1. Related literature

Party leaders are thought to exert significant control over rank-and-file members by setting the agenda and by whipping their members through various carrots and sticks (Rohde, Reference Rohde1991; Aldrich, Reference Aldrich1995; Cox and McCubbins, Reference Cox and McCubbins2007; Smith, Reference Smith2007; Evans, Reference Evans2018). Party leaders might also influence rank-and-file members through their informational or resource advantages (e.g., Krehbiel, Reference Krehbiel1992; Zelizer, Reference Zelizer2019) or by establishing an informal relationship of reciprocity (Fong, Reference Fong2023). Party leaders control resources, which may allow them to buy votes (Groseclose and Snyder, Reference Groseclose and Snyder1996; Banks, Reference Banks2000), and they have proposal power, which may allow them to offer their members just enough to win their support (Baron and Ferejohn, Reference Baron and Ferejohn1989; Ali et al., Reference Ali, Bernheim and Fan2019) while retaining a large share of rents for themselves.

The theory of conditional party government specifically posits that polarization and party effects go hand in hand. As the parties become more polarized, they are increasingly willing to hand over more power to party leaders, who in turn exercise that power, which further exacerbates polarization.Footnote ¹ Of course, the potentially cyclical relationship between party leaders and the polarization of rank-and-file members makes it difficult to empirically assess party effects. Krehbiel (Reference Krehbiel1993) provocatively asks, “Where’s the party?” and argues that although there could be large party effects, congressional polarization does not constitute convincing evidence of them.

Some researchers have taken up Krehbiel’s challenge and attempted to distinguish party effects from selection into parties (see Smith, Reference Smith2007 for a more thorough review of this literature). Nokken and Poole (Reference Nokken and Poole2004) find that the same members vote differently when they switch parties. Jenkins (Reference Jenkins1999) finds that the same members voted differently in the U.S. House than in the Confederate House, which did not have parties. Ansolabehere et al. (Reference Ansolabehere, Snyder and Charles2001) find evidence of party effects even after controlling for the stated preferences of members of Congress in surveys. Snyder and Groseclose (Reference Snyder and Groseclose2000) find evidence of party effects on close roll-call votes, even after controlling for the voting behavior of members on non-close votes. Canen et al. (Reference Canen, Kendall and Trebbi2020) find evidence of party effects using a structural model that exploits the fact that leaders from different parties sometimes vote in the same direction.Footnote ²

Although these studies are innovative and informative, concerns remain about whether these designs convincingly isolate the effects of party discipline (Krehbiel, Reference Krehbiel2003). Members likely switch parties as their constituencies are changing and as they were planning to change their roll-call voting anyway. Voting in the Confederate House may have differed from that in the U.S. House for additional reasons beyond parties. Surveys of candidates may not adequately capture all ideological differences between legislators from different parties. Bills with close votes are surely different from bills with lopsided votes. Bills over which the opposing party leaders agree are surely different from bills over which they disagree. Summarizing this literature and making a prediction about its future, Smith (Reference Smith2007) writes, “The search for direct party effects will prove frustrating” (p. 7).

2. Why party leaders?

Why do legislators form or join parties to begin with? Wouldn’t they prefer to operate independently, draft their own proposals, and decide for themselves how they vote on each bill? One common explanation for legislative parties is that they help members solve collective action problems and achieve better outcomes than they would otherwise achieve without cooperating (e.g., Aldrich Reference Aldrich1995; Schwartz, Reference Schwartz2021). Imagine two legislators voting on two bills that require support from both of them to pass. For each bill, one legislator is strongly in favor and the other is weakly opposed, but the legislator in favor differs between the two bills. If the members vote on each bill separately, both bills will fail. But if they could agree to cooperate and pass both bills, both legislators would be better off.Footnote ³ So if they agree to vote together, they can achieve this more preferable outcome.

Although scholars often think of legislative ideology as being largely one-dimensional, these opportunities for vote-trading surely arise frequently. One obvious example is pork-barrel spending. Each legislator might strongly prefer spending in their constituency but weakly oppose it in other constituencies, so by logrolling, legislators can pass spending projects in all of their constituencies and be better off. This logic likely arises in other policy areas as well. Suppose a slim majority of legislators would like to pass a set of new laws. On any given bill, a few members of the group would prefer to vote no for idiosyncratic reasons, so it wouldn’t pass on an independent vote. But if this group decides to tie their hands and support all of the group’s bills, they will all be better off. The same logic would apply to a slim minority trying to block the proposals of others in the legislature. So these groups form or join parties to tie their hands and help them to achieve more preferable outcomes.

How do parties tie the hands of their members in practice? The two legislators in the toy example above might promise to cooperate, but left to her own devices, the beneficiary of cooperation on the first bill would want to vote no on the second bill. Anticipating this, the other legislator votes no on the first bill, and the logroll unravels. To solve this dilemma, legislators elect a party leader and give her the resources and authority to reward or punish rank-and-file members (e.g., Calvert, Reference Calvert1987). Among other things, the party leader might control campaign funds and committee assignments, which provides an incentive for members to cooperate with the party on bills deemed important by the leader.

