Mandate Elections and Policy Change in Congress

Mandate Elections and Policy Change in Congress Amy E. Gangl Lawrence J. Grossback David A. M. Peterson University of Minnesota James A. Stimson University of North Carolina at Chapel Hill Paper prepared for presentation at the Annual Meeting of the Midwest Political Science Association, April 1998, Chicago, IL. The authors would like to thank Neal Beck, Jan Box-Steffensmeier, Brad Jones, and participants of the 1997 Summer Methodology Conference for helpful comments and suggestions. Portions of the data used in this paper were collected from the ICPSR data archives in Ann Arbor, MI. Abstract We postulate that members of Congress react to occasional elections which are unusual in outcome, unexpected, and carry a clear message about the direction of voter preferences. Our cases are 1964/65, 1980/81, and 1994/95. That reaction is movement in the direction of the perceived mandate, where both aggregate outcomes and individual behavior deviate from long-term norms under the temporary influence of perception of dramatic movements in voter preferences. We undertake analyses that show aggregate shifts in Congressional outcomes and individual departures from personal equilibria. We conclude with an event history analysis designed to capture the duration of this temporary phenomenon. 1 Despite much skepticism in the literature on individual voting behavior, journalists occasionally declare an election outcome to be a mandate, an election that carries a message about voter desire for policy change. Professional politicians, we assume, see the political world as commentators do, sensing from unexpected and dramatic election outcomes a threat to old assumptions and old strategies, a need to rethink behavior. This article is about such mandates. We ask when they occur, how much change they produce, and, particularly, how long they last. Mandates seem to have a standard dynamic. They begin around election time with perceptions of fundamental and permanent electoral change, invariably overstated. They proceed to fruition in the "First 100 Days" scenario, where a new Congress -- our focus in this article -- sets upon the task of revolutionary policy change. This is cheered by outside advocates of the mandate, encouraging ever bolder steps and threatening any who would impede the new movement with sure electoral annihilation. And then, after some period of time and much policy commotion, members begin to perceive that the mandate was overinterpreted. The voters, whose wrath was so to be feared, turn out neither to be particularly interested nor of one mind about what policies ought to be adopted. And then it is over, revolutionary talk and action become normal politics. Members revert to old assumptions, old strategies, and old behaviors. But impermanent though they are, these occasional episodes have lasting consequences for public policy, far beyond their limited durations. They produce opportunities for policy change of a sort that resists normal politics. A few such "100 Days" account for a good deal of what is important in American public policy. In those brief periods we locate the beginnings of “The Great Society,” (1965), the Reagan Revolution” (1981), and the “Contract With America” (1995).1 Without these “300 days,” post-War public policy in America would be vastly different from what it is. If they are as impermanent as we assert, it is worth asking why seasoned politicians seem to respond to them as if they were lasting, as if the election-night talk about what voters were saying were true. Why don't they, like us, expect a dynamic of short duration, ending in normal politics? The answer, we think, is that it is much easier for us to ignore the message. If wrong, we suffer at most modest embarrassment. Members risk career termination. They face, if wrong, not modest embarrassment, but the public humiliation of electoral defeat. Mandate elections are infrequent. Most members will have little experience with the normal dynamic. Probably experience matters. We expect members who have been through it before to be less strongly affected. The surging or declining electoral margin, for example, is easier to discount if you have seen surges or declines in the past that foretold little. And of course the original “First 100 Days” of 1933 was an event of some magnitude, but it lies beyond our data resources. 1 2 Why Mandates Matter We presume a world in which members constantly update their perceptions of public opinion, observing changes in opinion and evaluating the likelihood that such changes will work their way into electoral effects.2 Uncertainty attends both questions of what opinion is and whether it will have electoral consequences. Members try hard to get it right, we believe, and succeed most of the time. Mandate elections, in this context, come as decisive messages. Because they are unexpected—a defining characteristic— they are news that catches everyone's attention. But more than that, only election outcomes resolve both questions about public opinion, what it is and whether it will have electoral consequence. They tell all what it is, e.g., liberal or conservative, and that it did matter. It is a short step to the crucial premise that it will also matter in the future. The mandate election outcome in one sense is just one in a long stream of messages. But it is by far the most informative in that string. Elections And Voters: Do Electoral Mandates Occur? Pundits and politicians believe in mandates, that elections provide voters the opportunity to speak to a set of issues and that sometimes they do speak. Political science, addressing the issue mainly through the tradition of the voting studies (Sorauf, 1964; Kelley, 1983; Wolfinger, 1985; and Dahl, 1990), is altogether more skeptical. Our tradition is to deny that very many voters very often feel strong enough about electoral issues to register that feeling in their votes. But so long as some voters might wish to send a message with their votes, the micro research on what is typical of voting doesn’t quite answer the right question, the macro issue of whether the aggregate of votes cast is moved at the margin by issue concerns. Indeed, the question of whether mandate elections actually occur is not even itself quite the right question. For if politicians act on the perception of mandate, then in an important behavioral sense a mandate has occurred. And that would be true even if voters were not consciously sending a message. The issue then is are there occasions where there is a consensual understanding that election outcomes are properly interpreted to carry a policy message. This is an open question. The literature on the effect of mandates on policy voting is as skeptical as the voting studies literature. Weinbaum and Judd (1970) and Fiorina (1989) both address this general question and conclude that the policy effect occur not because members of congress change their individual voting patterns in response to the election, but simply because of replacement effects. The shortcoming of this research is that it only compares presidential agreement scores on the most highprofile legislation. More importantly, it explicitly assumes that the effects of the mandate will endure, aggregating the vote patterns for the entire session. In contrast, we expect This is in all respects the theory of Dynamic Representation, fully explicated in Stimson, MacKuen, and Erikson 1995. 