We thank Gary Jacobson, Keith Krehbiel, Doug Kriner, Jeff Lewis, and Keith Poole for data used in this project. We also gratefully acknowledge John Aldrich, Michael Bailey, Matthew Beckmann, Adam Berinsky, Chris Berry, Steve Callander, Charles Cameron, Brandice Canes-Wrone, Saul Jackman, Simon Jackman, Jeff Jenkins, Thad Kousser, Keith Krehbiel, Nolan McCarty, Eric Oliver, Eric Schickler, Boris Shor, Betsy Sinclair, Rob Van Houweling, five anonymous reviewers, and participants in seminars at Berkeley, Chicago, Duke, Notre Dame, Stanford, and Texas A&M for helpful comments and suggestions. For financial support, we thank the National Science Foundation, the Bradley Foundation, and the Program on Political Institutions at the University of Chicago. Mateusz Tomkowiak provided capable research assistance. All replication files can be accessed through the AJPS Dataverse at http://dvn.iq.harvard.edu/dvn/dv/ajps/faces/study/StudyPage.xhtml?globalId=hdl:1902.1/18254. Supplementary appendices are available at the AJPS website.
William G. Howell is Sydney Stein Professor of American Politics, Department of Political Science, Harris School of Public Policy, 1155 E. 60th Street, Chicago, IL 60637 (firstname.lastname@example.org.). Jon C. Rogowski is Assistant Professor, Department of Political Science, Washington University in St. Louis, Campus Box 1063, One Brookings Drive, St. Louis, MO 63130 (email@example.com).
An extraordinary body of scholarship suggests that war, perhaps more than any other contributor, is responsible for the emergence of a distinctly modern presidency. Central to this argument is a belief that members of Congress predictably and reliably line up behind the president during times of war. Few scholars, however, have actually subjected this argument to quantitative investigation. This article does so. Estimating ideal points for members of Congress at the start and end of the most significant wars in the past century, we find consistent—albeit not uniform—evidence of a wartime effect. The outbreaks of both world wars and the post-9/11 era—though not the Korean or Vietnam wars—coincided with discernible changes in member voting behavior that better reflected the ideological leanings of the presidents then in office. In the aftermath of all these wars, meanwhile, members shifted away from the sitting president’s ideological orientation. These findings are not confined to any single subset of policies, are robust to a wide variety of modeling specifications, and run contrary to scholarship that emphasizes ideological consistency in members’ voting behavior.
The larger scholarship on congressional voting behavior, meanwhile, wholly ignores war. In studies that meticulously document the impacts of ideology, partisanship, and the internal hierarchies of Congress on members’ voting behavior, war hardly makes an appearance (for a recent review, see Theriault, Hickey, and Blass 2011). Further, by emphasizing ideological rigidity and voting consistency, these studies take, if only implicitly, a rather dim view of the possibility that congressional support for the president flows and ebbs as the nation moves into and out of war.
This article puts the issue of war front and center. It investigates whether members of Congress tend to vote in ways that more closely approximate the policy preferences of presidents during wartime than they do during peace. Rather than focus on the highly selected subset of bills on which presidents take public positions, we cull all roll-call votes cast in those congresses during which major modern wars either began or ended. Employing a variety of techniques to facilitate comparisons over time, we then estimate peace- and wartime ideal points for every member of Congress.
We find substantial—albeit not uniform—evidence of a wartime effect. The outbreaks of both world wars and the post-9/11 era precipitated notable changes in member voting behavior that better reflected the sitting president’s ideological orientation. The beginnings of the Korean and Vietnam wars, meanwhile, did not generate clear evidence of congressional accommodation to the president. When transitioning from all of these wars to periods of peace, meanwhile, members of Congress appeared to shift away from the presidents then in office.
This article proceeds as follows. The first section briefly characterizes the relevant literatures on war, presidential power, and congressional voting behavior. The second explains our data and modeling strategies. The third and fourth sections present our main findings and a wide assortment of robustness checks for the 2001 war in Afghanistan, for which we have the strongest identification strategy. The fifth section reports the more provisional results for World Wars I and II and the Korean and Vietnam wars, and the sixth section concludes.
For centuries, no less, statesmen and scholars have argued that wars contribute mightily to presidential power. Shunning centuries of European precedent, such concerns guided the Founders’ decision to vest a legislature, rather than an executive, with the preponderance of warmaking authority. In Federalist 8, Alexander Hamilton recognized that “it is the nature of war to increase the executive at the expense of the legislative authority.” Echoing these sentiments, in Helvidius 4 James Madison argued that “war is in fact the true nurse of executive aggrandizement.” On this point, the Anti-Federalists essentially agreed with Hamilton and the other Federalists. The Anti-Federalists simply rejected the proposition that the constitution on offer adequately tempered the possibility that through war a president might eventually become, for all intents and purposes, a king.
