Winning the votes for institutional change: how discursive acts of compromise shaped radical income tax reforms in the United States

ABSTRACT Since the 1980s, many governments in the Western world have implemented radical income tax cuts which have become associated with soaring levels of inequality. The literature has focused on institutional accounts to explain these developments. However, institutions alone cannot account for the emergence of societal and political support necessary for radical change of this kind. Therefore, this paper explores the role of communication techniques directed at voters, interest groups and legislators to enable radical reform. Based on a content analysis of Congressional debates for the Reagan and Bush tax cuts, contextualised with archival documents from Presidential Libraries, this study shows the critical relevance of strategic acts of compromise to shore up legislative and voter support for radical tax cuts. It finds (a) that change actors have several different acts of strategic compromise (incorporation, compensation, and reconciliation) at their disposal which they use at different points in the legislative process. That (b) the most successful strategies consistently link the coordinative discourse (bargaining with interest groups behind closed doors) and communicative discourse (directed at the public and the minority congressional party). And that (c) change actors learn how to the use successful combinations of compromise over time and thereby enhance reform stability.


Introduction
Since the 1980s, many governments in the Western world have implemented significant tax cuts. Most of these reforms not only radically reduced corporate tax rates, but also curbed the top personal income tax rate and reduced taxes on financial incomes such as capital gains and dividends. While mostly designed to revitalize saving and economic growth, these reforms also had considerable effects on the distribution of income within societies. Tax progressivity declined, contributing to rising levels of inequality and a concentration of income among the top 0.01 percent (Piketty 2014;Piketty and Saez 2007). While a small group of individuals made significant gains, the vast majority received small gains and traditional values of ability to pay and redistribution were cast to the sidelines of the system (Scheve and Stasavage 2016;Limberg 2019). This one-sided approach to income tax reforms since the 1980s raises the question how administrations communicate radical and unequal tax reforms to their constituencies.
Institutionalism has so far emphasised continuity over change in taxation. The literature stresses that institutional context and long-term historical processes shape (and constrain) the evolution of tax systems (Martin, Mehrotra, and Prasad 2009;Morgan and Prasad 2009), or show how actors work around institutions outside of the legislative process enabling incremental change (Patashnik 2003;Howard 1997;Hacker and Pierson 2005). While providing important insights into tax developments, these accounts do not tell us much about the conditions for radical and unequal income tax cuts. Even Prasad's (2006) historically detailed investigation of policy makers' interaction with several different societal groups lacks a detailed analysis of how policy makers communicate with these groups. Therefore, this article examines the communication strategies of administrations and their specific discursive interaction with voters, interest groups and the minority congressional party.
This article argues that compromisesdefined as strategic acts of conflict-resolution (Bellamy 2012, 443-444) can function as critical discursive instruments which enable agreement with different actors in the legislative process. Its role in cross-party agreement has recently been examined (Bailey et al. 2021), but this paper goes further in exploring the role of different types of compromises, how these shape policy outcomes and how actors learn to combine different kinds of compromise to achieve policy stability.
To conceptualise interaction of policy makers, interest groups, the minority congressional party and voters, the paper embeds compromise into a larger theoretical framework of discursive institutionalism (DI). Different from many other ideational approaches, DI focuses on the procedural elements of discourse as well as the interaction of discourse with the institutional context within which it is generated (Schmidt 2008(Schmidt , 2010. Public policy research has recently discovered DI as a useful framework to make sense of the procedural issues of policy change (Bailey et al. 2021;Crespy and Schmidt 2014;Fitch-Roy, Fairbrass, and Benson 2020;James, Kassim, and Warren 2021;Widmaier 2016). One critical finding is that policy-maker agency sits in between "background ideational abilities"which allow actors to work with a given institutional and ideational frameworkand "foreground ideational abilities"which allow actors to think critically beyond the institutional context. Successful communication techniques develop novel approaches to a policy problem, but also acknowledging the traditional sentiments that have evolved within a particular institutional and ideational environment (Crespy and Schmidt 2014;Fitch-Roy, Fairbrass, and Benson 2020). Moreover, discourse is most successful if normative ideas about what is right are consistently integrated with cognitive ideas about the functioning of a policy instruments (Seabrooke and Wigan 2016). By combining these insights with concepts of policy learning (Dunlop and Radaelli 2013), the article considers both the content and the context of policy discourse as an act of learning.
This approach is tested on the basis of a case-study of income tax reforms in the U.S. The Bush and the Reagan tax cuts were the largest and most unequal tax reforms worldwide (Piketty and Saez 2007). Institutionalist accounts generally predict little change of taxation for the U.S. Due to the unusually high degree of dispersion of power in the U.S. polity, radical change is unlikely (Béland and Lecours 2014;Prasad 2006). The fact that radical change still occurred provides us with an interesting case to examine what drove change aside from institutional factors.
The relevance of strategic compromise is tested through a two-step empirical approach. First, a content analysis evaluates whether arguments of Republican and Democratic Congressmen for (or against) tax cuts became more similar. In a second step, the paper studies strategies to win support, how they changed over time and how this stabilised tax reforms. The study contrasts four reforms with different degrees of stability. While three reforms implemented lasting change, one reform (Reagan's first tax cut in 1981) did not withstand the test of time. Contrasting "stable" and "unstable" reforms allows for a test of the conditions of stability.
The argument proceeds as follows: The next section introduces the reforms that have been conducted by many OECD countries since the 1980s and how U.S. reforms compare to the general trend. It is followed by a review of the literature and a theory section which introduces an approach of discursive interaction and compromise for radical reforms. Next, the methodology is introduced. Section six traces the debates on tax reforms under Reagan and Bush Jr. and in the conclusion, the paper draws theoretical conclusions for the nature and the role of compromise in policy change.