How do party leaders behave when given this kind of power? Among other things, it likely depends on their ideology. More extreme leaders will want to whip their members on a larger set of bills to pass more (and more extreme) legislation or to block more (and more moderate) legislation proposed by the other party. Even if leaders know there is a trade-off between achieving their policy goals today and their probability of winning the majority in the next election as in the classic Calvert–Wittman model (Wittman, Reference Wittman1983; Calvert, Reference Calvert1985), more extreme leaders will likely make that trade-off in different ways depending on their ideology.

Anticipating this, who will rank-and-file party members select as their leader? Each rank-and-file member will want a leader who is extreme enough to impose party discipline on the bills important to them (or block the undesirable policy goals of the other party), but not too extreme that they have to support (oppose) too many bills that they don’t (do) like. And by the same Calvert–Wittman logic, each rank-and-file member faces a trade-off between policy victories today and their chances of electoral victory in the future, and depending on their ideology and that of their constituents, different members will make that tradeoff differently.Footnote ⁴

This theory of legislative leadership yields several testable predictions:

1. 1. Party leaders will tend to be more extreme than rank-and-file members because more extreme leaders can credibly commit to disciplining the party on a larger number of bills. Consider the two legislators in the toy example. They would prefer a party leader who supports both bills, so they would support a leader who is more extreme than either one of them.

2. 2. Parties will be more likely to select relatively moderate leaders when they are a small minority because, first, the benefits of discipline are muted when the party has too few members to pass or block legislation, and second, the incentive to win more seats is greater.

3. 3. More ideologically extreme party leaders should whip their members on more bills, generating more partisan and more ideologically extreme votes.

4. 4. Ideologically moderate members (who are most willing to defect in the absence of party discipline) should be affected more by their leader.

5. 5. Since a key function of party leaders is to tie the hands of rank-and-file members, leaders should influence the roll-call voting of even the rank-and-file members who disagree with them and opposed their bid to become party leader.

6. 6. Party leaders have an incentive to whip their members to support bills that the other party opposes or oppose bills that the other party supports, so the presence of a party leader should, on net, increase the extent to which members cast partisan or ideologically extreme votes.

3. Descriptive statistics

Before proceeding, I first show descriptive statistics that motivate the main analyses of the paper. Figure 1 shows estimates of the policy positions of congressional leaders and their rank-and-file members from the 76th through the 117th Congresses, 1939–2022. Following the literature, I measure the policy positions of leaders and rank-and-file members using their roll-call votes. For a simple and substantively interpretable measure of congressional roll-call behavior, I utilize Conservative Vote Probabilities (CVP), developed by Fowler and Hall (Reference Fowler and Hall2013) and used in many applied papers (e.g., Alexander et al., Reference Alexander, Berry and Howell2015; Feigenbaum and Hall, Reference Feigenbaum and Hall2015; Fowler and Hall, Reference Fowler and Hall2016; Fowler, Reference Fowler2020). CVP scores indicate how often an individual member of Congress casts a conservative roll-call vote relative to the median member of the same chamber in the same period. These scores account for the fact that not every member casts a vote on every bill and conservative votes are more common on some bills than others. To enable comparability over time, I adjust the CVP scores using the method of Groseclose et al. (Reference Groseclose, Levitt and Snyder1999), and I index the scores to the 117th Congress.

Figure 1.

Polarization of Party Leaders and Rank-and-File Members over Time. CVP stands for Conservative Vote Probabilities, which are adjusted using the method of Groseclose et al. (Reference Groseclose, Levitt and Snyder1999) and indexed to the 117th Congress. Scores for Republicans are shown in red, and scores for Democrats are shown in blue. The solid curves show the average scores for non-leaders. Dashed lines show linear fits before and after the 92nd Congress. Dots show the average score of the current party leader from the time before they became a party leader.

The solid curves in Figure 1 show the average adjusted CVP scores for Democratic (blue) and Republican (red) members, excluding the leaders of each party in each chamber from these averages. The dashed lines show separate linear trends before and after the 92nd Congress, which is approximately the lowest point of polarization. The dots show the average CVP score for the leader of each party in each chamber. Specifically, the Speaker of the House and the House Minority Leader are classified as the leaders of their respective parties in the House, and the Senate majority and minority leaders are classified as the leaders of their respective parties in the Senate. When measuring the positions of party leaders, I focus on their average CVP scores from the time before they became party leaders because some party leaders cast few roll-call votes, party leaders sometimes vote insincerely for procedural reasons,Footnote ⁵ and if becoming a party leader causes one to cast more partisan votes, I do not want to include that effect in this measure.

Some of the patterns in Figure 1 are familiar (e.g., see McCarty, Reference McCarty2019). Polarization in Congress, as measured by the difference between the average rank-and-file Republican and the average rank-and-file Democrat, reached a low point around 1970 and has been increasing since. However, this pattern is more pronounced in the House than in the Senate. The most striking anomaly in the figure is that the Republicans in the House were unusually moderate in the 1960s and 1970s. If one could explain why the House Republicans were unusually moderate during this period, they would explain a large share of the changes in polarization observed over the past century.