In this article, we subject such claims to empirical scrutiny. In particular, we consider how Congress—possessing war powers of its own—responds to the president during times of peace and war. To do so, we consider the possibility that, during war, members of Congress adjust their voting behavior to better reflect the president’s ideological commitments. When a Democrat [Republican] inhabits the White House, members of Congress may vote more liberally [conservatively] when the nation enters war, and more conservatively [liberally] when it exits war. Presidential power expands, by this account, because wartime presidents bargain and negotiate with congresses that exhibit greater support for their domestic and foreign policy agendas.2
Congressional Studies on War
Against a backdrop of sweeping claims about war and presidential power, few empirical studies have actually scrutinized the topic. Those that do, meanwhile, furnish little support for the contention that wars increase congressional support for the president. Where wartime effects are observed, they are confined to certain time periods, subsets of policy initiatives, and/or a single chamber of Congress. Cohen (1982), for instance, presents evidence that the wars in the late nineteenth century augmented president success in Congress, but not since. Examining roll-call votes taken between 1947 and 1988, Meernik (1993) finds that some wars bolster support in some chambers of Congress for the president’s foreign policy and defense proposals. Like Meernik and Prins and Marshall (2001), Peake (2002) suggests that the evidence of a wartime effect is limited to foreign policy bills. Finally, in a recent working paper, Schorpp and Finocchiaro (2010) present evidence that members of Congress were more likely to support elements of the president’s domestic policy agenda in the Vietnam and post-September 11 wars, but not in the Korean or Gulf wars. Moreover, Schorpp and Finocchiaro do not find any evidence that any of these wars increased the chances that members of Congress supported elements of the president’s foreign or defense agendas.
Based on the available evidence, it does not appear that members of Congress automatically line up behind their president during periods of war. It is much too soon, though, to draw strong conclusions about the causal impacts of war on congressional voting behavior. All of these studies, after all, face a basic challenge: they focus exclusively on samples of legislative initiatives on which the president took public positions. Ascertaining presidential power on the basis of roll-call votes on presidential initiatives, however, is extraordinarily difficult. Presidents do not randomly select elements from their policy agenda to put before Congress. Rather, presidents choose those policies that they think stand a decent chance of passage and set aside the rest. If such selectivity is a function of war—and there are ample reasons to believe that it is—then systematic biases are introduced that, uncorrected, may obscure war’s genuine effects on presidential power.3
The take-away point from these studies is clear: comparisons of individual members over short periods of time—and especially those that focus on trends within a single congressional term among members who did not switch parties—can be expected to reveal overwhelming stability in individual voting behavior; and no external event, very much including war, is likely to dislodge members from their chosen ideological footings. Accordingly, the empirical tests that follow constitute a tough test of the proposition that members’ voting behavior differs in war and peace.
Data and Methods
In subjecting claims about presidential power during war to empirical scrutiny, we examine whether members of Congress shift their voting behavior in the ideological direction of the president when the nation enters war and away from the president when the nation exits. To conduct such comparisons, we split roll calls based on the presence or absence of war. Whenever possible, we restrict the sample to a single Congress and thereby control for the many contextual factors that have been the subject of previous scholarship on congressional voting behavior. Crucially, by analyzing the universe of congressional roll calls within these time frames, rather than the subset of bills on which presidents have taken public positions, we substantially reduce the selection biases that confront previous research on war and congressional voting behavior.
Intertemporal Comparisons of Roll-Call Voting Records
To recover estimates of legislator ideologies using roll-call data, we rely upon the Bayesian approach recommended by Clinton, Jackman, and Rivers (2004). So doing, though, we confront a basic challenge. Because the scales may have shifted or stretched, ideal points that are estimated with different sets of data cannot be directly compared.4 To meaningfully compare estimates from separate samples of roll-call votes, therefore, Bailey (2007) and Bailey and Chang (2001) have recommended the use of “bridge” observations that serve as fixed reference points against which ideal point estimates can be compared.5 Using this approach, scholars have compared preference estimates between presidents, between senators and Supreme Court justices, bureaucrats, and members of Congress, and between state legislators and members of Congress (see, e.g., Bailey and Chang 2001; Clinton and Lewis 2008; Shor, McCarty, and Berry 2010).
Our chosen empirical strategy does not require us to compare different actors across different settings, but rather the same actors across different time periods—transitioning either from peace to war or from war to peace. The trick, then, is to identify bridge actors whose willingness to support different bills is plausibly unaffected by war. Though we explore a variety of options, we place the greatest confidence on those estimates that use interest groups to link peacetime and wartime congressional voting records. During war, after all, interest groups are less likely than any other political actor or reference group to assume different positions on pending bills. Moreover, as Poole (1981) demonstrates in an analysis of interest group positions on congressional roll-call votes, interest groups’ ideological orientations are generally consistent and stable over time.
Following Poole and Rosenthal (1997), we rely upon the American Conservative Union (ACU) and Americans for Democratic Action (ADA) as our bridges. The use of these interest groups offers several practical benefits. First, both groups take a fairly large number of positions (generally, 40–50) per chamber during each Congress, which enables us to estimate their ideal points precisely. Additionally, whereas some organizations such as the American Civil Liberties Union and the National Rifle Association are concerned primarily with roll calls that address a specific issue domain, the ACU and the ADA take positions that span a wide range of issues. Because we are interested in documenting shifts that can be meaningfully described as either “liberal” or “conservative,” it is important that a full range of issues define the continuum over which legislator estimates are compared.
Of course, there are potential downsides to assuming that any actor’s preferences remain constant over time. For several reasons, however, the assumption appears justified in this instance. First, we track the positions of interest groups and legislators within short periods of time, never more than two years. Second, the ACU and the ADA are the nation’s oldest existing conservative and liberal interest groups, which suggests some degree of ideological consistency that may not exist among organizations with shorter histories. And third, though interest groups may be strategic in their selection of roll calls (Fowler 1982; Snyder 1992), the recovered estimates should be consistent as long as the interest groups take ideologically consistent positions on the key votes that they select during peace and war.
