Realism and Responsible Parties

Realism can mean many things in political theory. This article focuses on “common-sense realism,” an approach to decision making under uncertainty characterized by its posture toward risk. Common-sense realist arguments have become popular in recent democratic theory. One prominent example is found in debates over the responsible party institutional model (RPIM). RPIM’s main features are two-party competition for full control of government and party organizations that empower officeholders, not activists. Proponents of RPIM defend it in realist terms. They claim that efforts to pursue more ambitious democratic ideals jeopardize goods that RPIM can readily secure. In this article I articulate a realist approach to institutional evaluation that assesses proposals on three dimensions: robustness, feasibility, and stability. Using this approach, I demonstrate that the realist argument for RPIM is weaker than it initially appears. The debate over RPIM is not a debate between realism and idealism but between competing democratic ideals.

P olitical parties present a thorny puzzle for democratic theory. On the one hand, parties can provide a way for elites to consolidate power and are commonly viewed, in both scholarly and popular theories of democracy, as creatures of an elite political class. 1 On the other hand, parties perform essential functions in modern democracies, organizing and mobilizing the mass public (Muirhead 2006;Rosenblum 2008;Sartori 2005;Schattschneider 1942;White and Ypi 2016). Parties may be the creatures of political elites, and elites may wield disproportionate power within them. Yet political parties are still essential parts of a democratic system that constrains the exercise of power and enables the citizenry to act collectively to pursue public projects. A crucial question for democratic theory, then, is how we can curtail parties' antidemocratic tendencies without hampering their democratic functions.
One approach seeks to prevent the concentration of power by dispersing it within or across parties. This approach may involve fostering a multiparty system that enables citizens who are dissatisfied with their party to join or form another. It may also involve dividing and distributing governing powers across multiple offices with different selection procedures. Or it may involve promoting less hierarchical forms of party organization.
However, one influential theory of party politicsthe theory of responsible parties-holds that these efforts to "democratize" party politics are counterproductive. The most thorough and sophisticated articulations of this theory can be found in two books: E. E. Schattschneider's classic Party Government (1942) and Frances Rosenbluth and Ian Shapiro's recent Responsible Parties (2018). Though these two accounts diverge in some respects, 2 they nevertheless overlap in several of their institutional prescriptions and in their arguments for them. Proponents of responsible parties claim that the mechanisms by which parties organize democratic politics work best when influence over party platforms, legislative agendas, and candidate selection is concentrated in the hands of a small number of office-holding (or office-seeking) party leaders. The two major statements of the theory of responsible parties also converge on the idea that two parties competing for full control of government creates the strongest incentives for elected officials to promote the public interest. Hereafter, I will refer to this institutional arrangement-two-party competition for unified control of government combined with a hierarchical party organization centered on the party in public office -as the responsible parties institutional model (RPIM).
Proponents of RPIM contrast responsible party systems with responsive party systems (Rosenbluth and Shapiro 2018, 21). Responsive party systems create mechanisms for many citizens and groups to influence party decision making and/or to easily find or form a more congenial party. But proponents of RPIM argue that these mechanisms of responsiveness make it harder for voters to hold representatives accountable for serving the public interest.
The debate over the value of responsiveness takes place on two fronts. First, proponents of RPIM may offer an account of democracy's value and purposes that differs from the account underlying many arguments for responsive party systems. Advocates of RPIM argue that the purpose of democracy is to serve "public" (as opposed to "sectoral") interests, and they identify these public interests with the median voter. Opponents of RPIM may either reject this majoritarian account of democracy altogether or insist that additional procedural values or goals should also play a role in how we evaluate democratic institutions.
The most common argument for responsible parties attempts to sidestep this disagreement about democracy's value, though. Instead, the argument is cast in realist terms. Proponents of RPIM often acknowledge that their institutional prescriptions do not sit well with popular ideals of democracy, so much so that they seem "paradoxical" (Rosenbluth and Shapiro 2018, 2). However, they argue that more democratic forms of party organization are unrealizable in practice (Schattschneider 1942, 59). As Schattschneider puts it, responsible parties "take from the people powers that are merely theoretical" (1942,52). Moreover, proponents of RPIM claim that attempts to realize a theoretically superior alternative end up producing a system that falls even further from democratic ideals than RPIM. Responsible parties, according to this argument, are "as good as we can get in a democracy" (Rosenbluth and Shapiro 2018, 13).
In this paper, I argue that the case for RPIM cannot succeed on realist grounds alone. I begin by articulating a realist framework for institutional evaluation. This framework assesses existing and proposed political institutions on three criteria: robustness, feasibility, and stability. In the second section, I explain and describe the general structure of the realist argument for RPIM. This argument rests on a claim that RPIM can reliably deliver a particular good-policy outcomes that serve long-term public interests-that alternative institutional arrangements jeopardize. In the third section I reconstruct in greater detail the argument for this claim. In the fourth section, I interrogate the assumptions underlying this argument and demonstrate that RPIM is not robust to predictable violations of these assumptions. Then, in the fifth section, I argue that RPIM is also vulnerable to failures of stability and feasibility. The upshot of this discussion is that RPIM does not reliably deliver the kinds of policies its proponents claim. In the sixth section, I consider and respond to a possible argument that, despite its poor prospects for success, RPIM limits catastrophic outcomes. Altogether, I argue, at least given available evidence, we cannot conclude that RPIM dominates models with more mechanisms of responsiveness. The debate between responsible and responsive party advocates, then, should not be viewed as a debate between "realists" and "idealists." Rather, it is a debate between two competing models of institutional reform that target different goods and carry different risks. In the final section, I discuss the contours of that debate and the challenges facing a nonrealist argument for RPIM.
The point of this paper is not to provide a dispositive argument against RPIM or in favor of some alternative model with more mechanisms of responsiveness. Rather, I demonstrate that this debate cannot be settled on the realist front. A successful argument for RPIM will need to be based on the strength of its democratic ideal and a claim that this ideal is sufficiently valuable that it is worth pursuing despite the risks of failure and other potential sources of value forgone in the attempt.

REALISM AND THE CHOICE OF POLITICAL INSTITUTIONS
Realism means many things in political theory. In characterizing the argument for responsible parties as a realist one, I invoke what we might call a commonsense notion of realism as an approach to decision making. This common-sense notion of realism is the one that a concerned parent might call on to dissuade their child from moving to Hollywood to pursue an acting career. If I enjoin someone to "be a realist," this usually means that I want that person to be more sensitive to facts when making decisions and to trim or adjust their ambitions. 3 Common-sense realists forgo opportunities to pursue the most valuable goods and instead seek to secure goods that may be less valuable but more attainable.