Figure 1 also shows, consistent with the observations of Harris and Nelson (Reference Harris and Nelson2008) and the first theoretical prediction above, that party leaders tend to be more extreme than rank-and-file members, and this is especially true in the House. Recall that my measure of party leaders is based on their roll-call voting before they were party leaders, so the figure essentially shows that extreme members of a party are more likely to be later selected as party leaders.Footnote ⁶ We also see an apparent correspondence between the extremism of the party leaders and the extremism of the rank-and-file members of a party, and subsequent analyses are intended to assess how much of this correspondence is attributable to the selection of party leaders versus the effects of party leaders.

Lastly, notice that the anomalous period of Republican moderation coincided with unusually moderate Republican Party leaders. Specifically, Gerald Ford and John Rhodes who served as House Minority Leaders from 1965 through 1980 were the most moderate Republican leaders in the modern history of the House, and Hugh Scott who served as Senate Minority Leader from 1969 through 1976 was the most moderate Republican leader in the modern history of the Senate.Footnote ⁷ Consistent with the second theoretical prediction above, these unusually moderate leaders were selected at times when their parties held a clear minority in their chamber. If party leaders strongly influence the voting behavior of their rank-and-file members, then these few individuals, along with the decisions of the party members to select these leaders, could explain a meaningful share of the decline and rise of political polarization in Congress.

4. Effects of party leaders on the roll-call voting of rank-and-file members

To assess the effects of party leaders on rank-and-file members, I leverage changes in party leadership. Although Figure 1 shows that party leaders tend to be more extreme than rank-and-file members, there have been notable instances when a party switched from a relatively extreme leader to a relatively moderate leader or vice versa. Do rank-and-file members change their roll-call behavior when the ideology of the party leader changes?

To answer this question, I analyze roll-call voting data from the 78th through the 116th Congresses, 1941–2020. To measure the ideology of congressional leaders, I utilize the same measure used in Figure 1, the average adjusted CVP score of a congressional leader from the period before they became a leader. This is the treatment variable of which I am hoping to estimate the effect.

My units of observation are roll-call votes taken by rank-and-file members. I classify each vote as conservative or liberal based on how the votes correlate with CVP scores. Specifically, if voting yea is positively correlated with CVP scores, I classify the yea vote as conservative, and if voting yea is negatively correlated with CVP scores, I classify the nay vote as conservative.Footnote ⁸ I exclude abstentions, and I drop all bills for which the rank-and-file members from both major parties voted unanimously in favor or in opposition. This leaves approximately 14 million member-votes.

I assess the effects of party-leaders by estimating regressions of the following form:

\begin{equation*}Conservative\,Vot{e_{ipbct}} = \beta *Leader\,CV{P_{pct}} + {\gamma _{ipc}} + {\delta _{bct}} + {\varepsilon _{ipbct}},\end{equation*}

where $Conservative\,Vot{e_{ipbct}}$ is in indicator for whether the rank-and-file legislator i from party p cast a conservative vote on bill b in chamber c and Congress t. $Leader\,CV{P_{pct}}$ represents the average adjusted CVP score of the legislator’s party leader from the period before that person became a party leader, ${\gamma _{ipc}}$ represents legislator-party-chamber fixed effects, and ${\delta _{bct}}$ represents bill-chamber-Congress fixed effects. The fixed effects implicitly make this a differences-in-differences design. Any idiosyncratic differences between different legislators, bills, chambers, or time periods are accounted for, and identification comes from within-member-party-chamber changes in the ideology of party leaders.

In subsequent analyses, I utilize three additional methods for classifying votes as more conservative or more Republican. I code an indicator variable for whether a member’s vote on a particular bill was more likely to have been cast by a Republican. For example, if Republicans were more likely to cast a yea vote than Democrats, I classify the yea vote as being in the Republican direction. I also code indicators for whether a member voted with the majority of Republicans or against the majority of Democrats. These four measures are all positively correlated with one another,Footnote ⁹ but by comparing results across these different measures, we can assess whether leaders are more likely to influence the extent to which members vote with their party rather than voting in a particular ideological direction.

The key identifying assumption of this design is parallel trends. Specifically, I assume that members who experience a change in the ideology of their party leader would, in expectation, experience the same change in their roll-call voting as members who did not experience a leader change had there been no change in leader ideology. The easiest way to think about this assumption is to think about the case in which one party selects a new leader (with a different ideology) while the other party does not. My parallel trends assumption is that members of the two parties would have, in expectation, shifted their roll-call voting in the same way before and after this change if not for any change in leaders. In other words, I assume that changes in the ideology of a party leader do not systematically occur at times when rank-and-file members from that party and chamber would have systematically shifted their roll-call voting relative to members of the other party in the absence of any change in party leaders.

The most likely violation of parallel trends arises from the selection of party leaders. Perhaps rank-and-file members elect a more conservative (liberal) leader at the same time that they would have otherwise shifted their voting behavior in a conservative (liberal) way. Another potential violation of parallel trends would arise if rank-and-file members and leaders are polarizing at the same time for reasons unrelated to the effects of leaders. I attempt to assess and rule out these concerns in several different ways. In some specifications, I also include, as an additional independent variable, the ideology of the leader in the next Congress. This allows me to test whether members of a party were already trending in the direction of their new leader before they selected the leader. Reassuringly, I find little evidence of this. If rank-and-file members tend to select more extreme leaders as they are becoming more extreme, we should find that the ideology of the leader in the next Congress appears to influence the voting of rank-and-file members in the current Congress, but we do not.