Radical tax reforms and rising levels of inequality in Western tax systems
Since the 1980s, income tax policies followed a surprisingly similar pattern across OECD countries. Many governments opted for tax cuts for corporations and top incomes comprised of five core elements: corporate income tax cuts, across-the-board income tax reductions, reductions in tax brackets, lower top marginal rates and cuts of dividends and interest income tax (Ganghof 2006b;Hakelberg and Rixen 2020).
The reforms came about as part of a larger conservative push to shift tax priorities from equity and economic growth towards lower tax to stimulate savings and investment (Swank 2006). In the post-war era, most countries combined relatively high marginal statutory tax rates with tax instruments that targeted certain behaviour and investments. Since the 1980s, taxes on top incomes and capital (on mobile, high-income earners) have been curbed and replaced with taxes on less mobile factors and activities including labour and consumption (Swank 2016). Reagan's tax cuts were the first which radically reduced tax rates and functioned as a role model in the OECD. Since the mid-1980s, the average top statutory rate fell from 70 to well below 50 percent and the headline tax rates on corporate income fell from 50 to 30 percent (Ganghof 2006b, 1).
Even though U.S. reforms initially mirrored this overall trend of reducing tax for corporations and top incomes, the composition of reforms changed over time. Reagan's first reform, the Economic Recovery Tax Act (ERTA) of 1981, contained tax cuts for individualsan across-the-board income tax cut which brought down top personal income tax rates from 70 to 50 percent (Martin 1991, 121)but also extended benefits for capitalintensive businesses through an accelerated cost recovery system (ACRS). These "big business" benefits had to be repealed in Reagan's second and third reform which then mostly focused on income tax cuts (Auerbach and Slemrod 1997, 598). Bush Jr further reduced the top statutory tax rate from 39.6 to 33 percent (Lewandoski 2008, 4-5).
Moreover, since Reagan's second tax cut, reforms increasingly incentivised private forms of social security and education finance for medium and lower incomes through deductions on self-employed retirement plans and IRA increases (Martin 1991, 135,159). Bush's 2001 reform also entailed significant deductions on education expenses and improved retirement savings (Graetz and Shapiro 2005, 202).
Several studies have shown that these reforms resulted in significant increases in income inequality (Piketty and Saez 2007). The Bush tax cuts of 2001 and 2003 had an estimated loss of revenue of $4.6 trillion between 2001 and 2013 and included a massive transfer of income from the lower and the middle classes to the rich. The total federal tax burden for the richest 1 percent of taxpayers declined by 25 percent and by 21 percent for the next richest 4 percent. The bottom 95 percent of the income distribution only received a tax reduction of 10 percent (Bartels 2008, 163).

Change in the existing tax literature
Institutionalists have so far stressed the role of institutional context and how it constrains radical change in tax systems. For instance, authors argue that conflicts inscribed into U.S. institutions in the nineteenth century have steered the evolution of tax. In contrast to countries with industrial-relations institutions which ensure consensus over taxation among capital and labour (Ganghof 2006a;Wilensky 2002), the lack of such institutions in the U.S. rendered interaction over taxation strongly adversarial (Morgan and Prasad 2009). While providing important insights into the historical development of tax, this approach cannot explain the more recent and radical changes in taxation.
In part to resolve this issue, Prasad (2006Prasad ( , 2012 examined how policy actors actively shape cross-class coalitions in favour of tax reforms. In her analysis, Reagan and Bush capitalised on electoral discontent which resulted from U.S. adversarial politics, radical partisan positions, and frequent shifts between the politics of the Left and Right (Prasad 2006;Prasad and Deng 2009). This is an important contribution which shows how the interaction of policy makers with societal groups can help overcome institutional hurdles. However, Prasad does not develop a deeper understanding of the role of policy maker communication with different stakeholders in society. This seems important considering how much the tax reforms contributed to rising inequality.
Processes of persuasion of voters, minority congressional party and interest groups are also rather neglected in the voter-attitudes literature. Here, scholars argue that citizens are most likely to request tax progressivity when they feel that other societal groups have been privileged. This is why top incomes were taxed at the highest rates during war times (Scheve and Stasavage 2016) and why demand for redistribution increased in the Financial Crisis in 2008 (Limberg 2019). But unless we understand how policy actors incorporate voter interests into their larger communication strategies and how voter interests matter for policy change, the causal link between opinions and policies remains elusive.
Only the "policy drift" literature explicitly engages with policy maker strategies to surpass institutional hurdles through communication. However, here scholars are mostly interested in incremental forms of change. For instance, change agents may expand tax exemptions instead of radical tax amendments because those do not have to go through the legislative process (Howard 1997;Patashnik 2003), or design reforms in a way that voters do not notice the inequality-raising elements Pierson 2005, 2010). While this is an important step towards a better understanding of strategies to surpass gridlock, persuasion might be the missing link which explains radical reform.
Learning strategic compromise: discursive institutionalism, policy learning and compromise This article follows recent public policy research which uses DI to show the contribution of discourse to policy reform and stability (e.g. Bailey et al. 2021;Crespy and Schmidt 2014;Fitch-Roy, Fairbrass, and Benson 2020;James, Kassim, and Warren 2021). Within this research, it has been shown that compromisedefined as strategic acts of conflict-resolution which require some level of commitment to the other side (Bellamy 2012, 443-444)can play a critical role in institutional change and stability (Bailey et al. 2021). While building on this research, this article attempts to further explore the nature of compromises, how they contribute to policy change and stability and how actors learn to use them more effectively over time.