To see how our bridging technique works in practice, consider the analysis of the 107th Congress. First, we split the set of roll calls for the 107th Congress at the date on which the war in Afghanistan began (October 7, 2001). We then construct separate matrices for the pre- and postwar roll calls in both the House and Senate, in which the rows correspond to legislators and the two interest groups.6 Column entries indicate whether the legislator or interest group supported or opposed each bill. By estimating the above statistical model in an unidentified state, we generate member ideal points for the roll-call votes that occurred prior to October 7, 2001. The data are postprocessed to constrain the ideal point estimates to have mean zero and unit variance, where negative ideal points identify more liberal members and positive ideal points identify more conservative members.7 After recovering the prewar estimates, we estimate the postwar ideal points for all members of Congress while constraining the estimates of the ACU and ADA to equal their prewar estimates. We then compare the pre- and postwar estimates at both the individual and chamber levels.
If war increases presidential power vis-à-vis Congress, as the scholars cited above argue, we expect to observe shifts toward the president then in office as the nation enters war and away from the president when the nation exits war. Table 1 translates these general expectations for the specific wars we investigate in this article.
Table 1. Expected Shifts in Roll-Call Voting Behavior
President’s Ideological Orientation
Expected Direction of Shift
World War I
World War I
World War II
World War II
When are we at war? Or, more to the point, when are we not? In the modern era, hundreds of thousands of troops are stationed around the world. Given its hegemonic status, the United States constantly monitors emergent threats to the nation’s economic and political interests from anywhere around the globe. And if that weren’t enough, modern presidents have waged “wars” against cancer, polio, drugs, and poverty. For purely practical reasons, therefore, we distinguish those actions that Congress takes during both world wars, the Korean War, the Vietnam War, and the war in Afghanistan that followed the attacks on September 11, 2001. By any metric, these constitute the most significant military actions taken by the United States in the last century.8
We define the beginning of war when actual fighting that involves U.S. troops either commences (in the case of World Wars I and II, Korea, and Afghanistan) or rapidly escalates (in the case of Vietnam). Wars end, then, when a formal peace treaty is signed, regardless of whether U.S. troops remained stationed in the region or intermittent fighting continues. With these rules, the following dates delineate the beginnings and ends of each of our wars: World War I (April 6, 1917 to June 28, 1919), World War II (December 7, 1941 to August 14, 1945), Korea (June 27, 1950 to July 27, 1953), Vietnam (February 7, 1965 to January 27, 1973), and the war in Afghanistan (begins October 7, 2001).
By construction, this definition of war does not distinguish mobilization and demobilization efforts, the varying levels of military engagements, or the changing size and immediacy of foreign threats. Despite these limitations, this definition offers important advantages. First, and foremost, it zeroes in on those military actions that have the best chance of reconstituting the domestic polity. If legislative behavior is insensitive to U.S. involvement in these wars, it seems unlikely that members of Congress are affected by participation in smaller military campaigns. Second, to the extent that it allots all kinds of military actions and threats to periods of “peace,” our definition constitutes something of a hard test for the proposition that members of Congress adopt a more deferential posture vis-à-vis the president during war. Operationally, what we are calling “peace” may, in fact, not be peaceful at all. If so, though, then the recovered estimates reported below most likely understate the true impact of war on congressional voting behavior.
Primary Results: Afghanistan
In the aggregate, we find strong support for the proposition that members of the 107th Congress were more likely to vote in ways that better reflected President George W. Bush’s preferences upon the commencement of the war in Afghanistan. In the top row of Figure 1, we plot the densities of members’ estimated peace- and wartime ideal points in the 107th Congress. The dashed lines indicate the peacetime scores, and the solid lines indicate the wartime scores. Scores are arbitrarily scaled from −2 to 2, where larger values indicate more conservative voting behavior. Both distributions are bimodal, with Democrats populating the left portion of the distribution and Republicans the right. In both the House (left panel) and Senate (right panel), we see marked shifts to the right. In the House, the movement appears to be concentrated among Republicans, while in the Senate both parties shift rightward. The unconstrained mean and median shifts in the House were +0.54 and +1.28, respectively; and in the Senate, they were +1.21 and +1.26.9
In the bottom row of Figure 1, we again plot the pre- and post- scores for every member of the 107th House and Senate. This time, however, individual scores before the outbreak of war are aligned on the x-axis, and wartime scores appear on the y-axis. If a member’s voting behavior did not change at all, then she will appear right on the 45-degree line. Scores above the 45-degree line indicate movement in the conservative direction, and scores below the 45-degree line reveal movement in the liberal direction.
Plainly, the vast majority of members in both chambers appear above the line. Indeed, all 95 members of the Senate and 323 of the 362 members of the House who reveal statistically significantly different pre- and post- scores document movement in the conservative direction. As points of reference, we have identified a handful of individual members. Compare, for instance, the locations of two Senate Democrats: Hillary Clinton and Paul Wellstone. As one would expect, Wellstone appears to the left of Clinton, indicating a significantly more liberal voting record. Importantly, though, Wellstone is much closer to the 45-degree line than is Clinton, suggesting the outbreak of war had less influence on his voting behavior.10
Though Democrats and Republicans in both chambers shifted to the ideological right, they did so to different degrees. Indeed, the intraparty variances of members’ ideal points actually increased during war. In the Senate, the variances increased for members of both parties, while in the House only the variance for Democrats increased. Interestingly, as the differences within parties increased, differences between the parties decreased. Party unity scores—which represent the percentage of the roll-call votes on which at least 50% of one party voted against more than 50% of the other party—decreased from 45% to 40% in the House and from 61% to 43% in the Senate.
