on a set of roll call votes that covers a wide range of issues; for a change in prferences to explain all this pattern of change across many issues, Jeffords would have to have abruptly and simultaneously changed his positions on many issues. This seems to lack face validity, and also conflicts with anecdotal accounts of the rift between Jeffords and his party most of which emphasize that moderate Senate Republicans like Jeffords had become increasingly marginalized within their own caucus as it moved to the right. Second, a change in observed voting behavior does not necessarily imply a change in preferences, which is only one of many factors, such as constituency and party loyalty, that contribute to voting behavior. Trying to parse the effects of each on a legislator s roll call behavior is notoriously difficult, and estimation measures such as 16

17 NOMINATE inherit this ambiguity 16 that is, they produce a single estimate for a legislator that is a product of preferences, constituency, partisan, and other effects. Given that partisan theories presume that party members will exhibit at least some degree of party loyalty on at least some votes, a change in Jeffords s NOMINATE score is consistent with partisan theories and, after all, Jeffords publicly announced that he would change his voting behavior on organizational votes. It is thus unsurprising from a partisan viewpoint that Jeffords s voting pattern differed somewhat after the switch. Third, regardless of whether we take roll call-based estimates as measures only of preferences, evidence suggests that the change in Jeffords s roll call behavior did not occur during the period we examine and therefore cannot explain our results. The estimates mentioned above use roll call data from longer time spans than the severalmonth window that we examine in our analysis, including at least the entirety of the 107 th Congress. Using Poole and Rosenthal s W-NOMINATE program, 17 we estimated W- NOMINATE scores for each day across the period from March 2001 until September 10 th In other words, for the first day in this time series, we began by estimating scores for each Senator based on all roll calls that had occurred in the 107 th Congress up to that point. For the next day on which there was at least one additional roll call, we then reestimated the scores, incorporating the new roll call data. For each day in the time series, we updated the scores in this fashion, so that we have a time series of estimated (firstdimension) W-NOMINATE scores across this period. Figure 2 here 16 See Smith (2007) for an overview of the problems with parsing effects. 17 The W-NOMINATE program is available at Voteview.com. 17

18 Jeffords s first-dimension score, shown in Figure 2, bounces around some, especially early when the scores reflect fewer votes. But, overall, the trend is relatively flat and centrist across the window as a whole. These cardinal estimates of his preferences may be misleading, since this methodology does not account for whether or how the estimated policy space itself bounces around across this period. In other words, Jeffords s relatively constant scores could be consistent with him becoming more liberal, if the policy space itself were moving to the right. To account for this possibility, we have also examined Jeffords s ordinal ranking on each day across this period. Viewed from this perspective, the results provide fairly compelling evidence that Jeffords was not moving to the left relative to other senators. He was the 49 th, 50 th, or 51 st most conservative senator for every day in the time series. In other words, he was at or very near the median position across the pre- and post-switch periods we examine We also estimated W-NOMINATE scores for each Senator in the 107 th Congress across the entire period of our analysis; that is, we use all Senate roll call votes from the beginning of the 107 th Congress until September 10 th, 2001, so that we get one estimate for each senator (except Jeffords) for the entire pre- and post-switch period. We treat preand post-switch Jeffords as two different individuals, in order to contrast his pre- and post-switch voting behavior. The result is an estimated pre-switch (first-dimension) ideal point of.14, and a post-switch ideal point of which is a very small shift in voting behavior. To put this change in perspective, compare it to the ideal points of a few other prominent Senators from across the political spectrum, estimated across the same time period: Wellstone (-.969); Kennedy (-.813); Daschle (-.719); Baucus (-.098); Chafee (.159); Specter (.289); Lugar (.715); Santorum (.825); Helms (1.000). 18