Discursive institutionalism and the role of compromise DI, the newest of the four "new institutionalisms", attempts to capture the role of preferences, values and strategies of actors and how ideational constructs, stories and narratives enable and constrain policy change (Béland and Cox 2016;James, Kassim, and Warren 2021;Schmidt 2002Schmidt , 2008Widmaier 2016). Different from other ideational approaches, the focus is less on the substantive content of ideas and rather on the interactive processes through which ideas evolve (Carstensen and Schmidt 2016;Schmidt 2008Schmidt , 2010. Through this procedural lens, DI is particularly suited to explain how actors interact in the discursive realm, achieving change or continuity in public policy (McCann 2014).
A critical contribution of DI is that it treats institutions as both "given" and "contingent" (Fitch-Roy, Fairbrass, and Benson 2020, 85). Institutions provide the context within which actors think, speak and act, but also result from thinking, speaking, and acting. Agency takes place in "background ideational abilities", which allow for action within the constraints of the institutional context, and "foreground ideational abilities", which allow actors to think critically beyond the institutional context and communicate with others to change or maintain institutions (Schmidt 2008(Schmidt , 314, 2017. Thus, strategic discourse often takes place simultaneously in actors' ability to work within the surrounding constraintstraditional institutional and ideational boundariesand pushing for change (Fitch-Roy, Fairbrass, and Benson 2020; Crespy and Schmidt 2014).
DI develops three types of ideational power. In power over ideasclose to coercive, structural, and institutional sources of powersactors impose a meaning onto ideas, shame others or resist accepting alternative ideas (Carstensen and Schmidt 2016, 326-328;Tsingou 2015). In power through ideas, actors persuade others of the cognitive validity and/or the normative value of an argument. Finally, power in ideas is interlinked with background ideas and the institutional and structural processes which determine which ideas find support because they suit the social, political or intellectual landscape (Carstensen and Schmidt 2016, 329-332). Since acts of compromise entail an inherent dependence of change advocates on the support of others, it is expected that power through ideas (a tactic of persuasion) plays a critical role. However, since reforms change institutions and discussion about taxation, acts of compromise likely also alter the power in ideas in future debates.
DI distinguishes different types of discourse which are more prevalent in certain institutional contexts: Coordinative discourse is an act of deliberation of the administration with interest groups, coalition partners or minority congressional parties in the policy processwhich often takes place behind closed doors (Schmidt 2010, 110;Widmaier 2016, 342). Communicative discourse justifies policies to the wider public including political leaders, the informed public, and policy forums (Schmidt 2008, 310). While communicative discourse is more likely in "simple polities" with powerful single-party governments, coordinative discourse is more prevalent in "compound polities" in which power is fragmented. With high fragmentation, the US polity constitutes a "compound polity", but the polarised two-party system obstructs agreement which typical for compound polities. Therefore, for US executives, communicative discourse is important to balance out the lack of an effective coordinative discourse (Schmidt 2009, 527, 534).
Finally, DI differentiates cognitive from normative ideas which correspond to different levels of discourse. Policy ideas are specific ideas, focused on policy, programmatic ideas underlie policy ideas and develop problems and solutions, and philosophical ideas constitute the "deep core" which undergird other levels and are rarely contested (Fitch-Roy, Fairbrass, and Benson 2020, 85;Schmidt 2008, 306). While cognitive ideas are most relevant for programmatic and policy ideas and are most persuasive if they accurately define problems and the applicability of a programme, normative ideas are relevant at all ideational levels and most persuasive if they correspond to traditional values held in the community. For successful policy proposals the two types of ideas must be consistent (Carstensen and Schmidt 2016, 324;Seabrooke and Wigan 2016).

Policy learning and compromise
While DI views learning and discourse as intertwined: "Discourse […] is one among several factors involved in policy change" and "can be seen as contributing to 'policy learning'" (Schmidt and Radaelli 2004, 189), the specificities of this relationship have not been teased out. This section incorporates different types of learning to see how they map onto different kinds of policy processes (and the corresponding discourses) and argues that aside from discourse contributing to learning, actors also learn to use discourse strategicallyaside from "speaking to learn", actors also "learn to speak strategically".
The literature on policy learningdefined as the updating of beliefs about policy through analysis, experience and social interaction (Hall 1993;Heclo 1974)has produced a plethora of different typologies (e.g. Dolowitz 2009;Checkel 2001;Dunlop and Radaelli 2013;May 1992). I follow Dunlop and Radaelli (2013) because they introduce strategic learning as a particular type of learning, explore the conditions under which it is effective and how it relates to other types of learning (see for applications Di Giulio and Vecchi 2019; Dunlop 2017; Dunlop and Radaelli 2018;Dunlop, James, and Radaelli 2020). The typology is structured around two variables: First, uncertainty and problem tractability determine whether learning is thick (changes beliefs) or thin (does not change beliefs) and whether learning can be achieved through technocratic or political processes. The second is actor certification or authority and describes whether actors hold expertise or institutional legitimacy.
Based on this distinction, four modes of learning are constructed: In "reflexive learning" radical uncertainty and low actor certification are present and contribute to an open dialogue of actors. In "epistemic learning" some actors are highly socially certified (experts) and are brought into the process due to radical uncertainty. "Learning through hierarchies" is characterised by low levels of uncertainty and high actor certification. Here, hierarchies in institutions shape the production of knowledge. Finally, "learning through bargaining" takes place under low actor certification and low uncertainty and involves interdependent actors exchanging information and bargaining over policy to achieve objectives different from the policy content. Since it can contribute to an agreement on mutually beneficial policy solutions but must not generate genuine agreement on an issue (Dunlop and Radaelli 2013, 607-613), it gets closest to strategic compromise.