2 3 that the effects will dissipate as the session moves forward and the members receive new information about their constituency. What combination of characteristics would we expect to see in a mandate election? We look for three. A mandate election is an occasion where outcomes are (1) unusual and (2) unexpected and where (3) the pattern of voting provides evidence of a clear policy message. These are the attributes that move pundits to declare mandates and move politicians to act on the declarations. An outcome which falls in the typical range is not a good candidate for a consensual understanding of a voter message. There both sides will point to local victories, and the lack of a sweeping verdict will leave ambiguity in any purported message. A victory long expected is one that does not cry out for novel explanation on election night. We look for messages rather in the cases where the outcome is unexpected. Factors such as presidential incumbency and midterm loss for the White House temper interpretation. When incumbent presidents do win re-election or when the White House party does lose seats at the midterm, politics is behaving as it should, and we don’t look for mandates in the result. But no election can be a mandate for policy change without a clear policy message. There we look to candidates and parties who say they are going to do something different or dramatic and voters who encourage them to do so. A Richard Nixon or a Bill Clinton seeking re-election without a clear plan of action has no mandate. A 1980 Ronald Reagan claiming to fundamentally change the course of American government does.3 For the post World War II period we believe three elections possess this combination of attributes. These are 1964, 1980, and 1994, the characteristics of which are sketched in Table 1. But this classification need not rely only on our judgement. The American press is the referee of mandate claims, and a review of it illuminates what an important set of actors said at the time. But clearly it isn’t just the candidate who matters. A Ronald Reagan running on “Morning in America” is 1984 claims no mandate. 3 4 Table 1. Outcomes, Expectations, and Messages in Three Elections 1964 Presidential Message Continue the New Frontier/Great Society expansion of government responsibility. Sweeping Democratic wins at all levels 1980 Presidential Cut taxes and “get government off our backs.” 1994 Midterm The “Contract with America” Outcome Unexpected easy defeat of an incumbent president and substantial gains in Congress Reagan Revolution “Government off our backs” Reagan “The Republic revolution:” first unified Republican control of Congress in 40 years. Republican Revolution Contract with America Newt Gingrich Spin Slogan Symbolic Winner Support for Great Society “Let us Continue” LBJ Which Modern Elections Are Mandates?: The Media Spin The strength of the mandate message is expected to depend, in large part, on media content. Because most people do not often directly experience political events, the content of news stories is usually the primary source from which they obtain knowledge about the political environment. Most politicians understand this, and, thus, look to press reports for important information about public opinion. The process by which media content is translated into public opinion is essentially elite-driven. Elite response to political events, such as the meaning of elections outcomes, gives rise to ideological position taking (Brody, 1991). These strategic responses among elites are communicated through the media, and, in turn, shape the public’s understanding of the meaning and implications of election outcomes. The influence of the content of new stories on public opinion also provides politicians with information about how their constituents expect them to behave. The predominance of news frames in 1980 and 1994 indicating that the electorate had ordered conservative action is likely to have led many legislators to believe that they would be evaluated by their constituents on how they responded to issues that were salient during the campaign. 5 We conducted a content analysis of news paper articles in the New York Times, the Washington Post, the Associated Press, and United Press International that contain references to an electoral mandate from 1980 through 1996.4 Figure 1 presents a raw count of election stories that use the word “mandate” for 1980 through 1996, grouping together election years and the odd years following them. Most frequent mentions of the issue of mandate to no surprise occur in our two cases of this period, 1980 and 1994. But the pattern is actually much stronger. In every other year that the idea of mandate is often broached (essentially in all other presidential elections), it is predominantly in stories that deny an electoral mandate. Figure 1. Media Depiction of Election Results, 1980-1996. Most press reports that discuss a mandate in all of the years that we looked at are based on quoted comments from politicians supporting or refuting such a claim. That a significant number of elites from both major parties interpreted the 1980 and 1994 elections as electoral mandates, and were willing to say so to major news organizations, is unique to the years we examined in our analyses. There is also a much larger number of editorials and opinion pieces discussing the presence or absence of a mandate in 1980 and 1994 compared to other elections year. Finally, 1980 and 1994 represent the only two election years examined in the content analysis in which there was a great deal of discussion in press reports about the presence of a mandate for the victors’ respective issue agendas.5 In 1980, the press 4 We examined the years accessible from Lexis/Nexis, which does not include the 1964 case. We will (but have not yet) analyze the New York Times and the Washington Post to include the media variable in our analyses of the 1964 data. We examined over a thousand stories discussing electoral mandates in all years from 1980 through 1996, including those not included in the analyses conducted here. We found considerable substantive differences in press reports of the policy substance of electoral messages attributed to Reagan in 1980 and the Republican Congress in 1995. In 1984 Reagan’s landslide victory was not interpreted by 5 6 reported that the electorate had granted Ronald Reagan a mandate for the conservative stances outlined in his campaign. The ratio of stories popularizing that interpretation over those contesting the mandate claim remained three to one from the day after the election into May of 1981. On the day of Reagan’s inauguration, Hedrick Smith wrote a piece for the New York Times contending that the electorate had issued Reagan a mandate to “reverse the New Deal’s legacy”: “The nation today arrived at a fascinating and quite remarkable moment in its political history: A 69-year-old-citizen politician who spent most of his working life in another profession has entered the White House and won the opportunity to lead a conservative political reformation that seeks to redirect the role of government in American life and perhaps reshape the national political landscape for the rest of the century (New York Times, January 21, 1981).” Smith's inaugural-day spin is representative of how Reagan’s victory was portrayed in the early months of his presidency. In no other year, except 1994, did we find considerable discussion in the press of an election providing leaders with an opportunity to redefine the structure and priorities of the federal government. In 1994, media reports touted the issue agenda outlined by the House freshman in their campaigns for office as a policymaking springboard for the 104th Congress, and the basis for their electoral mandate. In the months following the elections, Democratic leaders, including President Clinton, voiced little, if any, opposition to the mandate claim. Even Tom Daschle, who had one of the most consistently liberal voting records in Congress, tipped his hat to the Republican’s mandate interpretation on the opening day of the 104th Congress (Washington Post, January 5, 1995). The frame popularizing the elites or press commentators as a mandate. Most of the stories that include discussion of a mandate in 1984 and 1985 contrast the lack of substance in his second-term victory with the conservative message of the 1980 election. In the days and weeks after the election, both the New York Times and Washington Post ran a number of stories criticizing Reagan and his campaign advisor, Jim Baker, for running a substance-free campaign. And three weeks after the election, Reagan conceded in a Times story that he hadn’t yet sketched out an agenda for his second term (New York Times, Nov. 26, 1994). In 1988, the lack of discussion of policy issues in the race between Bush and Dukakis