In a subsequent section, I exploit a unique historical episode to test whether party leaders appear to influence the roll-call votes of members who did not support their bid to become a party leader. I find that even members who voted against the new leader are still influenced by the change in leadership. Therefore, within-member changes in roll-call voting following a new leader are not solely attributable to demand for change among the rank-and-file members.

Furthermore, previous evidence suggests that in the absence of leader effects, members of Congress do not tend to meaningfully change their roll-call voting over time. For example, Moskowitz et al. (Reference Moskowitz, Rogowski and Snyder2024) find that members’ responses to candidate surveys do not change meaningfully over time, and virtually all of the observed polarization is attributable to replacement rather than within-member changes. Therefore, within-member changes in ideology are unlikely to explain changes in the ideology of the leader. Changes in party leadership more likely coincide with changes in the composition of the chamber, which does not bias my results since I focus on within-member changes, or other idiosyncratic factors, some of which I discuss in a later section.

I also modify the design to test for heterogeneous effects. To test whether the effect of party leaders is immediate or takes some time to manifest, I include, as an additional independent variable, the ideology of the leader in the previous Congress. If there is an effect of party leaders, there might be reasons to expect that the effect is not instantaneous. Perhaps party leaders learn on the job and get better over time at influencing the behavior of their members. Perhaps rank-and-file members will only vote in line with their leaders after the leader has proven to be competent, trustworthy, or politically skilled. In other specifications, I include additional interactive variables to test for heterogeneous effects by party, chamber, or the ideology of members.

This design estimates the extent to which the ideology of a party leader influences the rank-and-file voting of members who were already in office. The motivation for such an analysis is that to the extent that party leaders influence the roll-call voting of rank-and-file members, we would expect more conservative (liberal) leaders to cause more conservative (liberal) roll-call voting.Footnote ¹⁰ Leaders likely have additional effects beyond the scope of this analysis. For example, some leaders may influence of the composition of their party-chamber by taking positions that influence the overall electoral fortunes of their party or by strategically doling out campaign funds to favored members of the party. We might also expect leaders to particularly influence the behavior of new members who joined Congress while the leader was already in power. For these reasons, the subsequent estimates likely understate the overall effect of party leaders on the behavior of the party’s rank-and-file members.

One mechanism through which party leaders can affect policy and roll-call voting is by influencing the agenda. However, I do not believe agenda setting is a key explanation for my subsequent results. First, I find that leader effects are just as great for the minority party, where the leader does not have agenda-setting power. Second, if there is an effect of agenda setting in this context, the direction of the effect is theoretically ambiguous. Consider an extreme Speaker of the House whose party holds a large majority. He or she might propose extreme policies that will just barely pass (Krehbiel, Reference Krehbiel1998), thus dividing her party and paradoxically making some of her rank-and-file members appear to be more moderate than they are. Unfortunately, my empirical approach, along with virtually all analyses of roll-call voting, is not well suited for measuring the extent to which leaders influence policy by strategically shaping the agenda.

The results of these analyses are shown in Table 1. All specifications include member-party-chamber fixed effects and bill-chamber-Congress fixed effects. First, let us focus on the top panel in which the dependent variable is an indicator for a conservative vote. In my baseline specification in Column 1, I estimate a coefficient of .389, implying that if the conservatism of a party leader increases by some amount, rank-and-file members from that leader’s chamber and party increase the conservatism of their voting by 38.9 percent of that amount.Footnote ¹¹ The coefficients of interest in the top panel of Table 1 can be interpreted in this way because the dependent and independent variables are measured on essentially the same scale.

Table 1.

Effects of party leaders on rank-and-file members, 1941–2020

Standard errors, two-way clustered by member-party-chamber and party-chamber-Congress, in parentheses; ** p < .01, * p < .05.

For my period of analysis, the typical range of leader CVP scores within a particular party and chamber is approximately .15. Therefore, the estimate in Column 1 of Table implies that if a party-chamber switched from the most moderate to the most extreme leader in its history, we would expect rank-and-file members to become approximately 6 percentage points more likely to cast ideologically extreme roll-call votes. To further interpret the substantive magnitude of this estimate, Figure 1 shows that the average member of Congress was approximately 12 percentage points more likely to cast an ideologically extreme vote in the 117th Congress—a recent period with relatively extreme leaders—than in the 92nd Congress—a period with relatively moderate leaders. Therefore, the increase in leader extremism combined with the estimated effects of leaders on rank-and-file members can explain approximately half of the increase in congressional polarization over the past 50 years.

In Column 2 of Table 1, I also include leading values of leader ideology to test for pre-trends. The coefficient associated with the leading value of the leader’s ideology is imprecisely estimated, so I cannot rule out the possibility of modest pre-trends. But the estimated coefficient is substantively small and statistically insignificant, suggesting that rank-and-file members were typically not strongly trending in a conservative (liberal) direction before selecting a more conservative (liberal) leader. The coefficient associated with the current value is largely unchanged. These results lend additional credibility to the parallel trends assumption, and they show that even if we allowed for pre-trends, the estimated effect of leader ideology on rank-and-file voting would be largely unchanged.