Common-sense realism abounds in political theory arguments about institutional choice. These arguments go beyond counseling political theorists to "undertake a kind of due diligence" in considering feasibility constraints and other relevant facts (Sabl and Sagar 2017, 270). Often, the main disagreement between realists and their opponents is not about what the facts are but about what we ought to do in light of them. This is especially true in the circumstances of uncertainty that surround the choice of political institutions. Put in rational choice terms, the realist approach to decision making under uncertainty is, at the very least, risk neutral, but most often it is risk averse or loss averse (Galston 2010, 394). Realists tend to focus their attention on preserving goods that have already been 3 This notion of common-sense realism is distinct from the strand of realist political theory that takes antimoralism as the defining feature of realism. Some contemporary realists argue that the common-sense notion of realism I describe here is better cast as "nonideal theory" (Rossi and Sleat 2014, 690). The label nonideal theory is too broad for the purposes of this paper, though. Although nonideal theories all seek to deploy realistic assumptions, they can vary greatly in their ambition and tolerance for risk (Valentini 2012). Common-sense realism is thus best understood as a position within nonideal theory. Meanwhile, the common-sense realist approach to political decision making under uncertainty co-occurs with antimoralist political theory often enough that some commentators identify both as primary themes in the family of contemporary realist theories (e.g., Galston 2010). Common-sense realism is logically compatible with both moralism and antimoralism (it doesn't assert any claims about the source of normativity we draw on in our evaluations). Likewise, antimoralist realism is logically compatible with both common-sense realism and idealism (Rossi and Sleat 2014, 691), but the two views no doubt share some common foundations, including a healthy appreciation of human fallibility. achieved (Galston 2010, 396), making small, adaptive gains (Shapiro 2016), and avoiding the worst outcomes (Bagg 2018, 896-7;Galston 2010, 394;Shklar 1989) rather than achieving the best.
Common-sense realism prescribes institutional arrangements that have relatively high prospects of success at securing some important (if not maximal) good or that have relatively low downsides should they fail to perform as intended. Realist arguments sometimes explicitly grant that the best possible outcomes of their proposals are theoretically inferior to alternatives. Nevertheless, they may argue that their proposal is preferable because it can more reliably secure intermediate goods. Realist arguments do not always concede the desirability of an alternative proposal's ideal, though. Common-sense realist arguments can be deployed where there is disagreement about the very best political outcomes. Realist arguments need only assume that participants agree on the value of some good. The hallmark of a realist argument, then, is an emphasis on demonstrating the positive claim that some agreed-upon good can be most reliably secured by the realist's proposal, whereas the alternatives place it in unacceptable jeopardy. In the debate over RPIM, the good in question is "good public policy" (Rosenbluth and Shapiro 2018, 4). Proponents of RPIM argue that the arrangements of responsible party government best secure policy outcomes that serve citizens' long-term interests. Institutions that promote responsiveness, on the other hand, jeopardize the good of public-interested policy. This is assumed to be an unacceptable trade-off for any other values that responsive institutions might realize.
Regardless of whether the proponent of a commonsense realist argument thinks that pursuing their preferred path involves settling, for the argument to succeed qua realist argument, it must be convincing to those that do. Insofar as realist arguments sidestep disagreements about which goods are most worth pursuing, their claims about the prospects of proposals to secure intermediate goods demand special scrutiny. Assessing the realist case for an institutional proposal, therefore, requires careful attention to the various sources of risk that might affect its probability of success.
In the remainder of this section, I provide a framework for assessing the prospects of institutional proposals. This framework identifies three qualities of an institutional arrangement that commonly play a role in realist arguments: feasibility, stability, and robustness. 4 Each of these three qualities corresponds to a different source of risk or uncertainty associated with an institutional model. This framework does not represent a unique or comprehensive breakdown of evaluative criteria relevant to the choice of institutional arrangements. Rather, it serves as a useful tool for identifying different types of risk that might arise with any institutional model. Feasibility refers to the prospects for realizing a world in which the prescribed institutions are in place. Realists often focus on feasibility constraints related to power dynamics and conflict (Kirshner 2022, 127). Institutional arrangements are only feasible if they can secure the compliance of veto players. Other theorists have observed that imagination shapes what is feasible; popular beliefs about what is possible are selffulfilling (Wright 2010, 23). Institutional arrangements that are more feasible have better prospects of success, all else equal.
The costs of pursuing infeasible arrangements that spring most immediately to mind are opportunity costs. Attempting to realize an ideal set of institutions takes time, effort, resources, and political capital that could be devoted to implementing some other set of institutions. Realists may argue that we should not waste scarce opportunities and resources for significant political reform in the pursuit of infeasible ideals. Instead we should take more modest, but more certain, victories.
Pursuing infeasible institutional arrangements presents another, often underappreciated, downside risk related to what has been called "the fallacy of approximation" (see Estlund 2019). Sometimes a partial success is worse than a total failure. Democracies are complex systems, and the political values that we care about are realized through the interaction of various institutions and political practices. Likewise, the political problems often arise from dysfunctional interactions of various institutions that are individually defensible. Dysfunctional approximations represent a downside risk of pursuing infeasible institutional packages because institutional reform is typically piecemeal. It requires victories in multiple jurisdictions or political arenas. Pursuing an ambitious program of institutional reform thus often results in something, but not everything, getting done. This is a problem if it's the wrong something.
Even feasible reforms may fail if they are unstable. Stability refers to the likelihood that the prescribed institutional arrangements, once in place, will stay in place. Institutions are stable when no (sufficiently powerful) actor has an incentive to alter or to circumvent the arrangement. Stability is a well-theorized criterion for assessing constitutions, and insights from theories of constitutional stability can also be applied to understand the stability of institutional arrangements (such as intraparty governance) that are not necessarily part of a state's constitution. Institutional arrangements become unstable when those with the power to change or circumvent them think they are better off without them and when deviation is not too costly. The long-term stability of an institutional arrangement depends on its being sufficiently adaptable to continue to make deviation unattractive even as circumstances change (Mittal and Weingast 2013, 280-3).
Like pursuing infeasible arrangements, implementing unstable institutional arrangements comes with opportunity costs. Instead of achieving a fleeting, insecure ideal, we could create a lasting regime to secure more modest political values. Furthermore, when ideal arrangements break down, they could produce outcomes that are worse than the status quo. A realist assessment of an institutional proposal thus requires asking not only how stable it will be but also in what ways it is most likely to degrade and how bad those degradations would be.
Finally, robustness refers to how an institutional arrangement performs across a range of likely circumstances. Even if a set of institutional arrangements is fully realized and stable, it may only produce desirable outcomes under specific, contingent circumstances. All else equal, we should prefer institutional arrangements that realize the values that purportedly justify them over a larger or more probable range of assumptions.
Failures of robustness, like failures of feasibility and stability, present opportunity costs and downside risks. If there are sufficiently numerous or sufficiently likely circumstances in which a set of institutions fails to perform as intended, we may prefer an alternative set of institutions that performs more reliably, even if it is less impressive at its best. And, again, when things go badly, we want to ask how badly.