From a research design perspective, the 107th Congress provides a nearly ideal venue in which to examine the impact of war on members’ voting behavior. The impetus for war was plausibly exogenous. Lacking either the impending threat of war (as characterized the beginning of World War II) or the steady buildup of troops (as in Vietnam), the dividing line between peace and war in 2001 was quite clear. Because the war began almost midway through the congressional term, large numbers of roll-call votes are available for us to generate precise estimates of members’ peace- and wartime ideal points. And because all the votes come within a single Congress, we hold constant most (though not all) intrainstitutional influences on voting.
Of course, by relying on observational data we unavoidably confront important identification challenges. In this section, therefore, we submit our findings to a host of robustness checks. In particular, we investigate the underlying assumption of interest group consistency, alternative ideal point estimation techniques, systematic changes in the congressional agenda brought on by war, and a variety of other issues.
Bridge Actors and Estimation Strategies
Our primary estimation strategy hinges upon the assumption that interest groups do not adjust their positions on pending legislation in ways that correspond with the presence or absence of war. As a first check on this assumption, we estimated a heteroskedastic item-response model (see Lauderdale 2010) using the entire set of roll-call votes and interest group positions in the 107th Congress. This procedure recovers uncertainty estimates of legislator ideal points, wherein larger values indicate that the legislator’s position on any given roll-call vote is predicted less well by their ideal point estimate along the primary dimension. The mean values of these estimates for the members of the 107th House and Senate are 1.08 and 1.04, respectively. Our interest groups have smaller estimates in both the House (ADA = 1.01, ACU = 1.01) and Senate (ADA = 1.01, ACU = 1.02).
Using interest groups to evaluate voting behavior in the 107th Congress would be problematic if one or both of the interest groups shifted to the ideological left, because it would falsely reveal shifts in member voting records in the conservative direction.11 It is improbable that the ACU would demonstrate more liberal voting behavior postwar than prewar, but this could be a concern with the ADA. Thus, we replicated our analysis, but this time substituted positions taken by the AFL-CIO for those taken by the ADA. In doing so, we again found that members of Congress had more conservative wartime voting records.12 The results using the AFL-CIO in place of the ADA even more clearly demonstrate that members moved to the ideological right upon the beginning of the Afghanistan War, and an even greater number of members of both the House and Senate exhibit significantly more conservative records in the postwar period.
We again obtain similar results when we utilize more than two interest groups as our bridges, in which we supplement the ACU and the ADA with the AFL-CIO, the American Civil Liberties Union (ACLU), the League of Conservation Voters (LCV), and the Public Interest Research Group (PIRG).13 We now observe mean shifts of +0.49 and +0.58 in the House and Senate, respectively.14
We also examined whether our results are a function of the kinds of bills on which the interest groups rated legislators. We reestimated pre- and postwar ideal points using only the votes either or both the ADA and ACU identified as a key vote. Even within these small subsamples, we observe a mean shift in the House of +0.20 and in the Senate of +0.45.15
Our findings for the 107th Congress also do not appear to be especially sensitive to the particular estimation procedure that we employ. As another check on the validity of our results, we replicated the estimation approach found in Bailey (2007) and Shor, McCarty, and Berry (2010) while continuing to use the ADA and ACU as our bridge actors.16 The results from this estimation procedure are broadly consistent with those reported above. We continue to observe substantial shifts in the conservative direction among members in both chambers of Congress.17 The mean shifts in the House and Senate are +0.26 and +0.50, respectively, and both figures are statistically significant.18
The Wartime Agenda
Our ideal point estimates, we recognize, may confound systematic shifts in the congressional agenda. This possibility takes two general forms. First, the onset of war may have coincided with wholesale changes in the kinds of issues considered by members of Congress. This is of some concern, as the ideological shifts in the conservative direction are observationally equivalent to agenda changes in the opposite direction. The second concern is that our estimates do in fact capture some differences in member voting behavior before and after the war began, but that the war opened up an entirely new dimension of ideological conflict that is not captured by the main liberal-conservative dimension. If true, then our recovered estimates may not reflect substantively meaningful changes in legislative behavior. To address these dual concerns, we conducted two sets of supplementary analyses.
First, recall that on May 24, 2001, Senator James Jeffords of Vermont switched his party affiliation from Republican to Independent and announced he would caucus with the Democrats, which awarded control of the Senate to the Democrats. This change in leadership may have affected the agenda on which senators voted. It may also have affected the House agenda, as House leaders may have changed their expectations about what they could get through the Senate. To account for this change in majority party, we replicated the original analysis with the full set of wartime roll-call data, but we limited the peacetime roll-call data to the period after Jeffords switched his party affiliation. Our substantive findings remain unchanged.19 With the outbreak of the Afghanistan War, we continue to see House Republicans and Democratic and Republican senators lurching to the right. Thus, to the extent that congressional agendas are a function of party control, we find no evidence that changes in the agenda are responsible for the findings described above.