prompted the media to spend the months after the election blaming itself for its focus on image rather than demanding policy substance from the candidates. And President Bush’s largely negative campaign strategy left him open to attacks throughout his term that he lacked a domestic agenda. Following Clinton’s victories in 1992 and 1996, there was relatively little discussion of electoral mandates in press reports. Most stories that included discussion of a mandate after both elections did so only to refute the notion, pointing out that Clinton had failed to win a majority of the popular vote. In press reports even many Democrats interpreted Clinton’s 1992 victory as a rejection of President Bush more than widespread electoral support for Democratic policies. And in 1996, press reports predominantly attributed Clinton’s victory to a prospering economy in combination with the lackluster campaign run by his Republican challenger, Bob Dole. 7 notion that House Republicans were granted a mandate to carry out their agenda remained seven to one from the day after the election continuing well into the first few months of the new session. The first paragraph of a front page Washington Post story by Robert J. Samuelson exemplifies the media’s initial interpretation of the elections. It also illustrates the expectations established among legislators of the 104th Congress during the first couple of weeks of the new session. On the first day of the new Congress, Samuelson wrote: “The new Congress that opens today may be the most momentous since the mid1960’s – not necessarily because it will pass momentous legislation (though it might) but because it may set a new tone and agenda for politics. In the mid1960s, Congress enacted a major tax cut, approved the Civil Right Act of 1964 and created Medicare and Medicaid. Government was to be society’s problem solver. The new Congress is best seen as ending this era (Washington Post, January 4, 1995).” Mandate Duration: An Aggregate Sketch Here we take a first crude look at the Congressional response to mandate elections. For each of the three terms immediately following our three mandates—and for each chamber—we present a moving average of all ideologically polarized votes for the term.6 The individual outcomes are scored by the percent voting the liberal position. Then we compute a moving average of all the votes in the sequence in which they occurred. Because we have no control of the vote content, we expect aggregate outcomes to vary over a considerable range. But the averaging procedure begins to smooth over that variation and converge on a normal outcome. Our purpose here is a modest one. We wish to present a descriptive portrait of the mandate pattern in congressional votes. The highly aggregated votes outcomes make for good description—but the critical test of the mandate response theory awaits the better data of the individual duration analysis to come. 1965: We begin with 1965, our worst case, and one that shows mandate effects in only one of the two chambers. The case which shows no evidence of mandate voting, the 1965 House (Figure 2a) serves to illustrate the null case. After large variation in the The procedure for each vote, borrowed intact from Stimson, MacKuen, and Erikson (1995), is to array the average ADA scores for the members who voted “Yea” against those who voted “Nay.” If the absolute value of the difference between them exceeds 30, the vote is considered ideologically polarized and its direction determined from the ADA scores. The threshold of 30 is approximately the expected ADA split on a pure party line vote. 6 8 initial votes, entirely an artifact of the averaging procedure,7 the 89th House settles into its equilibrium liberalism of around 60 and just stays there. Figure 2a. Aggregate Liberalism, House of Representatives, 1965. Figure 2b. Aggregate Liberalism, Senate, 1965. The 89th Senate, in contrast, shows exactly the expected mandate effect. Like the House, it has an equilibrium liberalism of about 60, but shows a marked deviation away from that equilibrium in the (liberal) direction of the 1964 “message” in the early session votes, an effect substantial enough that it does not dissipate until midway through the It is tempting to notice the very large effects of the first vote or two, but that temptation should be resisted. The votes do not become reliable evidence for a trend until the number of cases grows enough to discount unusual cases. 7 9 term. A story of a “First 100 Days” would be a good fit to the data of Figure 2b. Early votes respond crisply to the electoral message, and then the Senate settles down to business as usual. 1981: Following Ronald Reagan’s startling defeat of incumbent Jimmy Carter in 1980, the 97th Congress—ignoring its nominal Democratic control in the House—set about to give Reagan what he wanted. Most notably in the David Stockman engineered budget package (incorporating Reagan’s tax cut), the Congress engaged in a quick round of policy change that unraveled much of what it had done in the preceding two decades. The Democratic and moderate to liberal House is the more striking case. Here (see Figure 3a) we see a chamber that eventually converged on a pure moderate course of about 50 (and grew more liberal still when the same members returned in 1982) deviating notably toward the conservative message of the 1980 election in early 1981. We expect a more conservative Senate, where the 1980 elections were a bonanza for the GOP. But it too deviated from its conservative equilibrium in the direction of even more conservative voting (Figure 3b). Something needs to explain why a chamber is different at one time than another, and of course we think we know what does explain it. Figure 3a. Aggregate Liberalism, House of Representatives, 1981. 10 Figure 3b. Aggregate Liberalism, Senate, 1981. 1995: The 1994 midterm elections became a “Republican Revolution.” Declared so first by Newt Gingrich and his lieutenants, the spin ultimately was adopted by neutral media commentary. The Gingrich Republicans set out upon impressing America that it truly was a revolution, declaring that its “Contract with America” and much else would be enacted in an historic first 100 days. The data of Figure 4a are stark testimony to legislative success. The really dramatic conservatism of the early session, often credited to the 104th freshmen Republicans, cannot be understood without reckoning why so many Democrats who had campaigned against the “Contract” found it possible to vote for many of its elements. Indeed, the Republicans of the 104th House are so uniformly conservative from beginning to end (averaging about 95% in support of the conservative position) that they cannot produce change. Virtually all of the conservative swing of Figure 4a must be accounted for by Democrats responding to the message of 1994 (and then eventually drifting back to more accustomed positions). 11 Figure 4a. Aggregate Liberalism, House of Representatives, 1995. The 105th Senate shows more sedate patterns, as it did in life. The early session conservatism is not dramatic. We know that the Senate largely watched the ruckus of the Gingrich Republicans in the House, preferring the historic Senate “deliberative body” role. That is what can be seen in Figure 4b. Figure 4b. Aggregate Liberalism, Senate, 1995. 