Insofar as realist arguments counsel modest goals, they carry an elevated burden of empirical proof. The value proposition of a realist proposal is not that its goals are superior to those of the alternative(s) but that their achievement is more likely. Realist arguments, then, must be able to overcome doubts that their proposal really is the safer bet. A complete and successful realist argument requires that the realist proposal compare favorably to the alternatives, taking into account its prospects for success and costs of failure on each of these three dimensions. As I will argue, the realist argument for RPIM does not meet this standard. A successful argument for RPIM, then, will have to be based on the attractiveness, not the certainty, of its value proposition.

THE RESPONSIBILITY VERSUS RESPONSIVENESS TRADE-OFF
Advocates of the responsible parties institutional model contrast it with alternative models that promote responsiveness at the expense of responsibility. Political scientists offer various meanings for the terms "responsiveness" and "responsibility," and there are at least two distinct debates about the desirability of responsible versus responsive party systems that deploy these labels in different ways. One of these debates is about the kind of behavior we want from governing parties. In that debate, "responsive" systems produce policies that align with voters' immediate preferences, whereas "responsible" systems produce policies that best serve citizens' long-term interests (see Bardi, Bartolini, and Treschel 2014).
This paper intervenes in a separate debate about the strategy we should use for inducing desirable behavior on the part of governing parties. The two debates are distinct. RPIM's proponents argue that it promotes government that is both responsive to voter preferences and serves long-term public interests. 5 In the debate about institutional strategies for making sure that governing parties behave well, systems of responsive parties allow many stakeholders to exercise direct influence in party decisions. Because of their party's internal organization or because party supporters can credibly threaten to defect from the party, party leaders in responsive systems are closely tethered to the expressed wishes of many citizens and groups. In contrast, responsible parties are given a long leash. Party leaders have extensive discretion in how they campaign and exercise power once in office. They are also held responsible for what they do with that discretion. According to the theory of responsible parties, rational party leaders under RPIM anticipate the consequences of being held responsible for their actions, and this motivates them to use their power to serve the public interest.
Responsiveness and responsibility strategies both have their limitations. Responsiveness strategies require time-consuming supervision and intervention. Consequently, responsiveness strategies may be more useful for groups that are more motivated, are better organized, or have more resources. Responsibility strategies have their downsides too: where the interests of elected officials differ from those of the public and/or where there is asymmetric information, representatives will not be perfectly well behaved (see Ashworth 2012). Responsibility strategies also offer a more limited number and kind of opportunities for most citizens to exercise political agency.
The natural response to the limitations of both strategies is to look for ways to combine them. That is what happens in many democracies and what most democratic theories call for. Governing officials are theoretically held accountable for their performance at election time, and those deemed irresponsible or incompetent lose the power and privileges of their office. At the same time, in most democratic societies, citizens have other opportunities to intervene and influence officials' behavior through formal and informal intraparty governance, the ability to form or defect to more ideologically congruent parties, pressure tactics, or instruments of direct democracy such as citizens' initiatives.
Advocates of RPIM claim, though, that this hybrid approach is counterproductive. They argue that institutions for responsiveness undermine some of the critical prerequisites for the responsibility strategy to work. Instead of trying to have the best of both strategies, then, the theory of responsible parties insists that we commit to the responsibility strategy. Despite its limitations, RPIM represents the best available option.
For the responsibility strategy to yield well-behaved representatives, several conditions must be met. First, voters must be able to accurately attribute responsibility for outcomes to the actions of officials that produced them. Otherwise, voters will not be able to distinguish desirable from undesirable behavior. Second, voters must be able to credibly threaten to remove poor performing politicians. Following through with appropriate consequences must not be too costly for voters. Finally, the politicians who are being held responsible must have the capacity to rationally respond to incentives.
Problems with all three of these prerequisites for responsibility arise in the politics of electoral democracy. And proponents of RPIM argue that most institutions of responsiveness make these problems more severe. As a result, responsive institutions compromise the mechanisms by which popular elections are supposed to induce good public policy. According to this argument, to secure policies that best promote citizens' long-term public interests, we need institutions that maximize the effectiveness of the responsibility strategy. A system with two large, centralized parties competing for full control of government is supposed to do just that.
This argument for RPIM is a common-sense realist argument. It focuses on a particular good: policy that serves the public interest in the long-term. And it contends that RPIM can secure this good much more reliably than other institutional arrangements. The realist argument focuses on demonstrating the validity of this empirical claim rather than defending the importance of this variety of policy outcome relative to other evaluative criteria, such as "representation for its own sake" (Rosenbluth and Shapiro 2018, 18).

THE REALIST ARGUMENT FOR RPIM
The two most comprehensive accounts of responsible parties (Rosenbluth and Shapiro 2018;Schattschneider 1942) both tout the benefits of competitive two-party systems and of institutions that enhance party discipline and insulate officeholders from the pressures of ideological activists. These two books also provide similar arguments for preferring such systems to more responsive alternatives. They argue that the responsible parties institutional model sharpens incentives for representatives to pursue broad public interests, or as Rosenbluth and Shapiro put it, "to give most voters most of what they want most of the time" (Rosenbluth and Shapiro 2018, 26).
In this section, I will reconstruct the argument that RPIM secures the kind of public policy that best serves the public interest in the long term. This argument consists of three premises. First, RPIM establishes the prerequisites for responsibility, enhancing electoral accountability and enabling parties to act as effective collective agents in responding to electoral incentives. Second, when these prerequisites for responsibility are in place, elections will incentivize parties to engage in programmatic competition, campaigning on the promise of policy programs that they think will appeal to a majority of voters and working to implement these programs when in government. Finally, when parties are incentivized to engage in programmatic competition, they will produce public policy that best serves the public interest in the long term.
Most models of representation assume that, at least some of the time, the desires of representatives diverge from those of the citizens they are supposed to represent. Despite these diverging interests, citizens have a tool for incentivizing elected officials to serve the public: elections. At election time, citizens assess their representatives' performance and decide whether they should continue to entrust them with responsibility for governing. Officials who do not satisfactorily discharge their responsibilities will lose the power and privileges of their office. As they presumably want to stay in office, elected officials will do what citizens want so that they will be judged favorably and reelected. This is the theory of retrospective electoral accountability. 6 Of course, things are not really that simple. Problems with all three of the prerequisites of responsibility can undermine elections' effectiveness as accountability mechanisms. To begin with, voters sometimes struggle to correctly attribute responsibility to elected officials (see Healy and Malhotra 2013, 290-3). Citizens are unable to observe most of their representatives' actual behavior. Instead, citizens observe a subset of behaviors and outcomes. The problem is that what citizens can and do observe is often only loosely related to the attributes or behaviors of elected officials that they are trying to assess, especially when policies and policymaking processes are very complex. If voters cannot accurately determine when their representatives are doing a good job, it may not be worth it for officials to direct their efforts toward the public interest, especially if their own preferences lie elsewhere.