Next, we examined the distribution of roll-call vote cutpoints, which indicate where the separation occurs between supporters and opponents of a bill. We calculate cutpoints using the item parameters that are generated simultaneously with the legislator ideal points.20 If the findings we observe are due to changes in the agenda, then we should observe shifts in the distribution of cutpoints to the ideological left upon war’s beginning. We therefore estimate cutpoints for the periods before Jeffords’ switch, after the Jeffords switch but before the beginning of war, and once the war begins. Across the three time periods, the median House cutpoints all fall just to the left of zero. The results are shown in Figure 2.
The distribution of post-Jeffords switch cutpoints is flatter than that for those observed before the Jeffords switch. But the distributions of cutpoints immediately before and after the outbreak of war are indistinguishable from one another, suggesting that an agenda change is not responsible for the observed conservative shift in the voting records of House members. In the Senate we do observe changes in the cutpoint distributions. The cutpoints prior to the Jeffords switch were more liberal than those after the switch but before the war, indicating that the agenda indeed shifted to the left once the Democrats gained control of the Senate. Once the war began, however, the cutpoints shifted in the liberal direction, indicating that Democrats put forth a more conservative agenda once the war began. This fact suggests that comparisons of peace and wartime ideal point estimates based upon roll calls that occurred after the Jeffords switch understate the extent to which senators voted more conservatively once the Afghanistan War began.
More substantively, we also investigated the contents of bills voted on in the 107th Congress. So doing, we find little evidence that the presence of war induced significant changes in the kinds of issues taken up by Congress.21 To be sure, the wartime congress considered more items related to defense, foreign policy, and government operations and justice. These increases, though, came largely at the expense of roll-call votes that dealt with appropriations. Every other area of domestic policy was virtually unaffected by the presence of war.22
To examine whether our findings are primarily artifacts of shifts in voting behavior on issues of national defense and foreign policy, we examined shifts in voting behavior on purely domestic issues. Domestic issues were defined narrowly, consisting of roll-call votes that occurred on issues not related to national security or defense, foreign policy, or any sovereign state. We then estimated separate peace- and wartime ideal points for each member of the 107th House and Senate, restricting bridge observations to the subset of key votes within each domain.
Results for the 107th Congress are presented in Figure 3. Across both chambers, we again find widespread evidence of rising conservatism. Indeed, when looking at this subset of bills, the shifts are even more pronounced and far-reaching than those observed when using the entire samples of roll calls. These results weigh strongly against the claim that war’s effects on presidential power are confined primarily to the domain of foreign affairs.
Finally, war could spawn a whole new set of issues that is orthogonal to the main liberal-conservative dimension that characterizes so much of congressional voting behavior. We examine this possibility by estimating legislator ideal points in both one and two dimensions. During the prewar period, a single dimension correctly predicts over 91% of the votes in both the House and the Senate, and a second dimension correctly classifies another 1%. The same patterns hold true for the wartime period. Compared to a single dimension, a second dimension correctly classifies 1% more of the House roll-call votes, and 1.78% more of the Senate roll-call votes. These results further allay concerns that the shifts in roll-call patterns that we observe above are due to the presence of an additional agenda that is orthogonal to the main dimension along which we characterize legislative voting behavior.
Additional Robustness Checks
The previous analyses suggest that our findings are not an artifact of any particular estimation strategy, selection of bridge actors, or changes in the legislative agenda. In the remainder of this section, we explore still more explanations for our results.
Defining the Beginning of War. Changes in members’ voting behavior may not have coincided perfectly with the outbreak of war. Instead, perhaps, members may have lurched to the right either in the immediate aftermath of the September 11 terrorist attacks or later in the war. We also must scrutinize the possibility that factors having nothing to do with the war precipitated the observed shifts.
To wit, we reestimated member ideal points using arbitrary dates during 2001 to distinguish “control” and “treatment” conditions—specifically, the first and fifteenth of every month between May and November 2001—and then calculated the magnitude of the observed shifts for each set of estimates in the House and Senate. We find that the evidence of changes in member voting behavior strengthens as the designated partition date moves into the late fall.23 Though these findings do not allow us to discern whether different members moved just once though at different times, or whether members moved together but repeatedly during this time period, they do support two conclusions. First, factors unrelated to the war are probably not responsible for the observed shifts in voting behavior; and second, members continued to tack to the ideological right in the weeks and months that followed the September 11 attacks.
Rising Conservatism and War. It also is possible that the shifts observed after the outbreak of the Afghanistan War had less to do with the president, per se, and more to do with a rising conservatism evoked by war. To investigate this possibility—which we consider more fully below when examining previous wars undertaken by Democratic presidents—we assessed changes in voting behavior in the California legislature during the same time period.24 Using the same estimate technique described above, we find that, rather than observing a shift in the conservative direction upon the outbreak of war, members of the California assembly compiled strikingly more liberal voting records.25 These shifts occurred at the same time that every single congressional representative from California who showed significantly different voting patterns during this period (both senators and 37 of 51 members of the House, Republicans and Democrats alike) shifted in the conservative direction after the war began. These findings weigh against the notion that the Afghanistan War evoked a general conservative reaction among political actors whose electoral fortunes and policy agendas were not directly tied to the president.