How do these types of learning map onto different stages of policy processes? First, reflexive, and epistemic learning are "deeper" kinds of learning through which actors tap into novel sources of knowledge and apply them to a policy problem. This will likely be most relevant in the earlier stages of policy discussions when actors are confronted with a problem or a crisis and seek new information to resolve it and "speak to learn". Learning through bargaining and hierarchies, on the other hand, should arise later when first experiments with new ideas have been conducted and uncertainty is diminished. Learning through bargaining should generate procedural knowledge and help actors to "learn to speak". I argue that, while not solving policy problems, bargaining is of critical importance to solve political problems in the policy process and shapes policy development in its own way.
In the bargaining stage, actors can learn to strategically select different combinations of compromise to develop a vote-winning strategy. In incorporation agents include the other side's cognitive ideas about material benefits and the normative values they hold dear. Compensations appeal to losers (businesses or voter groups) by highlighting the material benefits included in the reform through cognitive ideas. And reconciliations offer certain (cognitive or normative) amendments of the reform (to downsize it, for instance) to enhance acceptance.

Methodology
To capture the impact of policy makers" compromise strategies, the paper uses a two-step empirical analysis: First, a content analysis tests whether there was growing agreement on tax cut elements over time. Because arguments alone cannot prove agreement on a bill, the content analysis is complemented with data on vote margins for every reform. In a second step, archival material unveils the compromise strategies deployed by administrations.
The content analysis portrays frequency of arguments used in plenary debates in Reagan's 1981 ERTA and in Bush Jr's 2001 EGTRRAthe two largest income tax cuts in U.S. history. The coding followed a hermeneutic-content analysis (Schmidt and Radaelli 2004). One coding unit equals one argument which may span several sentences. I selected the codes through an inductive test-round for 50 pages of each of the two reformsthose codes which comprised more than 10 percent of the entire population of codes were included in the codebook (see Appendix A for sample statements and a description of the coding process). Then, I analysed the main debates preceding the roll call votes in the House, Senate, and committee meetings (Ways and Means and Finance Committee) (see for further specifications Appendix A). In total, the text coded comprised approximately 600 pages and it yielded 719 coding units.
The codes were split into normative and cognitive arguments. In the normative realm, the concept of redistribution argues that burdening higher incomes with higher tax is fair (Scheve and Stasavage 2016, 5, 26). Horizontal tax justice argues that all incomes, irrespective of the source, should be taxed equally and that rates should tax incomes consistently (Murphy and Nagel 2002, 37). Compensations correct for other injustices in legislation (Scheve and Stasavage 2016, 5). Maintaining social security is closely tied to arguments of redistribution but does not explicate the distributive element.
In the cognitive realm, business support arguments claim that there is a need to support small businesses or industries. Economy contains statements about incentives to work and save which state that reduced top rates spur greater savings and labour (Musgrave 1959, 246). Economic growth states that tax cuts generate economic development. Different from the normative argument to maintain social security the balanced budget argument is an economic argument: higher deficits raise inflation and starve private investment.
To capture compromise strategies used by the administration, 5,932 and 1,655 administrative documents were collected from the Reagan and George W. Bush Libraries. The archival material covers all the documents that were released in the presidential libraries on the tax reforms and include documents which show the strategic planning of the administration and internal correspondence about policies, documents which demonstrate the ideas and demands carried to the administration from outsideincluding letters from interest groups, speeches, newspaper articles, and opinion polls to cover voter sentiment at each point in time (sources used in Appendix B). Table 1 contrasts the frequencies of arguments made for and against tax cuts in the U.S. Congress by Republicans andDemocrats in 1981 and. The table shows that arguments of incentives for savings and investments and for business support declined. Democratic arguments of redistribution were less frequent (fell from 30.25 to 17.73 percent), and both parties increasingly emphasized the concept of compensationfrom 1.8 (0.0) to 16.58 (14.28) percent among Republicans (Democrats). 1 Moreover, both parties used arguments of balanced budgets with higher frequency (from below 20 to above 30 percent). This indicates that the two parties increasingly shared arguments of balanced budgets and compensationwhich became critical for the tax cuts.

Conflict and reform instability: Reagan I
Reagan's income tax cuts constituted a critical break in US conservativism, not least because of the significant effect on the deficit. In the post-war era, Republicans had accepted rising marginal tax rates and welfare state expansion (Béland and Waddan 2015, 179), but positioned themselves as deficit hawks concerned about inflation and rising interest rates (Prasad 2018, 70-72). These conservative sentiments were also reflected in public opinion. Between the 1940s and 1970s, polls consistently suggested that voters supported deficit reduction over tax cuts (Erskine 1964, 161). In the immediate postwar era, balancing the budget was rarely an issue due to continuous economic growth, however, since the late 1960s, stagnation tossed policy makers into uncertain terrain of rising deficits. This combination of American "deep core" values of balanced budgets and rising deficits posed a critical obstacle for the stability of Reagan's tax cuts.