Even when citizens can correctly attribute responsibility and assess incumbents' reliability, they are limited in what they can do with that information. The incentives that incumbents face due to the expectation of being held responsible on Election Day depend on the quality of available alternatives. Representatives only need to outperform citizens' expectation of what they'd get if they elected the challenger rather than the incumbent.
According to its proponents, a system in which two parties compete for control of government can mitigate these two problems, thereby enhancing electoral accountability. Under RPIM, one party always has full control of government while another performs the essential role of opposition. This simplifies responsibility attribution. Without full control of government, incumbents will try to avoid taking responsibility for any perceived shortcomings of their turn in government. Full control of government, in contrast, should incentivize politicians to work harder to please voters because the governing party knows that all the blame will fall on them if things go poorly (Rosenbluth and Shapiro 2018, 36).
The flipside of this full control of government is unified opposition. A unified opposition serves two roles: (1) to help citizens monitor the party in power and (2) to serve as a government in waiting, a credible alternative to take over for a poor-performing incumbent party. Incumbents generally want to help citizens observe their successes, but to obscure their failures. The opposition, on the other hand, has an incentive to expose and publicize any bad behavior or incompetence on the part of incumbents (Rosenbluth and Shapiro 2018, 37). Meanwhile, the opposition seeking to win election also has an incentive to make themselves appear as credible an alternative to the incumbent party as possible, helping to solve the problem of replacement quality (Rosenbluth and Shapiro 2018, 37;Schattschneider 1942, 82).
Two-party systems are purportedly desirable, then, because they help mitigate two of the potential problems with the responsibility strategy. By simplifying control of government and creating an opposition with strong incentives to expose the governing party's shortcomings, two-party systems aid responsibility attribution. And by enabling a unified opposition to serve as a credible government in waiting, two-party systems make it more likely that citizens will electorally punish incumbents who perform badly.
The other major component of RPIM, which specifies a disciplined internal party organization with officeholders at the helm, addresses the third prerequisite for the responsibility strategy: parties' capacity to rationally respond to the collective incentives that elections create.
Reliable collective agency is often thought to require (or at least benefit from) centralization and hierarchy to overcome the social choice problems and collective action problems that bedevil large groups and, thus, to enable the group to "think" and act rationally. The argument for RPIM relies on this logic. It holds that parties are only responsible when they are controlled by a small number of elites who are motivated to win and maintain office and who are collectively capable of strategically responding to electoral incentives (Rosenbluth and Shapiro 2018, 20-1;Schattschneider 1942, 193, 207). Should they win control of government, they are able to implement whatever policy program they think will appeal to most voters. Meanwhile, when not in control of government, responsible parties form a unified opposition, pursuing a coherent strategy to unseat and replace the governing party.
Democratized parties might be able to perform these functions of responsible parties, but it is harder to constitute an effective group agent with a democratic structure. One reason is that democratized parties may find it harder to overcome collective action problems. Individual members of the party have their own projects that may conflict with the strategy that maximizes the party's chances of winning control of government. This can lead to free-riding and assurance problems. Coordination problems can also arise because of disagreement over the sorts of messages or policy programs that will best serve the public interest and/or maximize the party's electoral prospects. But a hierarchical party organization can dictate a common strategy and align the incentives of individual politicians with those of the party; it can discipline them to hold the party line.
In addition to the claim that RPIM enhances electoral accountability by establishing the prerequisites for responsibility, the argument for RPIM depends on two further premises. The second premise holds that parties will respond to RPIM's enhanced electoral accountability by pursuing coherent, broad-based, and long-term policy programs. The third premise holds that parties' adopting this strategy best serves the public interest.
In support of the second premise, proponents of RPIM argue that responsible parties will seek to secure durable majority coalitions in the electorate and that campaigning on and implementing long-term, broadbased legislative programs will best enable them to do that (Rosenbluth and Shapiro 2018, esp. 72;Schattschneider 1942, 206-10). Additionally, in support of the second and third premises, Rosenbluth and Shapiro cite the record of "Westminster in its hey-dey" (Rosenbluth and Shapiro 2018, 237), the best historical exemplar of RPIM, which, they assert, succeeded at "implementing policies with long-term benefits for most people most of the time over the long haul" (Rosenbluth and Shapiro 2018, 62). However, the bulk of both Rosenbluth and Shapiro's and Schattschneider's arguments for the link between RPIM, programmatic policies, and the public interest involves illuminating the pressures that lead to incoherent platforms and policies that serve narrow interests under other institutional models 7 -pressures that RPIM is designed to eliminate.
The realist argument for RPIM thus rests on three empirical premises about its ability to deliver a particular good-policy that serves public interest-more reliably than the alternatives. In the next section, I will argue that this argument fails common-sense realism's robustness test. Its premises rest on several contingent assumptions about how voters judge politicians, about the control that even unified party governments have over policy outcomes, and about the relative costs of different electoral strategies.

THE ROBUSTNESS TEST: EXAMINING RPIM'S ASSUMPTIONS
The argument for RPIM that I canvassed in the previous section rests on three empirical claims: (1) RPIM enhances electoral accountability, (2) RPIM's enhanced electoral accountability incentivizes programmatic competition, and (3) programmatic competition leads to long-term, public-interested policies. In this section, I argue that although there is evidence that the first claim holds true across a variety of contexts, the same cannot be said of the others. This is likely because of the long "causal chain" linking institutions to the promotion of particular policies (Ferree, Powell, and Scheiner 2014). The claim that electoral incentives produced by RPIM translate into a particular kind of policy outcome rests on many assumptions about the beliefs, capacity, and motivation of both voters and politicians. The value of RPIM is not robust to changes in these assumptions.
Proponents of RPIM have used historical case studies examining the performance of various regimes to provide support for their empirical claims. But this kind of evidence can only go so far in demonstrating the robustness of an institutional arrangement. Broad, cross-national studies of electoral systems present a more complicated picture. On the one hand, crossnational studies do lend support to the idea that retrospective accountability is greater in systems that resemble RPIM. Vote choice is more closely tied to satisfaction with government where voters expect elections to produce greater alternation of parties in government (Otjes and Stiers 2022). Perceptions of economic performance are also more predictive of vote share for parties that have more government control, especially when the party's position in government is at stake (Duch and Stevenson 2008, 285).
On the other hand, large cross-national studies have not found a consistent relationship between electoral institutions and either congruence with median policy preferences or objective measures of economic performance (a common proxy for long-term, publicinterested policy making). Though there have been many efforts to identify this relationship, they have produced different and contradictory results, in part because they are based on different sets of observations (Carey and Hix 2013, 53-4;Grofman 2016, 534-5). This does not mean that institutional arrangements do not affect policy outcomes. Rather, it means that these effects are mediated by factors that vary across time and across political contexts. The value of RPIM is not robust to these changes because it is based on a singular story about how electoral institutions structure policymaking incentives.