Congressional Voting Behavior and the Electoral Calendar. It also is possible that the shifts we observe have little to do with the outbreak of war and instead reflect typical changes in the voting patterns of members of Congress over the course of a congressional term. To investigate this possibility, we generated and compared ideal points for the first and second sessions of three congresses during which no major military actions occurred: the 95th (Carter, 1977–78), 99th (Reagan, 1985–86), and 103rd (Clinton, 1993–94). Examining each of these three congresses, we find little systematic evidence of changing patterns in voting behavior over the course of a single congress.26 To the extent that aggregate changes are observed, the movement occurs in the direction opposite the ideological position of the president—a finding that may reflect member posturing in anticipation of midterm losses. Were this pattern to hold for the 107th Congress, members would have more liberal voting records after the war began. The available evidence, however, suggests that just the opposite occurred.
War and Other Crises. We further recognize that war may be an element of a larger class of phenomena that induce systematic changes in support for the president. We therefore examined instances of other types of events that may do so: foreign crises (the Iran hostage crisis in 1979), smaller military deployments (Lebanon in 1982 and Bosnia in 1995), and presidential scandals (the news of President Clinton’s affair with Monica Lewinsky in 1998). We again find little systematic evidence of congressional accommodation to or rejection of the president upon the outbreak of these events.27 In none of these cases do we observe consistent effects across the House and Senate. Indeed, in every case that movement is detected toward the president’s ideological orientation in one chamber, movement away from the president is detected in the other. While these examples clearly do not exhaust the possibilities of crises (military or otherwise) that implicate the president, the collective evidence presented here suggests that U.S. involvement in a major military undertaking has the potential to alter patterns of congressional voting in ways that these smaller affairs do not.
In an effort to mitigate the deep selection and endogeneity issues associated with presidential position taking, we have chosen to examine all roll calls cast in a congress, rather than just the subset upon which the president publicly supported or opposed. Pursuing this strategy, however, we are unable to identify where, exactly, the president resides along an ideological continuum. And not knowing the president’s location, we cannot precisely measure each member’s relative proximity to him during war and peace. To address this basic problem, we take advantage of the historical variability in the partisan composition of Congress and the White House during the major wars conducted in the past century. By analyzing multiple wars conducted by presidents with very different ideological orientations, we can identify whether members of Congress shift in the general direction of the president then in office.
Earlier wars, however, present new identification problems. To analyze World War I and the Vietnam War, we must pool observations across congresses so that we have sufficient numbers of roll-call votes on both sides of the transitions between peace and war. The beginning of the Vietnam War, moreover, is contested, and the end is conflated by Watergate. The absence of interest group data for most of these earlier wars, however, poses the greatest challenge. Interest groups did not identify “key votes” for the congresses that served during either of the world wars or the Korean War.28 In place of interest groups, we apply two criteria to identify individual members of Congress whose voting records are least likely to have been affected by the onset of war. Online Appendix C describes this alternative strategy in greater detail. In the remainder of this section, though, we go to some length to clarify the unique challenges that each war presents and our efforts to ameliorate them.
Further Evidence of a Wartime Effect: The World Wars
World Wars I and II present a number of challenges to estimating changes in member voting behavior. Because Congress did not anticipate the Pearl Harbor attacks in December 1941, U.S. entry into World War II is likely to be exogenous to the congressional agenda; but this is less certain for World War I. Second, as we noted above, interest group positions are not available during these time periods, so we rely on individual members as bridge actors.29 Finally, because U.S. involvement in World War I began (April 6, 1917) and ended (June 28, 1919) early in new congressional terms, we lack sufficient numbers of peacetime and wartime roll-call votes to generate reliable estimates of legislators’ voting records. Thus, to assess changes in voting behavior upon the beginning of war, we analyze the voting behavior of members who served in both the 64th and 65th congresses, and we evaluate behavior upon the end of war using members who served in both the 65th and 66th congresses.
Even with these challenges, both world wars provide evidence of congressional accommodation to the president upon the beginning of war, and of congressional separation from the president at the end of war. Take a look at Figure 4, which presents our aggregate findings across all of the wars we examine. The plot on the left shows the extent to which congressional voting records shift in the ideological direction of the president then in office upon war’s beginning, and the plot on the right does the same for congressional voting behavior upon the end of war. The x-axis represents the magnitude of the shift in voting records and is scaled such that larger positive values indicate larger movements in the direction of the president, while negative values indicate movement away from the president’s ideological orientation. The large solid dots represent the average effect for each war and chamber, which are obtained by aggregating the shifts in member voting behavior across every method we used to identify changes between peace- and wartime (results from all such analyses are contained in section B of the online appendix). The horizontal bars indicate the 95% confidence intervals. The dashed vertical line is zero, which indicates that war did not produce an observable impact on member voting behavior. Our main interests are in observing whether the effects are consistent across both chambers within a given war and that the confidence intervals do not include zero.
The evidence is strongest for World War II, in which members of both the 77th House and Senate voted more liberally after the United States entered World War II. The end of World War II, meanwhile, yielded a Congress less willing to vote in ways that reflected the preferences of President Truman.30 Each of the analyses that use all roll-call votes shows shifts in the conservative direction upon the conclusion of World War II, though the findings for the Senate fall just short of statistical significance at conventional levels.
Though more tentative, the results for World War I are generally consistent with those for World War II. Once the United States entered World War I, both sets of bridging criteria indicate that members of the Senate shifted to the left of the ideological spectrum. The estimated mean shifts are −0.15 and −0.55, both of which are statistically significant. In the House, where only one bridging strategy was possible, the observed shifts are not statistically different from zero. Comparable findings appear at the end of the war: both estimates for the Senate show significant movement in the conservative direction—that is, away from the preferences of President Wilson. The results for the House are suggestive of a withdrawal of support for Wilson’s ideological orientation, though the estimated effects are statistically indistinguishable from zero.