12 Mandate Duration: A Model Of Individual Behavior The aggregate evidence is suggestive, but a far from conclusive story of mandate effects. Our theory is about individuals coping with the message of a mandate. We now elaborate that theory and turn to individual evidence for a test. A Behavioral Theory Of Response Presume members of Congress who are election and reelection seekers and who care about public policy. Because they care about policy outcomes, and because their electoral future is tied to a degree to the approval of primary electorates who also care about policy, their positions are generally not those of the median voter of their reelection constituency, but instead more extreme in the normal direction of their party, Democrats to the left, Republicans to the right. Deviation from the median is always in tension. Too far and too often and the member becomes a former member, “out of touch.” But hewing to the golden mean also creates tension. If the member counts on a more ideological base of party supporters for support that goes beyond mere voting (for example, financial contributions, campaign work, or simply the willingness to publicly endorse), such moderation exposes the member to occasional bad times without a base from which to fight them. From these dual tensions member equilibrium positions emerge. They represent member decisions of where to settle the difference between the two tensions. We observe the effects of such equilibria in roll call votes, where the Democrats line up strongly on the left and the Republicans are strongly on the right, even when both represent districts that are in the center. The U.S. Senate provides the strong case here where members representing the same constituency diverge left and right with their party. Add to this static story an information flow about public opinion, the general disposition of the national electorate, and we have a basis for dynamic equilibria. Members constantly calculate how far is too far—in either direction—and update such calculation with the flow of news about the changing dispositions of the public. 8 As members perceive change in public opinion, the compromise needs to be constantly renegotiated. A key feature of such logic is the understanding that members attend to Members, of course, care about public opinion in the district, the electoral unit that matters. But it is an easy matter to know where the district is, relative to the nation, and then assume that national trends will produce similar movements in the district. 8 13 future elections. These introduce a high level of uncertainty and continuing pressure to observe, think, and act to avoid the snags of electoral failure in a dynamic electorate. Elections as Information Flow: Elections are the target of this model of member behavior. Their typical function, we presume, is to serve as a check on strategy— normally as a confirmation of it. Members target elections with their position-taking behavior and then observe how well they did. Typically the result is reelection, but with tension from both primary and general election constituencies during the process. The tensions make members wary; the success provides evidence of the soundness of their strategies. All of this presumes the normal congressional election, more local than national, more influenced by reputation for constituent service than by positions or issue controversies. The mandate election is another story. There members will experience much greater threat in the general election (if they are on the wrong side of the mandate), a more issue-conscious electorate, and a more national result. Losers don’t return for the next session, so we need not ask how they might have altered their behaviors. But winners include both those advantaged by and those disadvantaged by the national trends, and both of them have seen information that suggests rethinking strategy. Mandate advantaged winners have new confidence in the electoral appeal of their off-center views. Mandate disadvantaged winners emerge from the experience still winners, but often shaken by shrinking pluralities and a new sense of threat from being on the wrong side of the electorate’s new path. Whatever they truly meant in the minds of voters, mandate elections come to be interpreted as permanent shifts of electoral preferences. “American politics will never be the same again” is the standard election night story following a mandate election. From a distance, in time and personal stake, it seems relatively easy to discount that story, to presume that things will return to normal. But members have a very big personal stake and do not have the luxury of distance in time. They have to act at the moment. They must entertain the idea that the political world truly has changed and ask how they must change to be survivors in the new environment. Since the center pushes and pulls on both the advantaged and disadvantaged, we expect both to shift away from established equilibrium voting patterns under the force of a perceived mandate. If the mandate were true—the shift genuinely permanent—then the equilibrium shift would be too. But the mandates we study, whether real or not, are always seriously exaggerated as election results are fresh, and thus we expect a decay of the mandate induced movement away from equilibrium as it become apparent in time that the change was exaggerated or one of temporary duration. Members scan the environment and eventually begin to see signs that the normalcy of the past has returned. In the aggregate we expect mandate effects to decay steadily over time. At the individual level we conceive the same effect as a dichotomy. Members are in a mandate state if they perceive a changed world and deviate from equilibrium. They are in a normal state when they cease to show the influence of the election event. 14 Conditioning Influences Our theory to this point applies equally to all members. But it has elements that suggest that different members will react differently to the stimulus of a mandate election. Three of these are particularly important. We expect asymmetries of threat along party and ideological lines, we expect that members who have witnessed the most striking electoral effects to respond more strongly, and we expect that political experience will moderate response. Party: A perceived moving center should affect the behavior of those on both sides of it. But there is asymmetry in the nature of the treats and opportunities of the advantaged and disadvantaged parties. The disadvantaged, or simply “out,” party members sense that their current behaviors may expose them to future (general) electorates in which their views are a disadvantage that may have fatal consequences. If the political world has changed and they are on the wrong side of the change, they should rebalance their positions in line with the now greater problem of the general election constituency. There is little cost to a bit of moderation in the new Congress. If the mandate proves ephemeral, they can revert to more comfortable positions. And if it does not, they set the stage to begin to establish a new and more viable position. (And if they are wrong about the mandate, the initial votes of the new Congress will conveniently be far from voters’ recent memories at the time of the next election.) “In” party members will sense an opportunity for free expression of their views, now seen to electorally advantaged. With the center seen to be moving toward their preferred position, the need to compromise is relaxed. Their threat—more apparent in the long than the short-run—comes from run-away ideologues on their own side who, having over-interpreted the electoral message, now demand purity and commitment of all. In party members are pulled by their primary constituencies into positions which can prove to be embarrassing if the mandate turns out not to be a permanent shift. Electoral Effects: Incumbent members are highly sensitive to the size of their victory margins. Either in or out party members who see big differences in the mandate election result should take the evidence more seriously than those who do not. Newly elected members, lacking any such baseline, may be most inclined to (over)credit their election to the policy attitudes of voters. This will prove hard to model, but the sort of effect that we anticipate is that out party members who just barely survive the election after previously comfortable electoral margins should be those most likely to change their behavior in its aftermath. For them the mandate story is not just opposition rhetoric, it is a quite immediate threat to their electoral career. Seniority: We can think of the world as composed of electoral veterans and novices. The veterans have seen party surges come and go in the past, and survivors of adverse times, it should be easier for them to discount the most recent case. Novices who lack such experiences seem much more likely to be moved the claims of the