Of course, most electoral systems research examines only one dimension of institutional variation at a time. But it is not necessarily the case that partial realizations of an institutional model yield proportional realizations of its value. Thinking otherwise, we fall victim to the "fallacy of approximation." If the benefits of RPIM depend on a sui generis interaction of its various components, then research on any one component of RPIM does not help us assess the claims made on its behalf. I will revisit this approximation problem in the next section when I discuss the stability and feasibility tests. For the remainder of this section, though, I will set it aside. Instead, this section discusses several explanations for why we might observe enhanced retrospective accountability without superior policy outcomes even in a fully realized system of RPIM. These explanations describe circumstances in which the second or third premise of the argument for RPIM might turn out to be false. The point of enumerating these potential explanations is to illustrate the complexity of the causal chain linking electoral institutions to policy outcomes and thus the many sources of contingency that might undermine the value of RPIM.
The first potential source of contingency in the link between electoral accountability and policy outcomes is multidimensionality in the space of policy competition. Competing on multiple issue dimensions means that parties can choose to emphasize issue areas where they are at a relative advantage. This option complicates the claim that two-party competition leads to better monitoring. Some issues are easier to monitor than others are, and strategic opposition parties will focus their monitoring efforts in areas where it is most cost effective regardless of whether those areas are most important to the public interest.
Proponents of RPIM argue that two-party competition incentivizes parties to engage in stable, programmatic policy competition as they pursue durable majorities. But multidimensionality also complicates this claim. First, multidimensionality enables parties to pull away from the median by constructing a majority coalition from multiple intense minorities. Second, these coalitions will not necessarily be stable. Miller and Schofield (2003) model how two parties in a multidimensional competitive environment select their political platforms. When a party faces a competitive disadvantage on one salient dimension of conflict, they can undertake a "flanking" maneuver to peel off some of their opponents' supporters by taking a more divergent position on another issue. 8 Instead of forcing the parties to improve their performance or build longterm policy programs, two-party competition can induce parties to circle around one another in pursuit of an issue that is easier to win. This is especially true as office-seeking party leaders are less constrained by policy-motivated activists within the party.
Governing parties can introduce new issue dimensions that scramble political debates, too, and they have an incentive to do so when they are performing badly and improving their performance is too costly or unreliable. Consider, for example, the 1987 New Zealand election in which the Labour party, under the leadership of David Lange, won reelection despite their unpopular economic program. The most salient issue in the election that initially brought Lange's Labour party into power was an economic crisis, and the Labour government implemented a sweeping program of economic policy changes in response. But the program was not popular with voters. In their 1987 reelection campaign, Lange and the Labour party did not expend much effort trying to change voter's assessments of their economic program. Instead, they focused the campaign on their popular antinuclear position and foreign policy. Multidimensionality, then, can sometimes allow governing parties to escape punishment for pursuing unpopular policies.
To be clear, this does not imply that multidimensional competition is bad for democracy. It does, however, complicate the translation of electoral incentives into campaign strategies and policy choices. It thus provides an explanation for why RPIM may not reliably produce broad-based and long-term policy programs.
The potential instability of party alignments under multidimensional competition can inhibit efforts to implement and maintain political programs that span many election cycles. But sharpening retrospective accountability may also more directly inhibit long-term thinking. Recency bias provides a second explanation for why enhanced retrospective accountability may not produce the policy outcomes touted by its proponents. The prevalence of recency bias in citizens' political judgment suggests that citizens' assessment of a party's tenure in office will be heavily weighted toward the months immediately preceding the election (Huber, Hill, and Lenz 2012). Researchers have demonstrated this most frequently with respect to economic performance, but it also appears in disaster readiness and response: voters reward spending on disaster relief but not on more cost-effective disaster preparedness (Healy and Malhotra 2009). Politicians anticipate and respond to these electoral incentives, providing more economic stimulus, relief, and other benefit spending in election years (Healy and Malhotra 2013, 298).
Even if party leaders are willing to accept short-term losses to enhance their long-term electoral prospects (Rosenbluth and Shapiro 2018, 4), they may not be able to do much to affect the party's future reputation. The problem is not just that voters punish decisions that impose short-term costs; evidence of recency bias suggests that voters will not reward those decisions later when they realize long-term benefits. Institutions that incentivize governments to respond more strongly to retrospective accountability do not necessarily promote longtermism.
A third source of contingency in the relationship between electoral incentives and policy outcomes comes from policy feedback effects. Public policies do not just reflect the political environment. They also shape it. Policy feedback effects occur when the implementation of a policy affects the preferences and political behavior of citizens. Policy feedback effects present a tool for parties to shape their long-term prospects without relying on voters to recognize and reward successful decisions in the past. They can do so by implementing policies that create new constituencies or that enhance (or diminish) the participatory capacities of others (Campbell 2012).
Parties who are concerned about their long-term electoral prospects may try not only to anticipate and please but also to shape their future electorate. Such activity might serve broad-based and long-term public interests. But it might not. Looking ahead to future elections could, for example, incentivize parties to drag their feet on implementing popular and beneficial programs (or to leave them incomplete or underfunded) so that they can continue campaigning on an advantageous issue (Howell, Krasa, and Polborn 2020, 562).
These first three sources of contingency in the relationship between electoral institutions and policy outcomes highlight different ways that electoral incentives might shape the policies that parties propose and implement. Additional complexity can arise from non-policyrelated campaign strategies that politicians might deploy. For example, electoral campaigns, especially in plurality rule systems with high-variance turnout, may focus on exploiting relative advantages in the turnout battle and strategically deploying mobilization resources. This aspect of campaign strategy may have also contributed to Labour's victory in the 1987 New Zealand discussed above. Labour conducted what has been called "New Zealand's first modern campaign," consulting with professional political tacticians to target constituencies with the greatest marginal value. The strategy was successful. As one commentary put it, Labour "lost the campaign but won the election" (Boston and Jackson 1988, 74).
Widespread political and social trends can also affect the strategic appeal of different campaign strategies. For example, increased personalization in media coverage of politics, along with new communication technologies, can both increase the benefits and decrease the costs of campaigning on personal qualities rather than the party's policy program. Centralized party organizations do not appear sufficient to resist the global trend toward personalization (Rahat and Kenig 2018). At the same time, the broad trend toward greater economic integration and interdependence diminishes the control that governments have over the outcomes that voters might otherwise hold them accountable for (Duch and Stevenson 2008, 205). This makes responsible policy making less reliable as a strategy for incumbents to win reelection, incentivizing alternative electoral strategies.