Mixed Evidence of a Wartime Effect: Korea and Vietnam
As with World Wars I and II, we rely upon members of Congress as our bridges for the Korean War. In this instance, though, we find mixed evidence of wartime congressional acquiescence to the president. Though the results show that the onset of the Korean War induced a liberal response from members of the House, neither set of Senate results confirms this finding. Indeed, members of the Senate appear to have voted in a more conservative manner upon the war’s commencement. Examining congressional voting upon the war’s end, however, we find that members of Congress were less likely to vote in ways that reflected Eisenhower’s preferences.31 Upon war’s end, members of the 83rd House and Senate moved in the liberal direction—away from the president’s ideological orientation. Thus, while the patterns of findings for the Korean War differ somewhat from those for Afghanistan and World Wars I and II, the general pattern is consistent with a wartime effect. While the beginnings of wars do not always induce systematic patterns of congressional accommodation, the ends of all wars reveal a decline in congressional accommodation to the president.
Unlike the world wars and Korea, interest group positions are available for us to compare voting behavior before and after the Vietnam War.32 The Vietnam War, nonetheless, presents several unique challenges to our analysis. First, unlike the other wars we examine, the gradual escalation of U.S. involvement makes it difficult to identify a clear start date. Following the lead of other scholars (e.g., Epstein et al. 2005), we mark February 7, 1965, as the beginning of the Vietnam War, which corresponds with the execution of Operation Flaming Dart, the first large-scale military initiative following the Gulf of Tonkin Resolution. Second, because the war begins and ends quite early in new congressional sessions, we again pool across congresses, examining those members who served in the second session of the 88th Congress (1964) and the first session of the 89th (1965), and the second session of the 92nd Congress (1972) and the first session of the 93rd (1973).
We find no evidence that the beginning of the Vietnam War yielded a Congress more willing to vote in ways that reflected the president’s ideological orientation. In the House we find that members voted more conservatively upon the beginning of war, while Senate voting patterns in the two time periods are not statistically different from one another. Consistent with evidence of a wartime effect, however, we do find that with the Vietnam War’s termination, the voting records of members of Congress shifted in the liberal direction—that is, away from the ideological orientation of the president then in office. In the analyses that include all roll-call votes cast in 1972 and 1973, members of Congress compiled significantly more liberal voting records upon the end of war.33
An extraordinary body of scholarship contends that wars constitute a boon to presidential power. And in virtually all of these accounts, Congress plays a prominent role. Members of Congress, it is supposed, predictably and reliably line up behind their president during times of war. And according to some, most notably Schlesinger, congressional accommodation to the president should promptly dissipate the moment wars end.
In this article, we present substantial support for such claims. At the outsets of World Wars I and II and the Afghanistan War, members of Congress began to vote in ways that better reflected the ideological orientations of the presidents then in office. All observed transitions from war to peace, meanwhile, coincide with shifts in congressional voting behavior away from the president’s ideology. The beginnings of the Korean and Vietnam wars, however, do not offer consistent evidence of a wartime effect on congressional voting behavior.
By characterizing members’ voting behavior on the universe of roll-call votes taken, these findings mitigate the deep selection and endogeneity issues associated with presidential position taking. Moreover, they are robust across a range of specification and estimation strategies, are not confined to any single subset of bills, and do not conflate agenda changes that coincide with the onset of war.
By design, though, our findings have limitations of their own. Most obviously, they do not permit explicit evaluations of the president’s location during war and peace. Hence, while we can assess whether members adjust their voting behavior in ways that broadly comport with the president’s ideology, we cannot measure the precise distances between members’ ideal points and the president’s. Similarly, not all of our estimates are equally reliable. We place the greatest confidence in those based upon interest group bridges within a single Congress, and less weight upon those that pool votes across congresses or that rely upon members of Congress to serve as bridges. And finally, these data are susceptible to the more general selection and endogeneity issues associated with roll-call voting.
We further recognize that our findings speak to changes in congressional roll-call voting as the nation moved into and out of war. Future research, however, would do well to explore how these patterns are affected by the specific conduct of war-related efforts. It is quite possible that congressional accommodation to the president increases with military escalations. Likewise, the pace of the withdrawal of resources, troops, and otherwise, from activities related to war may also influence the extent to which congressional voting patterns reflect the ideological leanings of the president.
Despite these limitations, the findings on offer challenge existing literatures that either consider the explicit relationship between presidential power and war or that emphasize the general stability of legislators’ voting patterns. Though few scholars have subjected claims about war and the expansion of presidential power to empirical scrutiny, those that do furnish little to no evidence of increased congressional support for the president during wartime. Meanwhile, while members of Congress may indeed “die in their ideological boots,” as congressional scholars emphasize, we find that members of Congress can—and often do—adjust their voting behavior upon the outbreaks and ends of wars.
With the evidence now before us, we enjoin scholars to pay closer attention to the micro-foundations of legislative decision making that lead members of Congress to acquiesce to the president during some wars while holding more firmly to their ideological commitments during others. Future scholarship would do well to scrutinize the ways in which wars reshape relations between executives and legislatures, paying particular attention to the mechanism that is responsible for whatever advantages presidents enjoy as a condition of war.