As described in Dunlop and Radaelli (2013), under conditions of radical uncertainty, successful epistemic learning can come about when experts are present. Since Reagan listened to a range of semi-experts in his first term, epistemic learning took place but was weak. A group of conservatives around Wall Street Journal editorial writer Jude Wanniski successfully framed the economic downturn in the 1970s as a consequence of high tax rates which had undermined incentive structures. While the discussion was kicked off by a group of economists, including Robert Mundell and Milton Friedman, critical ideasincluding the Laffer curve, starving-the-beast and trickle-down economicswere mostly developed by Wanniski and Jack Kemp, a Representative from New York (Block 2009;Morgan 2009, 103;Prasad 2018, 25-27). The Laffer curve became one of the most important frames later used for Reagan's reforms. It promisedin Wanniski's and Kemp's understandingthat tax cuts would not only not raise deficits, but help generate tax revenue (Béland and Waddan 2015, 180). 2 Even though it diametrically contradicted the Laffer-curve, conservatives simultaneously pushed the starving-the-beast narrativethe idea that cutting taxes would lead to lower government spending (Blyth 2002, Chapter 6). In sum, since epistemic learning was not opened up to the general academic field, but experts were often selected in a way that outcomes suited conservative political interests, some learning took place, but it was limited. When Reagan proposed his first tax cut, ERTA in 1981, he started the discursive process from a place of maximum reform ambition aiming at radically curbed tax rates and a retreat of the state from the private sphere. Instead of looking for Democrats' support, the administration directed a strong communicative discourse at Reagan's conservative constituency emphasising "trickle-down economics" and horizontal tax justice. Trickle-down economics argued that tax cuts for the rich created jobs for the poor because it incentivized the rich to save and invest more (Béland and Waddan 2015, 177). Horizontal tax justice argued that high levels of inflation, induced by government spending, had eaten up income increases of average Americans. This process of "bracketcreep" placed an uneven burden on lower and medium incomes: "All of you who are working know that even with cost-of-living pay raises, you can't keep up with inflation. […]. Over the past decade we've talked of curtailing government spending so we can then lower the tax burden" (Reagan 1981). The tax cut relieved individuals and reinstated the accuracy of the income tax schedule. Horizontal tax justice also featured strongly in Congressional speeches (see Table 1). It was argued for instance, that indexing tax brackets to price development would ensure that hard-working Americans paid stable rates, it "will prevent unlegislated tax increases in the future" (House of Representatives 1981, 17904).
The problem with this strong emphasis on horizontal tax justice in the communicative discourse was that it did correspond to the cognitive ideas debated in the coordinative realm. While income tax cuts (across-the board with a top rate reduction from 70 to 28 percent) were central to the reform, big-business tax cuts made up a significant (20 percent) share of the bill and included the implementation of the accelerated cost recovery system (ACRS) to help manufacturers invest in buildings and equipment (Martin 1991, 121). The Office of Public Liaison (OPL), responsible for organising interestgroup relations, concentrated on forging a large business alliance. Since close relations had already been forged with the National Association of Manufacturers (NAM) (Martin 1991NAM 1981, the OPL advised bringing in small businesses through additional estate and gift tax cuts and to offer capital gains and interest tax cuts to the American Banking Association (ABA) (Dole 1981a). While this coordinative discourse was successful at the timebusinesses initiated a telephone and ad-campaign which won 39 Congressional votes for ERTA (Small 1981a), the wedge between the outward normative communication and the increasingly business-oriented tax cuts limited lasting support.
The Democrats, on the other hand, developed a much broader appeal for their alternative programme by narrating a tax cut that was much closer aligned with traditional American values. Together with the trade union association AFL-CIO the Chairman of the Ways and Means Committee, Dan Rostenkowski, developed the Ways and Means Proposal. It was smaller (rates were reduced by 15 instead of 30 percent), and therefore less of a threat to a balanced budget, and entailed benefits for the middleclass. This approach allowed Democratic Congressmen to consistently present a strong communicative discourse based on redistribution (30.25 percent of arguments; Table  1) and balancing the budget (18.52%). In the normative realm, Democrats argued that Reagan's tax cut led to income inequality as "a taxpayer earning $200,000 would receive a tax reduction of $22,000 in 1982, while a taxpayer earning $15,000 would receive a deduction of $350", additionally interest and dividend tax cuts were "helpful only to the high-income people" (House of Representatives 1981, 18054). I believe a major tax reduction proposal must undergo careful scrutiny for fairness […]. I favor a reduction which will focus on the middle-income wage earners. Any reduction that shortchanges the working men and women of this Nation is unfair. The President's original plan would have given 30 percent of the benefits of the tax cuts to a group of only 4 percent of the taxpayers […], the committee's more equitable proposal gives 64 percent of the income tax relief to 54 percent of the taxpayers' (House of Representatives 1981, 18059).
Basing the plan on these normative frames of justice and equality facilitated the formulation of a strong communicative discourse among the Democrats.
However, formulating normative arguments in line with the basic party ideology was not enough to win a contest over an alternative tax bill. Thus, the Democrats knitted the justice frames developed in their communicative discourse together with strong cognitive arguments of generating jobs in small businesses in the coordinative discourse. Bigbusiness benefits and especially the tax credits for capital-intensive businesses were rejected as containing a risk of incentivising labour shedding. These measures gave "unwanted incentives" to multinational firms to engage in employment reducing "mergers, takeovers, plant-shutdowns, [and] overseas investments" (AFL-CIO 1981). It was more sensible to give tax incentives to the emerging small-firm high-tech sector with better job prospects (Valis 1981). Aligning the coordinative and communicative discourse through the concept of job creation in the small-business high-tech sector increased Republican support for Rostenkowski's tax plan. Republican Congressmen supported the idea that raising tax support for small businesses was more important than giving tax breaks to big manufacturers (mentioned 11.98 percent).
After Reagan lost votes to the Democratic programme, the administration expanded business elements which eventually bought in sufficient votes of Southern Democrats (Martin 1991). While the bill was passed with 48 (predominantly Southern) Democratic votes (by a margin of 238-195) (Weaver 1988, 203), its stability was compromised. In his second and third tax cut, Reagan was forced to reverse several elements of this reform (Martin 1991, 121).