The potential explanations for why enhancing retrospective accountability might not translate into superior outcomes that I have discussed thus far have raised doubts about whether parties' pursuing optimal electoral strategies under RPIM will reliably produce policy outcomes that serve voters' interests in the long run. As this discussion has demonstrated, there are many contingencies that could produce violations of this assumption. The realist case for RPIM is not robust to these violations because it is entirely predicated on the electoral incentives it creates for elites to serve voters' interests.
To be clear, this discussion does not show that RPIM's assumption about optimal electoral strategy never holds. And it certainly does not show that there is no accountability to voters' interests when it doesn't. The point, rather, is that RPIM's enhanced accountability does not reliably deliver the kinds of outcomes that purportedly justify it. This is a problem because enhanced electoral accountability is the main value proposition of RPIM and it comes at the cost of other potential sources of value.
Before moving on to discuss the stability and feasibility tests, it is worth raising one more potential concern about RPIM's robustness. In addition to its assumption about optimal electoral strategy, the realist argument for RPIM depends on an assumption about politicians' motivations. The argument assumes that politicians will seek to maximize their probability of winning office over the long term and will therefore choose maximally effective electoral strategies. This assumption, however, is also contingent.
Politicians presumably have a variety of professional, ideological, and personal interests. To the extent that winning office is instrumental to the satisfaction of their interests, we can expect party leaders to pursue effective electoral strategies, even if doing so is costly (Calvert 1985). But there are limits to this logic. So long as there is an element of randomness or incomplete information, politicians may sometimes choose electoral strategies that have a lower probability of success. They may do so if these strategies come with greater returns to winning, lower costs to losing, or lower costs in terms of other things they value (e.g., leisure, money, or pride). Even when programmatic politics is a relatively effective electoral strategy, then, other strategies that are less costly or have higher upside or lower downside risk may ultimately have more appeal.
The "cartel party thesis" holds that mainstream political parties across many modern democracies avoid engaging in full-throated competition with each other (Katz and Mair 2018). Instead, they cooperate to promote a number of shared interests, and they compete over an increasingly small and low-stakes set of political issues. A central claim from the cartel party thesis that I want to draw out here is that, in the right circumstances, party rationality shifts "away from maximizing the expected (average) pay-off or probability of victory and toward maximizing the reasonably anticipated minimum pay-off ('maximin') even in defeat" (Katz and Mair 2018, 16).
This shift in party rationality occurs when aggressive competition (maximizing the probability of victory) becomes less appealing and when cooperation with opponents becomes more appealing. Proponents of the cartel party thesis argue that this occurred across Western Europe around the turn of the twenty-first century. First, as discussed above, governments had less control over economic conditions because of global interdependence (Katz and Mair 2018, 9). This diminished control decreased the appeal of programmatic competition, as responsible policy making could not guarantee outcomes that voters would reward at the polls. Second, cooperation became more appealing because decision makers within the parties increasingly shared a set of professional interests that they wished to protect (Katz and Mair 2018, 72-8).
The cartel party thesis was developed to describe and explain a particular historical moment that may well be shrinking in our rearview mirror. Nevertheless, this theory still carries some important lessons for evaluating assumptions about politicians' motivations. Although cartelization is not a unique risk of RPIM, there is no reason to think that RPIM will make polities immune to structural forces that decrease the expected electoral returns to programmatic campaign strategies. At the same time the kind of party organization that RPIM calls for-centered on officeholders, not activists -makes it likely that politicians will develop a similar set of professional interests and an incentive to insulate them from the ups and downs of electoral fortunes (Katz and Mair 2018, 61-72).
The purpose of this section has been to demonstrate that the assumptions underlying the realist argument for RPIM are highly contingent. To do so, I have enumerated several scenarios where these assumptions might be violated. Most of these arise from issues that are endemic to any model of democracy-like multidimensionality and recency bias-or from global trends-like economic interdependence and technology change. But although these issues are not unique to RPIM, they nevertheless present a serious challenge to the realist argument for RPIM because they illustrate the contingency of its underlying assumptions. There are many common political forces that could lead to violations of these assumptions, and the realist case for RPIM is not robust to those violations.

FROM ROBUSTNESS TO FEASIBILITY AND STABILITY: WHY REALISTS SHOULD WANT REDUNDANCY
The realist argument for RPIM that I have reconstructed and critiqued in this paper defends this institutional model on the grounds that it reliably delivers good public policy. As I argued in the previous section, though, the assumptions underlying this argument are contingent, and there are good reasons to think that they are often false. When these assumptions are violated, RPIM cannot be expected to reliably deliver the value that purportedly justifies it. RPIM, then, fares poorly against common-sense realism's robustness test.
Proponents of RPIM may reasonably ask whether the alternatives are better in this respect. Politics is a complex and messy business. All of our arguments about it contain some contingency and uncertainty. Arguments for responsive institutions are built on their own set of contingent assumptions about the incentives these institutions create and how political actors will respond to them.
Contingency is a particularly serious problem for the realist argument for RPIM, though. This is true for two reasons. First, realist arguments counsel us to forgo the risky pursuit of some ideal to secure a (possibly) less attractive, but more certain, source of value. Realist arguments therefore bear a substantial burden of proof to demonstrate that the good they target really is more easily secured. Common-sense realist arguments are often deployed as a way of circumventing disagreement about the relative value of competing ideals. Even if the proponents of a realist argument do not think that pursuing their recommended path involves settling for a lesser good, for the argument to succeed qua realist argument, it must be convincing to those that Realism and Responsible Parties do. The certainty required to justify such settling (and thus the burden of proof that realist arguments bear) depends on the size of the gap between the value of the goods that a realist proposal would secure relative to the value of other potential goods that it closes off. A corollary is that the more empirical uncertainty in an argument, the more we need to know about the relative value of the competing ideals or goods at stake.
The second reason that contingency poses a problem for RPIM arises from the specifics of the model. RPIM takes a precision engineering approach to institutional design. The various pieces of the model-full party control of government, unified opposition, officeseeking party leaders, competitive districts, persuadable voters-all work together to produce the desired result: responsible party government. When it works well, this machine may be very good at governing in the public interest. The trouble is that the machine is delicate: because every piece is essential, there are many failure points. Contingency is a special problem for RPIM because the model lacks redundancy.
This lack of redundancy becomes even more of a liability when we move from the robustness test to the feasibility and stability tests. These tests invite us to think not just about what happens once an institutional arrangement is in place but also about how institutional arrangements come to be. Once we are thinking about this, we should be much more skeptical of precisionengineered institutional models.
We can examine two aspects of the feasibility of RPIM: the feasibility of realizing the complete institutional package and the feasibility of realizing incremental reforms that move toward a closer approximation of the model. Taken as a whole, the institutional package of RPIM does not fare well against the feasibility test. This is because institutional reform in modern democracies usually happens piecemeal and at the margins. This reality is reflected in the fact that many regimes exhibit some but not all of the characteristics of RPIM.