Wars, clearly, may augment other aspects of presidential power as well. Congress may grant more autonomy to bureaucratic agencies, staffed by employees of the president’s choosing; judges may be more willing to uphold presidential actions or policies that, during peacetime, would not pass constitutional muster; the Senate may be more willing to confirm a president’s judicial nominees; or Congress may enact (indeed, has enacted, and by the hundreds) emergency laws that trigger new presidential powers during times of war.
Members who served for short intervals of a given Congress were dropped from the analysis.
We fit a one-dimensional item-response model, running 300,000 iterations after discarding the first 50,000 and thinning by 500.
The 1991 Persian Gulf War might reasonably be classified similarly. However, its short duration presents a number of identification challenges, and thus we focus on these other wars that lasted significantly longer periods of time.
Additional summary statistics for all of the analyses can be found in the tables in section A of the online appendix. However, readers are cautioned not to compare the magnitudes of the observed shifts across congresses or between chambers or bridging criteria.
These senators’ public comments about the war are consistent with the observed differences in their voting records. The day after the September 11 attacks, Clinton declared on the floor of the Senate that “We will also stand united behind our President as he and his advisers plan the necessary actions to demonstrate America’s resolve and commitment . . . I have expressed my strong support for the President, not only as the Senator from New York but as someone who for eight years had some sense of the burdens and responsibilities that fall on the shoulders of the human being we make our President” (Congressional Record 2001, S9288). At the time, nothing similar could be heard from Wellstone. One year later, in spite of the prevailing political winds that led to a historic level of midterm success for Bush and the Republicans, Wellstone was one of only four Democratic Senators seeking reelection to oppose the authorization for the use of force in Iraq. Wellstone insisted that “Right now, despite a desire to support our president, I believe many Americans still have profound questions about the wisdom of relying too heavily on a pre-emptive, go-it-alone military approach” (Congressional Record 2002, 19025).
And, on the other hand, if one or both interest groups took more conservative positions during war than they did prior to the war, the magnitude of the shifts we observe understate the extent to which members of Congress voted in accordance with the president’s preferences once the war began.
These results are shown in Figure B-1 in the supplementary appendix.
These constitute the complete set of available interest groups that selected sufficient numbers of key votes to serve as bridges in both chambers in the 107th Congress. When also using the National Farmers Union (NFU) and National Right to Life Council (NRLC), which are only available for the House, we again observe significant shifts in the conservative direction.
These results are shown in Figure B-2A in the supplementary appendix. Because no interest groups other than the ADA and ACU take positions on a diverse assembly of bills over a long time span, we are reluctant to rely upon a broader population of interest groups as our bridge actors in our main analyses.
Further limiting the sample to only the key votes that addressed issues unrelated to national security or the war efforts, we observe mean shifts of +0.46 and +0.95 in the House and Senate, respectively.
Specifically, we created one roll-call matrix for each chamber, in which each legislator’s voting record is split over two rows. One row contains her prewar roll call votes, and the other contains her voting record once the war began. Each of our interest group bridges, meanwhile, appears in just one row of the data matrix and thereby provides the needed “glue” that facilitates comparable peacetime and wartime estimates.
See Figure B-2B in the supplementary appendix.
Note that the magnitudes of these shifts cannot be compared to those recovered from our own estimation procedure.
See Figure B-3 in the supplementary appendix.
Roll-call votes that are unanimous or extremely lopsided do not provide any information about member ideal points because every member (or virtually every member) votes in the same way. Such uninformative votes have discrimination parameters very close to zero, resulting in cutpoints that go to infinity in the limit. Consequently, we drop all roll calls for which the discrimination parameter is indistinguishable from zero. See Krehbiel, Meirowitz, and Woon (2005) for a similar analysis of legislative cutpoints.
The California legislature has the advantage of furnishing large numbers of roll-call votes in the 2001–2 session, the availability of high-profile liberal and conservative interests groups (the California League of Conservation Voters and Chamber of Commerce, respectively), and a Democratic governor (Gray Davis) then under its watch.
These results are shown in Figure B-5 in the supplementary appendix.
See Figure B-6 in the supplementary appendix.
See Figure B-7 in the supplementary appendix.
Though the ADA began issuing congressional ratings in 1947, the ACU did not do so until 1971. A conservative interest group, Americans for Constitutional Action, preceded the ACU, but it did not begin to issue ratings until 1959.
One of our two bridging criteria relies on presidential vote share to identify members from strongly Democratic and strongly Republican districts and states. However, presidential vote share is only available at the district level beginning in 1952, so for both world wars and the beginning of Korea we rely on just one member bridging criterion for the House.
We included only those roll-call votes cast during Truman’s presidency.
Interestingly, this development seems to have occurred even as Eisenhower, in ending the war, successfully fulfilled his 1952 campaign promise to do so.
However, because the ACU did not issue congressional ratings until 1971, we use positions taken by Americans for Constitutional Action (ACA) for the 88th and 89th congresses. The ACA is a reasonable substitute because it was founded in the mid-1950s as the conservative analog to the ADA, and it also takes a large number of issue positions on a wide range of bills. Scaling a number of interest groups based on their congressional ratings issued from 1969 to 1978, Poole (1981) also shows that the ACA and the ACU were virtually identical ideologically.
Because the end of the Vietnam War coincides with Watergate, we cannot rule out the possibility that personal scandal, rather than armistice, may be responsible for the observed changes in voting behavior.