Learning Strategic Compromise: Reagan II
The chances of passing a stable tax reform increase with the use of strategic acts of compromise which acknowledge and respond to the other side's normative and cognitive ideas. Although TRA of 1982 and TEFRA of 1986 further reduced the top income tax rate from 50 to 28 percent, the bills passed with strong Democratic support (Martin 1991, 159).
Initially, the two tax reforms had slim chances of passing Congress because voters were increasingly opposed to tax reforms. In 1981, Republicans had justified tax cuts in the communicative realm by emphasising the mostly cognitive ideas that reforms would be able to incentivise savings and work (22.30 percent of statements) and achieve a balanced budget (12.84 percent). In both discursive categories, the Laffercurve and starving-the-beast arguments condemned federal government deficits and promised a smaller state through tax cuts. In Congress, representatives repeated the starving the beast narrative. Tax cuts forced the government to "act explicitly to raise taxes" and hindered the federal government to indulge in the "luxury of an inflation-induced tax windfall to finance unnecessary, wasteful but politically-expedient Government support programs." This communicative discourse of "starving the beast" made the administration highly dependent on an actual reduction in sovereign debt. Together with the Volcker Shock, the tax cut of 1981 contributed to skyrocketing levels of debt -$200 billion in 1984and public support for Reagan's tax plans plummeted (Stockman 1994, 232). Polls indicated that 77 percent of Democrats and 70 percent of Republicans favored a balanced budget over further tax cuts (Meese 1982). If change agents wanted to implement additional tax cuts, they could no longer rely on communicating economic benefits of tax cuts to voters.
As described in Dunlop and Radaelli (2013), successful learning through bargaining can come about at low levels of uncertainty and low actor certification. This was clearly the case in Reagan's second and third tax cuts. At this stage of the reform process, actors had experienced the immediate economic effects of tax cuts and external advisers receded into the background of learning. Instead, the administration focused on solidifying the support for tax reforms through acts of compromise. While reaching out to business groups in a coordinative discourse again, this time, the OPL specifically incorporated the small-business idea developed by the Democrats and reached out to small businesses and the banking industry while proposing to repeal big business benefits such as the ACRS. The bill included a corporate tax reduction for small businesses, savers benefits and IRA expansions which benefitted finance (Shanahan 1986(Shanahan , 1147. While the reforms were still highly skewed to the top, supporting small businesses and savers was popular. The Democratic leadership supported the small business tax cuts and RNC polls showed that 81 percent of voters supported extensions of IRAs (RNC 1985, 4).
Next, Republicans offered reconciliations to the Democrats by acknowledging that big business benefits had generated inequalities (in the normative realm) and tax evasion (in the cognitive realm) and proposed the closing of loopholes, abolishing the ACRS, and implementing a 20 percent corporate alternative minimum tax (Martin 1991). Treasury Secretary James Baker communicated these measures with a framing of "nostra culpa" to the public. TRA had to be "far less offensive" and shift tax burdens "from individuals to corporations" because public frustration had "translated into a loss of respect for the government" (cited in Seaberry and Swardson, 1985). Reagan's speeches mirrored this framing: "Americans are dissatisfied with the current tax system" mostly because it is "unfair" (Reagan 1985, 1). Finally, TRA and TEFRA also tied cognitive frames about material benefits for lower incomes with normative rhetoric in a strategy of compensation. Even though the reform was still strongly skewed to the top, the administration added several measures to counter Democratic arguments that the tax reform was unfair. A standard allowance was increased from $3,670 to $5,000, the bottom rate was raised from 11 to 15 percent and the Earned Income Tax Credit (EITC) rate was increased (Auerbach and Slemrod 1997, 595). These measures allowed for an argument that the lower incomes were compensated for the losses they incurred through the tax plan.
The consequence of incorporation, reconciliation and compensation was that the Democrats and the AFL-CIO endorsed Reagan's second and third tax cut because "loophole closing" corrected for "some of the excesses of the 1981 corporate tax cuts" (Kirkland 1985, 2). TEFRA passed the House supported by 123 Democrats of which 96 had voted against ERTA (Martin 1991, 153). TRA passed Congress with a voice vote in the House and 44 of 47 Democrats supported the bill in the Senate (Rosenbaum 1986).
Since Reagan's policies were part of an international supply side movement also prevalent in the UK, Australia, New Zealand and Latin America, there is a possibility, that political strategies were learned from experiences made abroad. The most likely source of diffusion would be the UK, as Reagan admired Thatcher's sweeping reforms (Rossinow 2015). However, diffusion of ideas requires policy makers to be aware of and interested in the acts of their counterparts (Dolowitz 1997). The archival material presented in this section suggests that, due the confrontation with the Democrats, the discussions in the Reagan administration were firmly focused on what the Democrats were asking, and how they could appease them, and therefore functioned more locally than internationally.

Winning Strategies and Reform Stability: Bush Jr
Radical change was followed by a long period of relative continuity in conservative tax policy: Bush Jr did not only adopt many of Reagan's cognitive ideas about tax instruments, but his 2001 reform also used similar normative frames to legitimate them. As Reagan's later tax cuts, EGTRRA focused on individual incomes (an across-the-board income tax cut with top rate reductions to 33 percent (Lewandoski 2008, 4-5)) and financial incomes (Lewandoski 2008, 4-5). And it also included deductions on education expenses and improved retirement savings (Graetz and Shapiro 2005, 202). These similarities with Reagan's second and third reform are surprising as they were implemented under vastly different economic conditions. Reagan was confronted with a crisis of inflation, sovereign deficits and declines in manufacturing productivity. Bush Jr, on the other hand, was elected after a long period of economic expansion and sovereign surpluses (Béland 2006).