In Column 3, I also test for lagged effects of leader ideology. I find that there is little effect of the leader’s ideology in their first Congress as leader, but there is a substantively and statistically significant lagged effect. Perhaps leaders learn over time how to influence their rank-and-file members or perhaps members have to build trust with their leader before they change their voting behavior. Therefore, the average effect identified in the baseline specification likely arises from the combination of a small or null effect in a leader’s first Congress and a larger effect in subsequent Congresses.

In Columns 4–7, I test for heterogeneity across chamber, party, and the ideology of members. In Column 4, I find that the effects of party leaders are greater in the House than in the Senate. Specifically, I estimate a statistically significant effect of .161 in the Senate, and I also estimate that the effect is .443 greater in the House than in the Senate, and this difference is statistically significant. Therefore, focusing just on the U.S. House, I estimate an effect of approximately .6, meaning that for every change in leader ideology in the House, we would expect rank-and-file members to shift their voting behavior by more than half of that change.

In Column 5, I find suggestive evidence that leader effects are larger for Republicans than for Democrats, although this difference is not statistically distinguishable from zero. In Column 6, I estimate little difference in leader effects between the majority and minority parties. This suggests that to the extent that I detect leader effects, they are not likely explained by the ability of majority party leaders to set the agenda.

Lastly, In Column 7, I test for heterogeneity across the ideology of members. A canonical spatial model would predict that moderates will be easier to influence since they are typically closer to being indifferent between supporting or proposing a bill, and this was the basis for the fourth theoretical prediction mentioned above. Alternatively, Minozzi and Volden (Reference Minozzi and Volden2013) predict the greatest effects for extremists who benefit the most from a common party position. I classify each member as moderate or extreme relative to the other rank-and-file members with whom they typically served. To do this, I compute their adjusted CVP score from the first Congress in which they served. Then, I compute the median of these scores across all members who served in each Congress-chamber-party. If a member’s first CVP score was more moderate than that of their party-chamber median in more than half of the Congresses in which they served, I classify them as moderate. I find substantively and statistically significant leader effects for both moderate and extreme members, but I find that leader effects are greater for moderate members, and this difference is statistically significant.

In the subsequent panels of Table 1, I show the same analyses but using the three additional dependent variables described previously. The pattern of results is strikingly similar across all four measures of roll-call voting. Interestingly, the effect of leader ideology on party-line voting is even greater than the previously discussed effects on the ideological direction of voting. I estimate coefficients of .492 and .491 when examining whether members voted with the majority or Republicans or against the majority of Democrats, respectively, compared to coefficients of .389 and .385 when examining whether members votes in a conservative or Republican direction. This suggests that to the extent that extreme leaders influence the roll-call votes of rank-and-file members, they are more likely to induce members to vote with the party rather than induce members to vote in an extreme direction per se.

The previously discussed finding that leader effects are significantly greater for moderate members only applies to analyses of voting in a conservative or Republican direction. I do not replicate this result when analyzing voting with the majority of Republicans or against the minority of Democrats. The estimated effects are slightly greater for moderate members, but this difference is not statistically significant. This pattern of results is consistent with the idea that extreme leaders induce their members to vote with the party’s majority more so than they induce them to vote in an extreme direction. For relatively moderate members, voting with the party more often typically means casting more ideologically extreme votes. But for relatively extreme members, voting with the party more often may, in some cases, mean casting fewer ideologically extreme votes. So although extreme leaders shift the ideological direction of voting more for their moderate members, they might be comparably effective in inducing moderate and extreme members to vote more with the party majority.

Although voting with the majority of Republicans and voting against the majority of Democrats are not strongly correlated with one another (r = .30), these two dependent variables yield similar results across the analyses in Table 1. Although the majorities of the two parties often vote in the same direction, party leaders appear to exert pressure on bills in which the majorities of the two parties are voting differently—the cases in which these two variables are the same. Even when testing for heterogeneity by party, I still obtain similar results from either dependent variable. The estimated effect of leaders on roll-call voting is slightly greater for Republicans using either measure, although the difference is not statistically significant.

5. Extensions

One potential concern with this analysis is that I use previous roll-call votes of leaders (before they became leaders) to measure their ideology. Of course, if party leaders affect the roll-call votes of rank-and-file members, my estimates of leader ideology could be biased by the ideology of the leader in place when they cast those votes. To address this issue, I have implemented an iterative version of my analysis in which I estimate the effect of party leaders, adjust the estimated ideology of each future leader by accounting for the ideology of the leader and the estimated effect of leaders, and then repeat this process until the estimates converge. When I do this, I estimate an average effect of leaders on conservative voting of .340, similar to but slightly smaller than the estimate of .389 shown in Table 1. Therefore, the implications of this measurement challenge appear minimal for the main results of the paper. More details on this analysis are shown in the Appendix.