moment, to be they have witnessed a profound change, not merely another prelude to normal politics. All of these expectations are testable, and they will structure our analyses to come. 15 Testing The Theory: Modeling Individual Duration Effects To move from theory to test we need a means of observing when the appearance of mandate behavior is present or absent. For data we have the roll-call votes cast by all members of both houses for three mandate Congresses. What is needed is a means to translate a “yea” or “nay” vote into evidence of mandate influence, present or absent. The conception of the mandate state can be stated simply: a member is in a mandate state when he or she (1) deviates from personal equilibria (2) in the direction of the mandate (3) at the beginning of a session following a mandate election. That requires that we observe a sequence of votes with clear ideological content and polarity, estimate personal equilibrium positions for each member, and then observe whether a cumulated pattern of votes at the beginning of the session diverges significantly from that equilibrium or not. The result will be a sequence of binary codes in which each member is classified as in the mandate state or not as the session progresses. Research Design Coding the Votes: Our strategy for coding the ideological content of the votes is taken from Stimson, MacKuen, and Erikson (1995). It examines each individual roll call to determine if it is ideological and in which direction by mapping the vote distribution onto member ADA scores. A vote is liberal, for example, if members who vote the “yea” position have higher average ADA scores than those voting “nay.” The standard employed for “higher” is that the split between “yea” and “nay” must be greater than 30, an arbitrary threshold for judging ideological polarization. Where the threshold is not met, we presume that the voting alignment might be accidental with regard to ideology, the leading cause for concern being party against party votes that might have nonideological causes. We then select only the ideological subset of votes and aggregate them into sequences of 10, where we can observe moving member ideological position, if it occurs.9 Each block of 10 votes constitutes a legislative period—albeit one that does not map uniformly onto regular units of time. For each member in each period we then code the member as being in a mandate state if (1) the member was in the mandate state on the previous block (or it is the first block), and (2) the member’s observed liberalism for the block deviates in the direction of the mandate and is more than two standard deviations from his or her equilibrium liberalism. The equilibrium is estimated from the member’s This is a compromise between the two extremes of complete cumulation of votes, beginning to end, with a running “batting average,” and one vote at a time analysis. The former may exaggerate the effects of a few early session votes. The latter is so unreliable—in effect, all members are required to be 100% liberal or 100% conservative on a single vote—that deviation from equilibrium cannot be reliably discerned. 9 16 observed behavior in the second term of the same Congress.10 These decisions are each true to our interest in classifying individual members as under mandate influence or not. We restrict classification of deviations from equilibria, that is, to consistent early session behavior and to movements in the right direction only. Because we wish to be sure that we are modeling the election effect only, one particularly conservative decision is that mandate states terminate with a block of normal votes, and may not resume thereafter. We quite probably lose some genuine mandate behavior from this rule, combined with random variation and a small n. This coding scheme may best be understood through examples. Figures 5-7 present the voting patterns of Gary Hart, Bill Bradley, and Paul Tsongas in 1981. the solid line horizontal line is the second term equilibrium, the smooth dashed curve is the two standard deviation bound, and the jagged curve is the actual vote pattern. The vertical line denotes the vote block at which the member returns to equilibrium. In each case the Senator begins more conservative than in the normal year and slowly moves towards their equilibrium as the session progresses. Figure 5. Mandate Coding, Gary Hart 1981. We have tried other approaches, such as using first term averages (which presents circularity issues) or late first term averages. Insofar as we can tell, variations in these criteria do affect individual member classifications, but do not alter the general pattern of mandate behavior. 10 17 Figure 6. Mandate Coding, Bill Bradley 1981. Figure 7. Mandate Coding, Paul Tsongas 1981. We wish to ask, “How long do mandates last?” We now have the appropriate data in hand, sequences for all individual members of the form “111000000…” where the “1’s”—standing for ten vote blocks where the mandate influence is “on”—delimit the period of mandate influence. Our task now is to model this individual behavior. We begin with a descriptive look at some simple relationships and then proceed to a full model of mandate duration. 18 Patterns of Impact and Duration We are ready now to inquire into the mandate phenomenon. We have in hand 1,566 cases of members who served in either house of Congress in one of the three mandate terms, 1965, 1981, and 1995.11 We begin our sketch in Figure 8 with a look at the three terms, stratifying members into those who did or did not suffer a decline in their vote share over the previous election. We conceptualize behavior as consisting of only two states, voting normally (that is, close to personal equilibria) or deviating in the direction of the mandate, which we label “affected.” We expect those who lost vote shares to be more threatened by the mandate, more likely to deviate from normal behavior in fear that their careers might be cut short if they do not. And the figure provides some evidence that this is true.12 In the early periods, when the mandate is at full strength, those who lost vote shares are in each year more affected—more likely, that is, to deviate from their normal voting patterns in the direction of the mandate. In each year also the effect dissipates, leaving no differentiation between the threatened and unthreatened in the long run. Figure 8. Changes in Probability based on Lost Vote Share. In Figure 9 we continue to focus on electoral threat, this time looking at behavior in the two houses of Congress. We have no strong priors on chamber differences, but one might expect electoral threat to be generally weaker in the Senate, where members have six years before the next accounting at the polls. There is little differentiation in the 11 The is the expected (435 + 100)*3 minus a handful of cases lost due to insufficient data. Note that for these analyses the direction of the effect is whatever the mandate signal was, liberal in 1964 and conservative for the other cases. From the figures to come the Y axes map response to the mandate—in whatever direction it was. Thus up is always toward the mandate signal and down is away from it, toward individual member normal behavior. 12 19 Figure, where members of either body are strongly affected if their vote shares are declining, but no substantial differences exist between the two houses. Figure 9. Changes in Probability based on Chamber, Vote Share. We have postulated that less senior members are more influenced by the mandate phenomenon. Not likely to have experienced startling and unexpected election outcomes, they are likely to believe the mandate “spin” and act as if the election foretold a permanent shift. More experienced members would have seen cycles of such elections before and find it easier to discount the strong interpretation of them. In Figure 10 we break out incumbents by the number of two year spans they have served—terms for members of the House, and an artificial equivalent for Senators—into those serving up to their fifth term and the more senior members re-elected more than five times. There we see the expected effect. The most senior members do not respond as strongly to the mandate stimulus as the less experienced. Fewer catch the mandate fever in the first place and then they return to a closer to normal level than do the less experienced members. Figure 10. Change in Probability Based on Seniority. 