Previously tested strategies of compromise can serve as templates for the effective integration of cognitive ideas about material benefits (agreed upon in a coordinative discourse) and normative framing (developed in a communicative discourse). This process of discursive learning becomes apparent in Bush's strategy to win votes for his 2001 tax plan. During the election campaign voters had voiced preferences for tax cuts and increases in welfare spending (Bartels 2005;McCall 2013). The archival material clearly shows that the administration analysed Gallup polls which indicated that national policy priorities were child rearing (41 percent) and reducing the cost of health care (36 percent). Cutting federal taxes was only a minor priority (26 percent). 69 percent of voters supported an equitable tax cut with the largest share focused on the middleclass while protecting spending programs like Medicare (Gallup 2000, 77;Morgan 2009, 222). Taking these voter preferences into consideration in his communicative discourse, Bush developed an agreeable approach of presenting his tax cuts within the larger frame of "compassionate conservatism" in which low-income groups were not left behind.
Compassionate conservatism became a critical narrative in the 2000 presidential campaign which incorporated values of classical liberalism (including the critique of state power) and traditionalism (the emphasis on traditional social and religious values). Different from Reagan I, Bush developed a narrative based on a coherent set of values including support for the poorfor instance through tax incentives to work and safe, while focusing on personal responsibility. This was tied in with the narrative of the ownership society which suggested that private ownership was superior to state ownership. These two concepts allowed the Bush administration to develop an alternative to Big Government and generate a positive narrative for a conservative programme (Béland and Waddan 2007, 772;Waddan 2010, 174).
Like Reagan in his second and third tax cut, Bush Jr carefully orchestrated the coordinative discourse by ensuring that the normative demands among voters and the Democrats could be incorporated. Forging a coalition of small businesses, small farmers, finance, and savers, allowed Bush to stress that this was a reform for the average American and not big business. Tax instruments requested by manufacturing associations were not considered (NAM 2001). Instead, the tax plans proposed measures for small-and medium-sized businesses including top rate reductions and estate tax cuts (Chamber of Commerce, 2001;Graetz and Shapiro 2005). 3 After the Enron scandal, big business tax cuts would have been highly risky (McCall 2013), but two-thirds of Americans supported estate tax cuts because they were understood to help individuals and small businesses. Next, Bush added dividend tax cuts, expansions in IRA tax allowances and education-savings investments to the bill (U.S. Congress, 2001, 104). These measures were not only presented as supporting the finance sector (in the cognitive realm) but also as compensating hardworking lower and middle-incomes (in the normative realm). Seniors "receive [d] half of all dividend income in America" (Bush, 2003, 2). Tax cuts for the financial sector allowed for the average American to fund "education, retirement investment" (Press Office, 2002). The measures did not only revive the New Economy but made "a significant dent in a family's mounting credit card debt" (McGrath, 2001, 1-2).
In Congress, Republicans used this coordinative discourse to find a coherent communicative discourse directed at voters. Table 1 shows that arguments of incentives to work and save were much less frequent in 2001 compared to 1981 (fell from 25.15 to 5.88 percent). The main reason was that the top rate reductions were now increasingly justified through compensations for lower and medium incomes. The critical narrative developed by the Republicans was that they had included many minor tax measures which ensured that hard-working Americans benefitted. Congressmen argued that the instruments of "education savings incentives, charitable giving, and the permanent extension of the R&D credit" could "expand savings and investment opportunities" and were "a real and lasting relief to the American people" (U.S. Congress, 2001, 5). The fact that this resonated with Democratic Senators, who used the rhetoric to push for further compensations, demonstrates its effectiveness. Senators considered to support the Bush tax cuts if further tax credits for payroll taxes were included in the bill: "I would like you all to consider the concept of doing something on the payroll tax by giving a credit against income taxes paid to take care of that bottom group" (Committee on Finance, 2001, 14).
Finally, Republicans also learned from their experience of winning Democratic support by ensuring a balanced budgetwhich was offered as a reconciliation to Democrats. The main concern of the Democratic party was that the Bush tax cut would jeopardise the balanced budget (31.03% of all speech acts in 2001). Thus, in the committee debates, Republicans accepted the Democrat's cognitive ideas to reduce the volume of the tax cut from $1.6 to $1.2 trillion. The tax cut passed the Senate by 58:33 margins with 12 Democratic votes (Rosenbaum 2001).

Conclusion
This paper finds that strategic acts of compromise can be critical to enhance reform stability, while shaping policy outcomes. Reagan's first attempt to implement radical tax cuts partially failed because the cognitive ideas (big-business benefits) used to bring in business support in the coordinative discourse did not match the communicative (normative) discourse presented to voters. In the two-follow-up reforms, big-business tax cuts had to be repealed but a new approach of compromise allowed for the passing of further income tax cuts. The acts of incorporation, compensation, and reconciliation not only facilitated the formation of a consistent coordinative and communicative discourse, garnering higher levels of support of the minority congressional party, but also considerably changed the composition of the reform.