My main analyses focus on the highest ranking leader from each party and chamber—the Speaker of the House, House Minority Leader, Senate Majority Leader, and Senate Minority Leader. I extend this analysis to also estimate the effects of the next highest ranking party leaders, specifically the House Majority Leader and party whips. On one hand, we might expect the party whips to be particularly influential since their job titles imply that they will ensure that rank-and-file members vote the way the party prefers on important legislation. On the other hand, we might expect that these lower ranking leaders will largely take instructions from the highest ranking party leaders, in which case, even if these lower level leaders do a lot of work, it might be the highest ranking leaders’ positions that influence roll-call votes. When I simultaneously estimate the effects of the highest ranking leaders and these other leaders, I find that the estimated effects of the highest ranking leaders are essentially unchanged, and the effects of other leaders are substantively small and indistinguishable from zero. More details are in the Appendix.

6. What happened with the republicans in the 1960s and 1970s?

Three Republican leaders stand out in Figure 1, and they likely contribute significantly to the estimates in Table 1 because they were so much more moderate than the Republican leaders who served before or after them. They are also interesting because their leadership coincides with a historic period of Republican moderation and low polarization. Gerald Ford replaced Charles Halleck as House Minority Leader in 1965 and was notably more moderate than any previous Republican leader in the House. When Ford became Vice President and soon after President, he was replaced by the comparably moderate John Rhodes. When Rhodes retired, he was replaced by the more typically conservative Bob Michel in 1981. Similarly, Hugh Scott replaced Everett Dirksen as Senate Minority Leader in 1969 and was more moderate than any previous Republican leader in the Senate. When Scott retired, he was replaced by the more conservative Howard Baker in 1977.

If not for the moderating influence of Ford, Rhodes, and Scott, we might have a very different historical impression of congressional polarization. The Democratic Party controlled both chambers of Congress during this period, but it was internally divided on major initiatives such as civil rights legislation—see Peabody (Reference Peabody1976) for a more detailed discussion of congressional politics in this era. Other Republican leaders had pursued a ‘Southern strategy’ and largely opposed civil rights, but Ford, Rhodes, and Scott supported many civil rights initiatives and allowed their members to do so. Furthermore, these minority party leaders teamed up with Democrats to pass major reforms and initiatives during the Nixon administration, such as the creation of the EPA and OSHA.

Why did Republican members of Congress select these relatively moderate leaders at this time? Republicans suffered a major electoral loss in 1964, which harmed the reputation of House Minority Leader Charles Halleck, who was viewed by Republican members as simply opposing Democrats rather than putting forward new ideas (Green and Harris, Reference Green and Harris2019).Footnote ¹² Gerald Ford was perceived as not especially extreme or moderate, which made him a reasonable compromise candidate, and he was younger, well-liked, and eager to improve the electoral standing of the party. In the contest for minority leader, Ford narrowly defeated Halleck with 73 votes to 67. Green and Harris (Reference Green and Harris2019), analyzing data from internal tallies, found that Republican members who were from Michigan, on a committee with Ford, who were electorally less secure, who were younger, and who were more moderate were more likely to vote for Ford.

This account suggests that the forces that led to Ford’s election were somewhat idiosyncratic, and some of Ford’s success may be attributable to the fact that many Republicans didn’t appreciate just how moderate he was—at least relative to previous party leaders. However, consistent with the second theoretical prediction above, a big part of the story is surely that Republicans were in electoral trouble—they had won less than a third of the seats in the 1964 House elections, which was enough of an impetus for them to change course and elect a more moderate leader.Footnote ¹³

7. Do leaders influence members who do not support them?

To investigate the mechanisms through which party leaders influence rank-and-file members and to address an alternative explanation for my estimates of party leader effects, I estimate the effect of the Halleck-Ford transition on the roll-call voting of rank-and-file Republicans in the House. Utilizing data from Green and Harris (Reference Green and Harris2019) on which members supported Ford versus Halleck in the 1965 leadership contest, I test whether the effects of the leadership transition were greater for those who supported or opposed Ford.

To do this, I focus on roll-call votes taken by rank-and-file members of the U.S. House between the 87th and 91st Congress. This is the longest period during which the only leader transition in either party was from Halleck to Ford. I include only members who served in both the 88th and 89th Congresses, meaning they served during the leadership of both Halleck and Ford. Table 2 shows the results of these analyses. I code an indicator for Ford’s leadership, which takes a value of 1 for Republicans in the 89^th Congress and onward and 0 for Democrats and for Republicans before the 89th Congress. Following the approach in Table 1, I utilize four different dependent variables indicating a conservative or Republican vote, and all regressions include member-party fixed effects and bill-Congress fixed effects.

Table 2.

Effects of ford leadership on opponents and supporters

Standard errors, two-way clustered by member-party and party-Congress, in parentheses; **p < .01, *p < .05. The analyses include rank-and-file members of the House who served in both the 88th and 89th Congresses. “Ford leadership” is an indicator for Republicans during the period during which Ford was the House minority leader (89th Congress and onward). “Ford supporter” is an indicator for whether a member supported Ford in the internal election between Halleck and Ford. Columns 2 and 5 only classify those as Ford supporters for whom the historical record provides clear information, and Columns 3 and 6 also include those who leaned toward Ford but for whom the historical record is less clear. Columns 1–3 include the 87th–91st Congresses, the longest period during which the Halleck–Ford transition was the only leader transition in either party. Columns 4–6 include the 88th–89th Congresses, the immediate Congresses before and after the Halleck–Ford transition.