19 Finally, even if Jeffords s preferences did change, it is unlikely to have changed the location of the median legislator enough to explain the pattern of roll rates we observe across a range of issues. In preference-based theories such as Krehbiel s (1998) pivot model, the location of the ideological median legislator is a key determinant of legislative outcomes. In Poole and Rosenthal s DW-NOMINATE estimates for the entire 107 th Congress, Jeffords s score shifts from as a Republican to -.34 as an Independent, in the process changing from the 51 st to the 30 th most liberal senator. However, the chamber median hardly shifts, from to -.062, which seems unlikely to explain the oberved changes in roll probabilities for Republicans and Democrats. Do Republicans do better than Democrats? One intriguing aspect of these findings, which goes beyond our hypotheses but suggests an interesting possible characteristic of Senate decision-making, is that the postswitch Democratic majority clearly did not seem to do as well as had the pre-switch Republican majority nor did the post-switch Republican majority seem to do as badly as the pre-switch Democratic majority. From the point of view of partisan theories of legislative decision making, there seem to be two most likely explanations. First, from a Conditional Party Government perspective, one might immediately suspect that the Republicans were more homogeneous in their preferences than were the Democrats, leading to more delegation to party leaders among the Republican majority than among the Democratic majority. Though we make no pretense of answering this question definitively, we can take a crude look at whether this conjecture holds up to scrutiny by comparing the heterogeneity of 19

20 preferences among senators of each party in the 107 th Congress. We do so in two ways by taking the standard deviation of first-dimension common space NOMINATE scores for senators of each party, and by calculating the range between the minimum and maximum score for each party (we exclude Jeffords from each party). 19 For Democrats, the standard deviation is.134 and the range is.651; for Republicans, the standard deviation is.142 and the range is.653. There is thus little evidence that greater Republican unity produced the Republican majority s seemingly higher level of success. The other possibility is that the result is due to divided government. This is consistent with Campbell, Cox, and McCubbins (2002) finding that divided government weakens majority party agenda control in the Senate, since Republicans held the house and the presidency during the post-switch Democratic majority. But, of course, the Jeffords switch research design that we use throughout this paper, though well designed for testing our hypotheses regarding the effects of majority status, is not well designed for testing hypotheses about variations in majority party power. We include this discussion merely as an aside, partly in hopes of prompting future research dealing with the questions we have raised. Conclusion In the literature on Senate decision making, the role of parties and the value of majority status are often murky. Yet, the preceding results are strikingly clear. We have presented a test of hypotheses predicated on the assumption that the Senate majority party 19 See Aldrich, Berger, and Rohde (2002) and Aldrich, Rohde, and Tofias (2005) on measuring the extent to which CPG conditions are met. 20

21 can systematically and significantly bias the Senate agenda to its advantage. Our findings unequivocally support these predictions. By focusing on the short time frame - just a few months on either side of the Jeffords' switch - our results give us a narrow but potent window to isolate the effect of a majority status "treatment" in the Senate. During our period of study, member preferences (including Jeffords preferences, as we have shown), the preferences of external actors, the legislative agenda, and numerous other potentially confounding factors remained constant. Yet, Democrats probability of being rolled was markedly worse than Republicans during their time as the majority party at the beginning of the 107th Congress, but those fortunes were reversed following the Jeffords switch. In contrast with some Senate literature claiming the contrary, it is clear that the Senate majority party exercises some level of influence over decision making within the chamber. By no means do we intend to suggest that the Senate majority party is akin to the strong, cohesive parties posited by the responsible-party Westminster ideal (Lijphart 1984). And, of course, none of the partisan theories of Congress would suggest this either. Rather, these theories posit that majority party leaders wield various powers in particular, powers to influence the agenda that allow them to manipulate outcomes. The results presented here seem to offer substantial evidence that the Senate majority party, through some means, manipulates outcomes for the benefit of its own members, and does so in ways that have often gone unrecognized by Senate scholarship. 21

22 References Aldrich, John, Mark M. Berger, and David W. Rohde The Historical Variability in Conditional Party Government, In David Brady and Mathew D. McCubbins, eds., Party, Process, and Political Change in Congress: New Perspectives on the History of Congress. Palo Alto: Stanford University Press. Aldrich, John, and David W. Rohde The Republican Revolution and the House Appropriations Committee. Journal of Politics 62: Aldrich, John, and David W. Rohde The Logic of Conditional Party Government: Revisiting the Electoral Connection. In Lawrence Dodd and Bruce Oppenheimer, eds., Congress Reconsidered, 6 th edition. Washington, DC: Congressional Quarterly Press. Aldrich, John, David W. Rohde, Michael W. Tofias One D is Not Enough: Measuring Conditional Party Government, In David Brady and Mathew D. McCubbins, eds., Process, Party and Policy Making: Further New Perspectives on the History of Congress. Palo Alto: Stanford University Press. Bargen, Andrew Senators, Status Quos, and Agenda Setting: A Spatial Story of Policy Making in the U.S. Senate, Paper presented at the Annual Meeting of the American Political Science Association Pennsylvania Convention Center, Philadelphia, PA, August Beth, Richard S., Valerie Heitshusen, Bill Heniff, and Elizabeth Rybicki "Leadership Tools for Managing the U.S. Senate." Paper presented at the Annual Meeting of the American Political Science Association. Toronto, ON, August 31, Binder, Sarah A Minority Rights, Majority Rule. New York: Cambridge University Press. Binder, Sarah A., and Steven S. Smith Politics or Principle? Filibustering in the United States Senate. Washington: Brookings Institution Press. Brady, David W. and Craig Volden Revolving Gridlock: Politics and Policy from Carter to Clinton. Boulder, CO.: Westview Press. Campbell, Andrea C., Gary W. Cox, and Mathew D. McCubbins Agenda Power in the U.S. Senate, 1877 to In David Brady and Mathew D. McCubbins, eds., Party, Process, and Political Change in Congress: New Perspectives on the History of Congress. Palo Alto: Stanford University Press. 22