The second finding it that change advocates learned how to use strategic acts of compromise and thereby stabilised reform initiatives over time. In the first stage of income tax reforms, the Reagan administration predominantly learned epistemically, however, learning was limited and was not sufficiently oriented towards strategic goals of getting the bill through Congress. In Reagan's second and third tax cut, the administration learned which types of acts of compromise increased the number of Democratic votes in favor of tax cuts. This type of "strategic learning" or "learning through bargaining" was thus a critical ingredient which stabilised the reforms. Bellamy (2012) argues that compromise enhances the democratic fiber of a reform if it genuinely incorporates requests of the other side. However, this study shows that incorporating ideas of the Democrats enabled the passing of highly unequal tax cutsnot in the interest of Democratic representatives or voters. While Republicans offered tax instruments which compensated losers, these measures did not change the overall distribution of tax cuts. If tested against the concept of input legitimacy, defined by Fritz Scharpf (1970) as policy makers' responsiveness to citizen demands, legitimacy did not increase either. The administrations used discourse to pass reforms which increased inequality against the voters' will. Thus, strategic acts of compromise can instill an impression of policy makers conforming with the voter will, while disguising the actual content of a reform.
Several practical implications arise from this analysis. If policy advocates want to realise stable reforms, they should make themselves aware of the institutional (and ideational) context within which they operate and where powerful opposition may originate. Do they operate in a simple polity, in which communicative discourse is dominant, in a compound polity in which coordinative discourses is dominant, or in a combined system, as the US polity, where both discourses are required? Next, actors have to identify which traditional sets of values should not be violated and should listen closely to what powerful critics demand. Then, they can incorporate critical cognitive and normative ideas held dear by voters and the minority congressional party and make them "fit" their own strategy. The result must be a discourse which consistently aligns cognitive (material elements of reform) and normative elements (value-based justifications).
The current discussions to raise taxes for higher incomes and the wealthy to finance COVID-related spending are often countered by the argument that businesses will lose out from more state intervention. Policy advocates could incorporate demands for better investment opportunities in the US economy and offer new business support programmes as a compensation allowing for a consistent normative and cognitive narrative of state-led economic revival. Notes 1. However, the decline in Democrats' arguments of redistribution were balanced out by an increase in the argument of maintaining social security in 2001. This indicates that social justice as a whole remained an important concept for the Party. 2. The Laffer-curve showed that an intermediate level of taxation could raise tax revenue and implied for some supply side economists that tax cuts not necessarily lowered the level of tax revenue. 3. These measures brought the two largest small-business associations, the NFIB and the Chamber of Commerce, into Bush's Tax Relief Coalition.
be directed at most Americans, not at the small percentage of wealthy families, and must be designed to increase incentives to work and save" (House of Representatives 1981, 17906). Here both "horizontal tax justice" and "incentives for saving and work" were given an equal weight in the argument. An example where one coding was given despite two arguments is this one: "I do not object to spending more money on education. The Democratic budget provided for more money for education. But I do object to us passing legislation that is going to add to red ink. That is where we are heading, to larger tax cuts, larger spending, and what we will give is our ability to pay down our national debt" (House of Representatives 2001Representatives , 2207. Although the speaker talks about spending on education, this is just presented to contrast their argument from the administration's core argument which is about the budget.

A2examples for each coding category
Normative Arguments -Redistribution: "Next, the vast bulk of every other American, the average American, they only get a grand total of 16 percent of the total tax cut, but he says it should go directly back into the pockets of big oil and gas and electricity companies across the country to pay for people's energy bills. So no tax cut in people's pockets" (House of Representatives, 16.05.2001Representatives, 16.05. , 2205. -Horizontal tax justice:"So I think a lot of people, when they read that we have a surplus at the Federal level, that they feel that this would be one way to target tax relief to working families, especially those where both spouses work to sustain the family economically. So I am very pleased and honored to be able to join Congressman Weller and the other Democrats and Republicans on this legislation towhat we hope to achieve is to have a fairer Federal tax code. Yes, it will provide tax relief, but there is simply no rationale why a couple should be punished for being married if a couple that lives together without the benefit of marriage is not susceptible to that same tax burden" (Ways and Means debate, 21.03.2001, 22).
-Maintain social security: "But what troubles me about the present budget chairman and what is going on the House floor today is if we should have been embarrassed for Congressmen writing checks on money that was not there, should we not be ashamed that we are passing tax cuts on a day when we owe the Social Security system $1.1 trillion? We have taken their money, we have spent it on other things and now when we have a small surplus, instead of putting that money aside for Social Security, we are giving some Americans a tax break" (House of Representatives debate, 16.05.2001, 2206).
-Compensation: "I believe increasing the standard deduction for a married couple filing jointly so that it is twice that of a single person and widening the 15 percent tax bracket so that we can at least alleviate the pain at that lowest level and it will give some relief to every couple that either does not itemize and takes the standard deduction or anyone who is paying taxes would get some relief from the 15 percent bracket." (Ways and Means debate 21.03.2001, 3).

Cognitive arguments
Economy. -Balanced budget: "I had occasion to say when we were discussion the budget that it was a "drop dead America" budget. This tax bill drives the nail into the coffin of America because when we have the kind of rate of increase in defense expenditures, three times the rate of Vietnam, $1,635 trillion over the course of the next 5 years, and instead of increasing taxes to take care of that kind of expenditure, we cut taxes, we are asking for chaos in the American economy" (House of Representatives, 16.05.2001Representatives, 16.05. , 2206. -Economic growth: "American jobs are exported overseas. That is what we are talking about today; we need to encourage people to invest and be productive" (House of Representatives, 29.07.1981, 18068).
-Savings and investment: "Cutting marginal tax rates encourages individuals to work harder and to take risks. For the small businesses who pay taxes on the individual schedule, these tax cuts will make it possible for them to expend the capital necessary for them to continue to grow." (House of Representatives, 16.05.2001Representatives, 16.05. , 2214