In Column 1 of Table 2, I estimate that the transition from Halleck to Ford caused Republicans to decrease their probability of casting conservative votes by 13.7 percentage points, and the estimates are similar for the three alternative dependent variables. In Columns 2 and 3, I also interact the indicator for Ford’s leadership with an indicator for whether a member was a Republican and supported Ford in the contest between Ford and Halleck. These votes were not public, so we do not know with certainty which members supported Ford, but I utilize data from Green and Harris (Reference Green and Harris2019) who infer member’s votes from Donald Rumsfeld’s whip sheet and from interviews conducted by Peabody (Reference Peabody1976). Column 2 presents results using a more stringent measure that only classifies members as Ford supporters if Green and Harris are highly confident, and Column 3 uses a less stringent measure that also classifies members as Ford supporters if they leaned toward Ford on the whip sheet or in interviews. In these regressions, the coefficient associated with Ford’s leadership can be interpreted as the effect of the Ford–Halleck transition for rank-and-file Republicans who supported Halleck, and the interactive coefficient can be interpreted as the difference in the effect for Republicans who supported Ford versus those who supported Halleck. Consistent with the fifth theoretical prediction above, the estimates in Columns 2 and 3 suggest that even the Republicans who supported Halleck were influenced by the leadership transition. And the estimated effect of the leadership transition is not significantly different for those who supported Ford.

Columns 4–5 repeat the analyses in Columns 1–3 but include only the immediate Congresses before and after the transition from Halleck to Ford (88–89th). Consistent with my finding in Column 3 of Table 1 that the effects of leader transitions are somewhat delayed, I find a smaller effect of the transition when focusing just on the first Congress of Ford’s leadership. Nevertheless, the pattern of results is similar. I find substantively meaningful and statistically significant effects for those who supported Halleck, and the effect is not significantly greater for those who supported Ford.

The analyses in Table 2 show that the transition from Halleck to Ford caused Republican members to cast fewer conservative or party-line votes. They do not necessarily tell us whether Halleck whipped members in a conservative direction, Ford whipped them in a liberal direction, or a combination of the two. The next section provides evidence on this question. If we think Ford whipped in a liberal direction, then the results of Table 2 suggest that his whipping was effective even among those who didn’t want him to be their leader. If we think Halleck whipped in a conservative direction—which is more consistent with the theory of party leaders laid out in this paper, then the results of Table 2 suggest that Halleck’s whipping was effective even among those who didn’t want him to be their leader and later supported Ford. Either way, the results suggest that party leaders successfully tie the hands of their members, influencing even those who are not aligned with them.

8. What happens when there is no party leader?

Qualitative accounts from the time suggest that any moderating effect of Ford might be attributable to the fact that he whipped his members less aggressively than other Republican leaders. Indeed, party unity voting was especially low during the leadership of Ford, Rhodes, and Scott (see Crespin et al., Reference Crespin, Rohde and Wielen2011). Peabody (Reference Peabody1976, pp. 104–105) quotes two members of the Wednesday Club, a group of younger, moderate House Republicans, who report supporting Ford over Halleck in 1965 because Halleck had worked too hard to achieve near unanimity among House Republicans, forcing moderate members to take positions that were unpopular in their districts. They felt that Halleck often pressured them to vote no on Democratic proposals not because they were bad policy but simply because he wanted to oppose the other party.

To further investigate this question, I looked for historical periods during which roll-call votes were cast while one party had no leader. I found only one such instance in the modern era of Congress, but fortuitously, it happens to be an interesting one from the perspective of this study. Everett Dirksen died in office on September 7, 1969, and Hugh Scott was elected to replace him on September 24. In the few weeks during which the minority party had no leader, the Senate voted on nine bills.

I estimate the effect of Scott’s leadership relative to Dirksen, and perhaps more interestingly, I also estimate the effect of having no leader relative to Dirksen. I exclude all party leaders from this analysis, and I analyze each rank-and-file U.S. Senator’s votes across all bills in the 91st Congress. I regress conservative voting on an indicator for Republicans during the period in which they had no leader, an indicator for Republicans during the period of Hugh Scott’s leadership, member fixed effects, and bill fixed effects.

Table 3 presents the results of this analysis. Consistent with the sixth theoretical prediction above, I find that Republicans were 10.1 percentage points less likely to cast a conservative vote, 11.0 percentage points less likely to vote in the Republican direction, 14.5 percentage points less likely to vote with the majority of Republicans, and 21.3 percentage points less likely to vote against the majority of Democrats when there was no party leader relative to when Dirksen was their leader. Although these estimates are imprecise because there were only 9 votes taken with no leader, two out of those four estimates are statistically significant. I also find that Republicans were somewhat less likely to cast conservative or Republican votes under Scott than under Dirksen, but these estimates are not statistically significant, and they are substantively much smaller than the estimated effect of having no leader.

Table 3.