20 Our expectations about member ideology involve a mix of considerations. Those who are most moderate we think often are so because they represent heterogeneous and insecure constituencies which would not tolerate more extreme positions. They should therefore be highly sensitive to the shifting center implied by mandate. More extreme ideologues, on the other hand, have little to gain from a strategic shift of position. Partly because their extremity is probably associated with a safe electoral base and partly because their claim to moderation lacks credibility, they have less to gain by moving toward the mandate effect in the early session. In Figure 11 we break members into three classes from most moderate to most extreme, where the “folded” ADA score, (absolute value of actual ADA rating –50, that is) is our criterion. Moderates (ADA between 25 and 75) indeed prove to be much more responsive to the mandate than our relatively in between (10-25 or 75-90), who are in turn more responsive than the most extreme (0 to 10 or 90-100). Figure 11. Change in Probability Based on Ideology. If we give up exploring the dynamics of mandate response (for the moment) and simplify our question to whether or not members respond to the election message in the first block of votes, we can put together our whole explanatory structure for a multivariate look at the phenomenon. We consider the effects of loss of victory margin, this time measured as inter-election change in margin percent,13 from seniority (number of terms served), ideological extremity (the folded ADA score), adding specification dummies for party (Democrats), chamber (House of Representatives), and year (dummies For U.S. Senators not at risk in the current election, we estimate the margin change from the aggregated change for the Senator’s fellow partisans in House races in the state. For newly elected members, this was based on the performance of the candidate of the same party in the previous election. Alternative coding rules were tried as robustness checks and in no case did the substantive conclusions change. The Senate electoral data was from Elections Research Center (1992, 1994) and the House data is from King (1994). 13 21 for 1981 and 1995). Our dependent variable is a dichotomy (affected or not) and so we estimate the model with a logit setup. Probability of being affected by the mandate, we see in Table 2, is heavily influenced by the threat of lost re-election margins, very heavily by ideological extremity, and not at all (in this analysis) by seniority. Among the specification dummies, the most notable effect is that 1981 shows much stronger response than do the other two years. Democrats respond more than Republicans, but given the asymmetry of our three cases it would be unwise to put much interpretation on that fact. Model fit for discrete choice models is difficult to address, but the percent of the cases correctly classified is 82.3 percent and the chi-square test clearly demonstrates that the model as a whole is significant. We now have a picture of who responds to mandate elections. The responders are political moderates who have seen evidence of the most relevant sort, declining electoral security, that the mandate needs to be treated as a real phenomenon. On seniority we have conflicting evidence. Table 2. (Logit) Model of Whether Member Shows Mandate Effect in the First Period Variable Change in Victory Margin Seniority (Number of Terms Served) Ideological Extremity (Folded ADA Rating) Party (Democrat) Chamber (House) Year (1981) Year (1995) Intercept N Likelihood Ratio χ2 /p Coefficient -0.738 0.002 -0.028 0.272 0.058 1.223 -0.138 -1.436 1566 142.62 / 0.000 Robust Standard Error 0.247 0.017 0.005 0.129 0.197 0.176 0.228 0.289 Z -2.984 0.915 -5.323 2.108 0.294 6.986 -0.603 -4.960 We highlight the effect of two key parameters from Table 2 in Figure 12. There we hold constant the specification variables (at their means) and examine the effect of jointly varying ideological extremity and change in voting margins. The Figure displays 22 change in margin over its full possible range (which greatly exaggerates the empirical range) for three hypothetical ideological extremity cases, the pure moderate (ADA rating 50), the in between ideologue (ratings of 25 or 75), and the full ideologue (ratings of 0 or 100). Figure 12. Percent Affected by Ideology and Vote Share. We know something now about who responds. Now we turn to dynamics, asking how long lasting is the response. To answer that question we model the duration of member response with an event history formulation. Estimating A Discrete Time Duration Model Prior work looking into congress’ response to mandate elections compared the member’s voting behavior across entire years. We believe that it is a mistake to assume a given duration for the effects of a major policy signal. We therefore use an event history analysis to explore the differences in the length of the effects across members of congress. Event history analysis allows us to assess both the patterns and causes of change in political behavior (Yamaguichi, 1991). Specifically, we are interested in knowing how the duration spent in one social state affects the probability some entity will make a transition to another social state (Box-Steffensmeier and Jones, 1997). In the analysis at hand we are interested in how the duration spent in the mandate state affects the probability that a member of Congress transitions back to his or her equilibrium voting pattern. We are also interested in how the conditioning factors of party, electoral effects, ideology, and seniority affect the movement from the mandate state back into a normal voting pattern. Event history models predict the probability, termed the hazard rate, of an event occurring at a specific time given that it did not previously occur. We define the event of 23 interest as a member's movement from the mandate state into a normal voting pattern. We determine this by the member's voting behavior in each legislative period (ten vote block). Given that these periods are measured at discrete intervals, our research centers on whether an individual exits and, if so, in which time period the non-repeatable event occurred (Singer and Willet, 1995). The nature of our data led us to estimate the probability of exiting the mandate state via a discrete time hazard model. Several characteristics of our data help to justify our choice of a discrete time specification. First, the data contain a large number of ties, or in our model a large number of members exiting the mandate state during the same legislative period. Continuous time methods often yield biased estimates when a large number of ties are present. Second, the discrete time model easily handles our time varying media variable. Finally, our time intervals, i.e. vote blocks, are relatively short and thus mimic the underlying continuous time process. The basic estimation technique is a standard logit model, using what amounts to binary time series cross section data. Each member of Congress has a line of data for each time period they are in the mandate state and one line for the period in which they exit back to their equilibrium voting pattern. The media variable is the only variable that varies over time. The Logit Based Hazard Model The discrete time hazard is defined as the probability that event T occurs at time t, given that the event has yet to occur. If we let λ(t) be the hazard function we can state this formally as: λ (t ) = P(T = t | T ≥ t ) This probability distribution can then be generalized to include a constant and a λ (t ) = P(T = t | T ≥ t ;α , β ′Χ ) vector of covariates. Cox (1972) has shown that the continuous time proportional hazard model can be modified such that the discrete time equivalent has a logistic dependence on the covariates. 