No doubt for this reason, some proponents of RPIM recommend incremental reforms that move political systems closer to the ideal (Rosenbluth and Shapiro 2018, 239). Even this approach faces significant feasibility challenges. Many of the recommended reforms are unpopular. Although they may be frustrated with the state of electoral democracy, voters and interest groups are hardly clamoring to give back the power over candidate selection and platform creation that they believe they have wrested from politicians.
The more serious problem, though, is that it is unclear how much value there is to gain from such partial successes. As discussed above, political systems that more resemble RPIM do not clearly outperform those that resemble it less. This is likely because the value of each component of RPIM depends on how it interacts with the others. For example, scholars have long observed that the single-member districts RPIM calls for can produce disproportionate electoral outcomes and distorted electoral incentives due to nonuniform geographic distribution of political preferences (Powell 2004, 239). To address this, Shapiro and Rosenbluth's specification of RPIM recommends pie-slice districts that each include parts of urban, suburban, and rural areas to create constituencies that are more similar to each other (Rosenbluth and Shapiro 2018, 239). But representative districts are not easy to achieve or maintain. In fact, they are the one element missing from what Rosenbluth and Shapiro regard as the closest historical approximation of RPIM (Rosenbluth and Shapiro 2018, 92). The absence of this one element attenuates the value of the institutional arrangement, and something similar can be said for the various other components.
The feasibility test, then, presents something of a catch-22 for RPIM. On the one hand, if we focus on the complete institutional package, it appears highly infeasible. On the other hand, reforms that might realize some but not all components of RPIM raise the approximation problem. Because the value of RPIM arises from how its various components interact to produce a desirable pattern of electoral incentives, it is not clear that better approximations of the model yield proportionately better approximations of its value. Institutional reform packages that can be decomposed into independently valuable proposals will typically fare better against the feasibility test.
Supposing, though, that we could get RPIM in place, could we keep it? When it comes to stability, again, RPIM's prospects look poor. In the previous section, I discussed incentives and opportunities that officeholders might have to avoid full-throated programmatic competition, even within the framework of RPIM. But there is another important strategy that they can pursue, which is to change the institutions to secure an advantage or entrench their position. It may be easier, for example, for parties to win votes by adopting popular intraparty governance reforms than to deliver to voters what they want. Especially when ambitious social and economic programs are hard to deliver, institutional reforms and changes to the party organization can provide a way for party leaders to maintain the support of their otherwise disenchanted voting base (cf. Wolkenstein 2019). RPIM's lack of redundancy rears its head again here: if one part of the model is unstable, it compromises the value of the whole package.
After examining RPIM's prospects for success across all three dimensions, then, the realist case does not look very strong. RPIM is not obviously more robust, feasible, or stable than the institutional arrangements it would supplant or the marginal reforms that proponents of greater responsiveness might pursue. The main culprits are contingency and lack of redundancy.
It is worth noting here that this discussion does not imply that proposals for more responsive institutions will necessarily have greater prospects for success. To begin with, there are many such proposals, which no doubt vary in terms of their feasibility, stability, and robustness. Furthermore, redundancy has its own pitfalls, most especially the potential for negative interactions among different mechanisms to constrain power. Even if the argument for RPIM does not demonstrate its superiority on realist grounds, it does highlight legitimate concerns about responsiveness that democratic reformers must take seriously. The point of my argument here is not to show that responsiveness strategies outperform RPIM or that there is no case to be made for responsible parties. Rather, it is to show that RPIM does not represent a clear realist choice. It does not reliably deliver its purported value. A successful argument for RPIM, then, must involve comprehensive consideration of the various goods at stake. In the final section of this paper, I sketch the shape of such an argument and the challenges it will have to face. But first, I respond to a different sort of realist argument that proponents of RPIM might make: that pursuing RPIM represents a maximin strategy for avoiding catastrophic democratic collapse.

WHAT ABOUT DOWNSIDE RISKS?
Proponents of RPIM might argue that there is another realist case to be made for RPIM. Realists do not just care about a proposal's prospects for success; realists also want to know about the costs of failure. Even if RPIM cannot be counted on to deliver policies that serve broad public interests, its proponents might nevertheless argue that the potential for catastrophic failure is lower for RPIM than for its alternatives (including the status quo).
At first glance, this argument seems plausible; the most spectacular example of democratic collapse, the Weimar Republic, occurred in an institutional context as different from RPIM as possible. Democratic backsliding in Hungary appears to reinforce this cautionary tale. In both cases, a multiparty system yielded a weak and fragmented opposition, unable to resist an authoritarian leader's consolidation of power.
On the other hand, there may not be similar examples of spectacular failure in RPIM simply because the sample size is much smaller. The handful of good examples of RPIM lasted for a few decades during the mid-twentieth century. As Rosenbluth and Shapiro observe, they seem to have performed well. However, as they likewise observe, many democratic regimes seem to have performed well under the favorable conditions of the postwar period (Rosenbluth and Shapiro 2018, 19). To characterize the downside risks of attempts to realize RPIM, we need to consider how it would perform under the same pressures that have caused other institutional arrangements to falter (cf. Kuo 2019). Some scholars, citing the example of recent protests in Chile, have cautioned that systems that limit opportunities for citizen participation and insulate governing elites from interelection influence are more susceptible to populist backlash (Rhodes-Purdy and Rosenblatt 2021).
There is another reason to doubt that pursuing RPIM is a maximin strategy. Given the difficulties in establishing and maintaining the full institutional model that I discussed in the previous section, consideration of RPIM's downside risks should focus on the risks of institutional arrangements with some but not all of RPIM's components. One potentially catastrophic outcome occurs when a system asymmetrically approximates RPIM: it features two-party competition and grants the victorious party full control of government; however, only one party manages to maintain a centralized party discipline. If the disciplined party electorally dominates the other, its leaders will be able to operate free from not only electoral accountability induced by a strong competitor but also the constraints of intraparty democracy and other mechanisms of responsiveness. The result of this approximation of RPIM is thus very similar to the catastrophic results that its proponents worry about in multiparty systems without a unified opposition.
Again, this discussion does not show that RPIM's downside risks are necessarily more severe than those of any existing regime or proposed alternatives. Rather it seems that, given the available evidence, we cannot say with confidence that pursuing RPIM is a reform path that minimizes the risk of catastrophic failure.

THE IDEALIST DEBATE: WHAT ARE THE GOODS OF DEMOCRACY?
In this paper, I have argued that the realist argument for the responsible party institutional model does not succeed. RPIM is not sufficiently robust, feasible, or stable, nor is it a clear maximin strategy. This does not mean that there is no argument for RPIM. But if there is such an argument it must be based on the attractiveness, not the modesty, of its objectives.