Effect of no republican senate leader in the 91st congress

Standard errors, two-way clustered by member and bill, in parentheses; ** p < .01, * p < .05. Each observation is a rank-and-file U.S. Senator by bill in the 91st Congress. “No leader” is an indicator for Republicans during the period in which there was no minority leader. “Scott leadership” is an indicator for Republicans during the period in which Hugh Scott was the minority leader.

The results in Table 3 further reassure us against the concern that our estimates of party leader effects are biased because rank-and-file members endogenously elect a new leader precisely when they otherwise plan to change course. When Dirksen unexpectedly died in office, rank-and-file Republican senators quickly cast fewer conservative or Republican votes. When Scott was elected to replace Dirksen, the same members voted more conservatively, but not as conservatively as they did under Dirksen. So even in the rare case when exogenous events cause a within-Congress leadership change, rank-and-file members change their roll-call voting accordingly.

The results in this section suggest that—left to their own devices—members of Congress would vote against their party more often, but party leaders prevent them from doing so. These results also suggest that to the extent that leaders like Ford, Rhodes, and Scott had a moderating effect on their parties, it was not because they persuaded them to cast liberal votes that they didn’t otherwise want to take, rather, they were more willing than other leaders to allow members to vote in a moderate way.

9. Discussion and conclusion

This study contributes to the debate about party discipline in Congress. Individual members of Congress change their roll-call voting in response to the ideology of their party leader. Therefore, at least in some cases, parties and leaders do meaningfully influence how their members vote. On average, I find that when the ideology of a party leader changes, rank-and-file members who were already serving will shift their voting behavior by 40 percent of that change. This effect is less than the hypothetical effect of 100 percent that we would expect if members naively mimic the positions of their leader, but it’s large relative to other factors thought to influence the roll-call votes of members of Congress. For example, studies typically find small effects of even electoral incentives on the roll-call voting (Poole, Reference Poole2005; Fowler and Hall, Reference Fowler and Hall2016; Stone, Reference Stone1980; Fowler, Reference Fowler2024; Fowler and Fu, Reference Fowler and FuForthcoming; Kim, Reference Kim2025), so party leaders might be one of the most significant factors affecting roll-call voting beyond perhaps the personal preferences of the member.

Analyses of several unique historical events bolster the findings in Table 1. After a competitive internal contest for House Minority Leader between Gerald Ford and Charles Halleck, I find that the switch from Halleck to the relatively moderate Ford significantly shifted the roll-call votes of rank-and-file Republicans, and it affected Halleck’s supporters approximately as much as it influenced Ford’s supporters. This suggests that party leaders influence even members who do not strongly support them, and it also suggests that the appearance of party leader effects is not entirely explained by selection.

When the Senate Minority Leader Everett Dirksen unexpectedly died in office, he was replaced by the relatively moderate Hugh Scott, and rank-and-file Republican senators accordingly voted in less partisan and extreme ways. Perhaps even more interesting, during the interim period with no party leader, rank-and-file Republicans were notably more moderate than they were under either leader. This case suggests that when left to their own devices, members of Congress would prefer to be more moderate and less partisan than they are even under a relatively moderate leader.

These analyses shed light on an interesting, historically anomalous period of Republican moderation in the 1960s and 1970s. The results suggest that a small number of congressional leaders caused a meaningful share of this moderation. Interestingly, I detect meaningful effects of leaders in this setting with low polarization and low party cohesion—the situation in which the theory of conditional party government would predict the weakest effects of party leaders (but see Rubin, Reference Rubin2025).

This study also contributes to an ongoing debate about the explanations for the fall and rise of political polarization in Congress over time (see McCarty, Reference McCarty2019 for a review). Because leaders significantly influence the voting behavior of their members and because parties are increasingly selecting extreme members as leaders, leader effects may explain a meaningful share of the rise in polarization over time. My estimates suggest that the increasing extremism of leaders can explain approximately half of the increase in polarization over the past 50 years. A fuller analysis of the causes of congressional polarization is beyond the scope of this analysis, but party leaders may be one important explanation for both the level of polarization and its increase over time.

Of course, the empirical analyses of this study do not explain why members of Congress typically select extreme leaders and why they are increasingly doing so over time. One potentially relevant explanation is that only experienced members of Congress are typically considered serious candidates for party leadership (Wolfinger and Heifetz, Reference Wolfinger and Heifetz1965), and because of increased geographic polarization (Nall, Reference Nall2018; Rodden, Reference Rodden2019) and the nationalization of elections (Hopkins, Reference Hopkins2018), experienced members are especially likely to come from particularly partisan states and districts, which tend to elect relative extremists (Fowler, Reference Fowler2024). Therefore, even if members would prefer more moderate leaders, the viable choices with many terms of experience are likely to be extremists. However, the theoretical discussion about why legislators select party leaders suggests that even in the absence of these forces, members may have an incentive to select a relatively extreme leader.

Despite strong claims that party leaders discipline their members, convincingly distinguishing between party discipline and selection effects has proven difficult. This study presents several different forms of evidence that allow us to distinguish leader effects from the selection of leaders. I find that party leaders significantly influence the behavior of rank-and-file members. When a party selects a relatively extreme leader, its rank-and-file members become more likely to cast ideologically extreme votes, and they are particularly more likely to cast votes in line with the majority of their party.