1 1 + exp −[α + β ′Χ ] We can then invoke the standard logistic transformation. λ (t ) = 24 ln λ (t ) = α + β ′Χ 1 − λ (t ) Estimating Duration Dependence The simple logit specification assumes duration independence (Beck 1996). That is, it assumes the hazard rate is not an explicit function of time. Since we expect the passage of time to affect the probability of exiting the mandate state we had to incorporate a method of estimating the main effect of time (Willet and Singer 1993). We will avoid two techniques, including a dummy variable for each legislative period and parameterizing the dummy variable as lying on a low order time polynomial. Beck (1996) discusses the limitations of these methods at length, the main weakness being that they are inflexible and may not pick up the correct pattern of duration dependence. We have chosen the method of fractional polynomials to capture the time aspect of our data. Fractional polynomials estimate a family of curves using power terms that are limited to a small predefined set of integer and non-integer values (Royston and Altman, 1994). The method is designed to capture a rich set of curves while providing a parsimonious means of parameterizing the time effects.14 A fractional polynomial of degree m has m integer and/or fractional powers p1 < p2 < … < pm, , and is of the form β0 + β1 x ( p1 ) + β2 x ( p 2 ) + ...+ βm x ( p m ) Where for a power p,  x p if p ≠ 0 x = lo g x if p = 0 p This form can be generalized to allow for repeated powers. The powers are restricted to the set {-2, -1, -0.5, 0, 0.5, 1, 2, 3}. This limitation is not intrinsic to the approach, experience has shown that it is not worthwhile to include extra powers. A systematic search is conducted for the best power or set of powers that fit the time variable. 14 Beck (1996) recommends the use of a cubic smoothing spline as an alternative to a fractional polynomial. Both techniques represent a non-linear means of capturing the main effect of time and neither stands out as the best method at this time. Our choice of the fractional polynomial is based primarily on its ease of use (STATA includes a set command) and the fact that it has yet to be used in the political science literature. In the future we intend to do a careful comparison of these two techniques. 25 Event History Results We believe that the flow of information members of congress receive via the media is important to understanding when they will return to their normal voting patterns. These flows of information for 1980-81 and 1994-95 are depicted in Figures 13 and 14. In 1980-81, the depiction of the election as a mandate was most prominent immediately after the election. After November, both the number of stories and the proportion proclaiming a mandate declined steadily. For 1994-95, there is a similar decline in the quantity of stories, but the perception of a policy mandate increases early on in the session. It takes until well into the session for the media coverage to turn away from there being a clear mandate. Figure 13. Media Coverage, 1980-1981. Figure 14. Media Coverage, 1994-1995. 26 The media information now enters our model as a time varying covariate. It measures the balance of stories proclaiming a mandate in each of the legislative periods. These graphs show the extent to which the media’s interpretation plays up the idea that a clear policy message came out of the election returns. They also show how the message dilutes over time as competing interpretations of the election come into play. The Dynamics of the Individual Responses Before modeling the individual transitions, we can gain some insight into the dynamics of a policy mandate by looking at the length of time members of congress were voting in accord with the mandate message. Table 3 presents the number of members affected for the given number of legislative periods. Clearly the majority of members remain unaffected in all three congresses. This is not unexpected given our belief that members of the party supported by the mandate are unlikely to have to adjust their voting behavior. In each of the three years we see that a good number of members are affected for a short number of votes, and, more importantly, that a sizable number of members are affected for a lengthy period of time. In 1981 nearly 75 members of congress voted in a more conservative direction for over 150 votes. The average number of periods affected also shows the extent to which members adjusted their policy behavior early on in the period. In 1981 and 1995 the average number of votes cast in accordance with the mandate was close to 30. Given the importance of early votes in a typical session, these results lend support to the idea that members reacted to the mandate message and that the mandate will have real policy consequences. 27 Table 3. The Total Number of Legislative Periods Members Were Effected Periods Affected 0 1 2 3 4 5 6 10 11 12 13 14 15 20 24 42 45 62 1965: N = 530, Mean Number of Periods Affected = 0.84 1981: N = 531, Mean Number of Periods Affected = 2.66 1995: N = 538, Mean Number of Periods Affected = 2.89 1965 470 21 1 2 5 1 1 1 1 1 18 8 1981 374 42 16 8 13 1 1 1 2 59 0 0 0 0 14 1995 495 10 1 0 2 1 1 0 0 0 0 1 1 1 0 3 1 21 We can get a second view of the duration of the mandate effects from Figure 15. Figure 15 shows two things. The solid line is a prediction of the period within which a member exits the mandate state based solely on time (the fractional polynomial). The circles represent the residual between the time only prediction and a prediction based on the entire event history model. Two things deserve mentioning. 28 Figure 15. Effects of Time on Probability of Returning to Equilibrium Fractional Polynomial (-2 .5), adjusted for covariates 5 Estimate from full model 0 -5 Estimate based on time only -10 1 Period 62 First, the pattern of how long members are affected is clearly seen in the predictions. A sizable number of members are affected by the mandate for the first portion of the session. The number of members affected drops off quickly leaving only those who are affected for close to the entire first half of the congress. The second point regards the importance of our having parameterized the main effect of time via the fractional polynomial. Clearly time matters. Members are more likely to exit the mandate state the farther they from the day of the election. Time, however, does a poor job of predicting when members exit early in the session. Other factors matter in how long members are affected, factors we capture with our multivariate analyses. Modeling the Individual Effects In addition to the media variable and the specification for the effects of time, we have included the same set of explanatory variables as in the logit estimated above. The 29 only difference is the single variable indicating that the year is 1995. 15 Our expectations for the effects of each variable in the event history analysis are the same as the logit model which assessed only whether a member was affected or not by the mandate message. It should be noted that interpretation of the coefficients is the opposite of what it was for the logit model because we are now modelling the probability of exiting the mandate state. For example, in Table 2 the negative coefficient for ideological extremity meant that a member who is more extreme is less likely to be affected by the mandate. In Table 4, the positive coefficient means that the more extreme member will exit the mandate state more quickly. In other words, coefficients with opposite signs in the two models have the same substantive interpretation. Table 4. Discrete Time Hazard Model of Duration of Mandate Effect Variable Time Period One Time Period Two Balance of Media Coverage Change in Victory Margin Seniority (Number of Terms Served) Ideological Extremity (Folded ADA Rating) Party (Democrat) Chamber (House) Year (1995) Intercept N Likelihood Ratio χ2 /p Coefficient 0.021 -4.810 -0.166 0.382 0.039 0.042 0.927 -0.136 -1.182 6.166 1566 769.64 / 0.000 Robust Standard Error 0.003 0.641 0.031 0.234 0.017 0.005 0.137 0.184 0.375 1.555 Z 6.986 -7.505 -5.382 1.633 2.262 8.621 0.676 0.459 -3.154 3.966 Note: Time periods 1 and 2 represent the best fitting fractional polynomials to dependent variable “period.” Period 1 = x-2 and Period 2 = x .5 where x = period/10 This is only because the media data for 1965 has not been coded yet. We fully intend to include that year in this estimation. 