23 Carroll, Royce, and Henry A. Kim Party Government and the "Cohesive Power of Public Plunder. American Journal of Political Science 54(1):pp Carson, Jamie L., Nathan W. Monroe, and Gregory Robinson Unpacking Agenda Control in Congress: Individual Roll Rates and the Republican Revolution. Political Research Quarterly 64(1) Chiou, Fang-Yi, and Lawrence S. Rothenberg When Pivotal Politics Meets Partisan Politics. American Journal of Political Science 47(3): Clinton, Joshua, Simon Jackman, and Douglas Rivers The Statistical Analysis of Roll-Call Data. American Political Science Review 98: Cox, Gary W., and Mathew D. McCubbins Legislative Leviathan: Party Government in the House. Berkeley: University of California Press. Cox, Gary W., and Mathew D. McCubbins Agenda Power in the U.S. House of Representatives, 1877 to In David Brady and Mathew D. McCubbins, eds., Party, Process, and Political Change in Congress: New Perspectives on the History of Congress. Palo Alto: Stanford University Press. Cox, Gary W., and Mathew D. McCubbins Setting the Agenda: Responsible Party Government in the US House of Representatives. Cambridge : Cambridge University Press. Daschle, Tom Like No Other Time: The 107 th Congress and the Two Years that Changed America Forever. New York: Crown. Den Hartog, Chris, and Nathan W. Monroe The Value of Majority Status: The Effect of Jeffords s Switch on Asset Prices of Republican and Democratic Firms. Legislative Studies Quarterly 33: Evans, C. Lawrence and Walter J. Oleszek Message Politics and Senate Procedure, in The Contentious Senate. Colton Campbell and Nicol Rae, eds. Landham, Maryland: Rowman and Littlefield. Evans, C. Lawrence, and Daniel Lipinski Obstruction and Leadership in the U.S. Senate. In Congress Reconsidered. 8 th ed., eds. Lawrence C. Dodd and Bruce I. Oppenheimer. Washington, DC: CQ Press. Gailmard, Sean and Jeffrey A. Jenkins Negative Agenda Control in the Senate and House: Fingerprints of Majority Party Power. The Journal of Politics 69:

24 Goodman, Craig, and Timothy P. Nokken Lame-Duck Legislators and Consideration of the Ship Subsidy Bill of American Politics Research. 32: Jayachandran, Seema, "The Jeffords Effect." Journal of Law and Economics, vol. 49(2), Jeffords, James M My Declaration of Independence. New York: Simon and Schuster. King, Gary Unifying Political Methodology: The Likelihood Theory of Statistical Inference. Cambridge; New York: Cambridge University Press. Koger, Gregory "Pivots For Sale: Endogenous Rules, Transaction Costs, and Pivotal Politics." Paper Presented at the Annual Meeting of the American Political Science Association, Chicago, IL. Koger, Gregory Filibustering: A Political History of Obstruction in the House and Senate. Chicago, IL.: University of Chicago Press. Krehbiel, Keith K Pivotal Politics. Chicago: University of Chicago Press. Lawrence, Eric D., Forrest Maltzman, and Steven S. Smith Changing Patterns of Party Effects in Congressional Voting. Paper presented at the 2005 Annual Meeting of the American Political Science Association, Washington, D.C., August 31-September 4, Lawrence, Eric D., Forrest Maltzman, and Steven S. Smith "Who Wins? Party Effects in Legislative Voting." Legislative Studies Quarterly 31 (1): Lee, Frances E Beyond Ideology: Politics, Principles, and Partisanship in the U.S. Senate. Chicago, IL: University of Chicago Press. Mann, Thomas E., and Norman J. Ornstein The Broken Branch: How Congress is Failing America and How to Get It Back on Track. New York: Oxford University Press. Martinez, Gebe "Moderates torn by parties; Senate defection shows conflict of middle ground." The Detroit News, May 27, Matthews, Donald R U. S. Senators and Their World. Chapel Hill: University of North Carolina Press. Nicholson, Stephen P The Jeffords Switch and Public Support for Divided Government. British Journal of Political Science 35:

25 Nokken, Timothy P., and Keith T. Poole Congressional Party Defection in American History. Legislative Studies Quarterly 29: Oleszek, Walter J Congressional Procedures and the Policy Process, 6 th ed. Washington. D.C.: Congressional Quarterly. Palmquist, Brad Analysis of Proportions Data. Unpublished typescript. Vanderbilt University. Poole, Keith T Recovering a Basic Space From A Set of Issue Scales. American Journal of Political Science 42 (3): Poole, Keith, Nolan McCarty, and Howard Rosenthal Income Redistribution and the Realignment of American Politics. Washington, D.C.: AEI Press. Poole, Keith T., and Howard Rosenthal Congress: A Political-Economic History of Roll Call Voting. New York: Oxford University Press. Ripley, Randall B Power in the Senate. New York: St. Martins. Roberts, Jason M The Statistical Analysis of Roll-Call Data: A Cautionary Tale. Legislative Studies Quarterly XXXII(3), Rohde, David W Parties and Leaders in the Postreform House. Chicago and London: University of Chicago Press. Rohde, David W Electoral Forces, Political Agendas, and Partisanship in the House and Senate. In Roger H. Davidson, ed., The Postreform Congress. New York : St. Martin s Press, pp Schiller, Wendy J The Art of Manipulation: The Use of Senate Parliamentary Procedure to Change Policy Outcomes. Paper presented at the Annual Meetings of the American Political Science Association, Chicago, August 31-September 3. Schiller, Wendy J Trent Lott s New Regime: Filling the Amendment Tree to Centralize Power in the U.S. Senate. Paper presented at the Annual Meeting of the American Political Science Association, August 31-September 3, Washington, DC. Schiller, Wendy J Majority and Minority Rights in the Senate and the Role of Party Leaders in Internal Governance. Paper presented at the 2001 Meeting of the American Political Science Association San Francisco, August 30-September 2. Sinclair, Barbara The Transformation of the U.S. Senate. Baltimore: Johns Hopkins University Press. 25

26 Sinclair, Barbara The Senate Leadership Dilemma: Passing Bills and Pursuing Partisan Advantage in a Nonmajoritarian Chamber. In The Contentious Senate. ed. Nicol C. Rae and Colton C. Campbell. Lanham, MD: Rowman & Littlefield. Smith, Steven S Party Influence in Congress. Cambridge, MA: Cambridge University Press. Tiefer, Charles Congressional Practice and Procedure: A Reference, Research, and Legislative Guide. New York; Westport, Connecticut; London: Greenwood. Wawro, Gregory J. and Eric Schickler Filibuster: Obstruction and Lawmaking in the U.S. Senate. Princeton, NJ: Princeton University Press. 26

27 Table 1. Effect of Jeffords switch on senators probability of being rolled Dem (Predicted to be positive) PostSwitch (Predicted to be positive) Coef, (SE), p-value 2.558*** (.139) *** (.157).000 Dem*PostSwitch *** (.198).000 Distance 2.153*** (.293).000 Constant *** (.171).000 γ -.020*** (.002).000 PostSwitch + Dem*PostSwitch (Predicted to be negative) Dem + Dem*PostSwitch (Predicted to be negative) *** (.121) ** (.142).029 N 198 Pseudo R Log-likelihood Extended beta binomial coefficients shown in each cell; p-values are for one-tailed tests. 27

28 Table 2. Estimated probability of a Senator being rolled on a final passage vote before and after Jeffords switch, by party Probability of being rolled on a final passage vote is: Member s party Pre-Jeffords Post-Jeffords Republican Democrat Distance it is set at the median value for members of each party. 28

29 Figure 1. Change in senators roll rates going from pre- to post-switch 29

30 Figure 2. Jeffords s estimated W-NOMINATE scores on all votes through the given date. 30