In this final section, I will describe one such argument 9 that is built on a distinction between "public" interests, and "private" or "sectoral" interests, along with a normative claim that we ought to choose political institutions that promote public rather than sectoral interests. Public interests relate to costs and benefits that accrue to everyone, such as the benefits of clean air and water, or the cost of inflation. In contrast, private or sectoral interests relate to benefits and costs that disproportionately accrue to only a subset of citizens, such as the benefits of religious accommodation or the cost of industry regulation.
Underlying the case for RPIM is the idea that responsible parties tend to promote public interests, whereas responsive parties tend to promote sectoral interests (Rosenbluth and Shapiro 2018, 92-4;Schattschneider 1942, 193, 206). In fact, responsive systems often seek to promote sectoral interests by design. Responsiveness mechanisms sometimes disperse power among 9 This is not the only available argument for the attractiveness of RPIM. Robert Goodin, for example, has argued for a similar model of party politics on the grounds that it is uniquely capable of generating a coherent ratio for the laws, something that he argues is necessary for citizens to be truly self-governing (Goodin 2008, 216-23). A similar argument might be derived from accounts that identify coherent narratives as among the chief contributions of political parties (Rosenblum 2008). This argument faces its own challenges in responding to alternative conceptions of self-government and deliberative virtues that might be better realized by alternative institutional models (e.g., Bonotti 2017;Wolkenstein 2016). It also faces the same challenge as the argument from public interests in addressing countervailing procedural values. different groups known to have conflicting "private" interests. These mechanisms are meant to facilitate equitable bargains-"logrolls" to their detractors (Rosenbluth and Shapiro 2018, 72, 130)-among various sectoral interests. The debate between proponents and critics of RPIM, then, is not just about empirics, it is about the relative value of these two visions: a programmatic politics of the "public" interest or a more transactional politics built on equitable bargaining.
Proponents of responsible parties argue that bargains among sectoral interests are often undesirable even if they are equitable. Rosenbluth and Shapiro (2018) argue that private interests are less stable than public interests, meaning that political bargains among competing interests may end up disproportionately harming large groups of citizens in the long run, even if they seemed like fair bargains initially (160-77). They also suggest that promoting public interests tends to produce more efficient outcomes than bargains among private interests (21-2). Schattschneider (1942), on the other hand, adopts a principled stance that public power should be used to promote public interests rather than an agglomeration of private interests (204). This not-so-realist argument for RPIM, then, emphasizes the superior value of a politics of public interest over the muddled compromises of responsive party systems.
For this argument to succeed it must overcome three kinds of challenges. The first challenge disputes the claim that a programmatic majoritarian politics is necessarily superior to a more consensual model of politics. Advocates of RPIM claim that the point of democracy is to give "most people most of what they want most of the time" (Rosenbluth and Shapiro 2018, 26). But others might argue that we should also consider preference intensity-what people most want. 10 There are several reasons why we might care about preference intensity. First, ignoring extreme differences in preference intensity may create political instability (Dahl 2006, 92-9). Second, unless we assume that public interests never conflict, a democratic justification for prioritizing one public interest over another will require reference to either preference intensity (Ingham 2019) or to procedural values. Finally, there is the intuition that some of our wants just matter much more than others to how our life is going.
The second kind of challenge takes aim at the distinction between public and private interests itself. A large body of democratic theory holds that political interests and preferences only take shape as a result of politics (Disch 2011). A valuable form of democracy, then, must include citizens in the creative work of defining public interests. Party systems should be evaluated with this ideal in mind (Chapman 2022, 150).
A third set of challenges that any argument for RPIM must overcome arises from the additional sources of value that might be found in responsive systems apart from their tendency to produce a certain pattern of policies. Much of what makes responsive party systems attractive to their supporters is the opportunity for direct participation and the chance to choose from a wider range of attractive options at the ballot box. These features of responsive systems may have some inherent value as opportunities for individuals to exercise political agency. The theory that RPIM enhances democracy has been described as paradoxical (Rosenbluth and Shapiro 2018, 20) precisely because it requires eliminating opportunities for influence that are thought to have inherent democratic value.
Responsive systems may also have desirable effects on political culture that should be considered when evaluating electoral institutions. Proportional representation systems tend to enjoy higher voter turnout, perhaps because they create stronger incentives for parties to invest in broad mobilization infrastructure than systems with single-member districts (Cox 2015). Yet another source of value to be considered is the nature of political-and especially partisan-socialization that occurs in responsive party systems. Affective polarization, for example, may be lower in systems with proportional representation (especially where governing coalitions are unpredictable; Westwood et al. 2018, 335). The nature of citizens' attitudes toward and relationship with those on the other side of political cleavages features prominently in many accounts of democracy's value (Gutmann and Thompson 1996). Therefore, assessing alternative political institutions should involve considering how different institutions affect the development of democratic attitudes.

CONCLUSION
Realist arguments can be a useful way of getting around intractable disagreements about ultimate ideals. People who may disagree about what the best possible version of democracy looks like may nevertheless agree on the value of a more modest good. And if that modest good is readily attainable, we may reasonably decide to forgo opportunities to pursue more ambitious ideals. But precisely because of their opportunity costs, realist proposals face an empirical burden of proof. We need to know that the realist proposal really is the safer bet. The higher the opportunity cost, the higher the burden of proof.
Some realist arguments about democracy can meet this burden of proof. There is, for example, a great deal of evidence that competitive elections constrain elite power and limit certain very bad outcomes. Consequently, we can reject, on realist grounds, radical proposals like epistocracy (Bagg 2018) or lottocracy (Chapman 2022;Kirshner 2022) that jeopardize these accomplishments in pursuit of more uncertain political ideals.
But when it comes to the choice among different kinds of competitive electoral systems, we cannot have the same degree of confidence. Realism sets limits on our reform ambitions, but it will rarely determine them. The debate over the responsible parties institutional model illustrates this point. The case for RPIM is built on modest and plausible assumptions about political motivations. But because these assumptions are contingent and because RPIM has no redundancies, there is substantial uncertainty around its prospects for success. Pursuing RPIM, then, just like pursuing any other reform agenda, carries risks. In fact, because RPIM entails concentrating power, it carries the kind of risk that realists typically worry about most. Choosing between the pursuit of responsible or responsive party systems means choosing which risks we'd rather take. To make that choice, we will have to move beyond realist arguments to a more comprehensive debate over the goods at stake.

ACKNOWLEDGMENTS
Previous versions of this paper were presented at the American Political Science Association Annual Meeting, the European Consortium for Political Research General Conference, and the Northwestern Political Theory Workshop. I am thankful to participants for helpful comments and conversations. I am also grateful to Andy Sabl, Sam Bagg, Alison McQueen, Avi Acharya, Gary Cox, Lise Herman, and three anonymous reviewers for feedback and suggestions on earlier drafts.

CONFLICT OF INTEREST
The author declares no ethical issues or conflicts of interest in this research.

ETHICAL STANDARDS
The author affirms this research did not involve human subjects.