15 30 Table 4 shows that the probability of exiting the mandate state varies over the course of the session, both as a function of time and as the messages being sent by the media change. As the balance of media coverage changes from clear support for a policy mandate to reporting that the Congress has misinterpreted the results, the members listen and react. Clearly, they pay some attention to the messages they receive from the media about the state of public opinion. The second main difference between these results and those reported for the logit regards the importance of seniority, which is now significant. Seniority may not predict whether or not the member is affected initially, but serving more terms does make the member more likely to return to their equilibrium quickly. The learning curve for more senior members appears to be relatively short. Results for the other conditioning factors more closely parallel the findings reported in the initial logit model. The member’s electoral status matters. Members who saw their margin of victory decrease are affected by the mandate for a longer period of time. Members who are more extreme ideologically return to their normal voting pattern more quickly than do moderate members. The only variable included for specification purposes which reaches statistical significance is the dummy variable for 1995. The members serving in 1995 return more slowly than the members who served in 1981. This may be a function of the number of roll call votes in 1995 as much as an indicator that 1995 contained a stronger electoral signal. The House in 1995 cast almost 800 roll call votes, dwarfing the numbers cast in other chambers and in other years. Since the coding of the time that it takes to move back is a function of the number of roll call votes, and this is not normalized across elections, this result only means that it took more roll call votes to move back. If the variable were coded in time units, there probably would be no effect for the different years (as evidenced in the logit results). Conclusion The results reported here begin to demonstrate the importance of the policy message members of congress take from significant election returns. While political scientists may disagree over the possibility of finding a policy mandate in electoral returns, members of congress seem to have little difficulty finding one. Members adjust their behavior, in the direction of the policy message, as a means of insulating themselves from the changing electoral landscape. The changes members make in their voting behavior are rational and follow a predicted pattern. The greater a member’s electoral safety, seniority, and ideological extremity, the more quickly they return to their normal voting patterns. In addition to the message from the voters, our results also suggest that members of congress respond to how the media interprets public opinion. The balance of media coverage significantly affects the length of time members are affected by the mandate. 31 The results suggest that the media also alter how politicians perceive messages from the electorate. This does beg the question of the direction of causality between Congress and the media. We presume that the media messages alter how the members perceive the electoral results. It is possible that the true relationship is more complicated, where the actions of Congress alter the coverage, which, in turn, alters the perceptions of the members. Tests for this type of self-fulfilling prophecy needs to be developed. Several additional avenues for research are called for. A first step would be a closer examination of other elections. While the media depictions of the previous election is an important indicator of the strength of the message received, it is worth asking whether or not these voting patterns, at least in the aggregate, exist in other elections. They may not fit our criteria of a mandate election, but if the pattern of policy votes exists, it may create doubt about the importance and special status of these mandate elections. Secondly, it may be possible to focus the content of the mandate more directly to specific policy areas. We believe the effects exist for general policy liberalism, but it could be that they are directly associated with specific policies that framed the campaigns of that year. Beyond the important result that mandate elections lead to changes in congressional policy making, our research has important implications for broader questions in political science. Perhaps the most important is the evidence for the individual level mechanism behind the aggregate results of dynamic representation found by Stimson, MacKuen, and Erikson, (1995). Members of Congress have perceptions of what the ideological make up of their constituency is and they alter their policy votes in anticipation of the electoral consequences. Mandate elections are the most forgiving of test cases for these results. The ideological message is clear and the electoral consequences are clearly evident. 32 References Allison, Paul D. 1982. “Discrete-Time Methods for the Analysis of Event Histories.” In Samuel Leinhardt (ed.) Sociological Methodology. San Francisco, CA: JosseyBass, Inc. Beck, Nathaniel. 1996. “Modeling Space and time: The Event History Approach.” Draft prepared for Elinor Scarbrough and Eric Tanenbaum (eds.), Research Strategies in the Social Sciences. Box-Steffensmeier, Janet M. and Bradford S. Jones. 1997. “Time is of the Essence: Event History Models in Political Science.” American Journal of Political Science 41(4):1336-1383. Brody, Richard A. 1991. Assessing the President: The Media, Elite Opinion, and Public Support. Palo Alto, CA: Stanford University Press. Congressional Quarterly, Inc. 1994. Congressional Quarterly Almanac. Washington, DC: Congressional Quarterly Press, Inc. Dahl, Robert A. 1990. “Myth of the Presidential Mandate.” Political Science Quarterly 105(3):355-372. Elections Research Center. 1992. America Votes. Vol. 20. Washington, DC: Congressional Quarterly Press, Inc. Elections Research Center. 1994. America Votes. Vol. 21. Washington, DC: Congressional Quarterly Press, Inc. Hershey, Marjorie Randon. 1994. “The Meaning of a Mandate: Interpretations of “Mandate” in 1984 Presidential Election Coverage.” Polity 27(2):225-254. Kelley, Stanley Jr. 1983. Interpreting Elections. Princeton, NJ: Princeton University Press. Kiefer, Nicholas M. 1988. “Economic Duration Data and Hazard Functions.” Journal of Economic Literature 26(June):646-679. King, Gary. 1994. Elections to the United States House of Representatives, 1892-1992. Ann Arbor, MI: Inter-University Consortium for Political and Social Research. King, Gary and Lyn Ragsdale. 1988. The Elusive Executive: Discovering Statistical Patterns in the Presidency. Washington, DC: Congressional Quarterly Press, Inc. Royston, Patrick and Douglas G. Altman. 1994. “Regression using Fractional Polynomials of Continuous Covariates: Parsimonious Parametric Modeling.” Applied Statistics 43(3):429-467. 33 Singer, Judith D. and John B. Willett. 1993. “It’s About Time: Using Discrete-Time Survival Analysis to Study Duration and the Timing of Events.” Journal of Educational Statistics 18(2):155-195. Stimson, James A., Michael MacKuen, and Robert Erikson. 1995. “Dynamic Representation.” American Political Science Review 89(2):543-65. Wanta, Wayne and Yu-Weid Hu. 1993. “The Agenda-Setting Effects of International News Coverage: An Examination of Differing News Frames.” International Journal of Public Opinion Research, Vol 5. Weinbaum, Marvin G. and Dennis R. Judd. 1970. “In Search of the Mandated Congress.” Midwest Journal of Political Science 14(2):276-302. Wolfinger, Raymond E. 1985. “Dealignment, Realignment, and Mandates in the 1984 Election.” In Austin Ranney (ed.) The American Elections of 1984. Washington, DC: American Enterprise Institute, Duke University Press. 34