The link between public opinion and public policy is of special importance in representative democracies, as we expect elected officials to care about what voters think. Not surprisingly, a large body of literature tests whether policy is a function of public preferences. Some literature also considers the mechanisms by which preferences are converted to policy. Yet other work explores whether and how the magnitude of opinion representation varies systematically across issues and political institutions. In all this research, public opinion is an independent variable—an important driver of public policy change—but it is also a dependent variable, one that is a consequence of policy itself. Indeed, the ongoing existence of both policy representation and public responsiveness is critical to the functioning of representative democracy.
Extended scope of review to include new lines of research, added citations to new research, and included discussions of findings.
The representation of public opinion in public policy is of obvious importance in representative democracies. Not surprisingly, there is a considerable body of research addressing the connection between the public and the government. This work is broad and varied. Some research focuses on “descriptive representation”—whether the partisan and demographic characteristics of elected politicians match the characteristics of the electorate itself. Other studies examine the positions of policymakers, observed, for instance, through the roll call voting behavior of politicians. Still other research focuses on policy more directly, concentrating on legislative (and budgetary) outcomes.
Descriptive traits, roll call voting behavior, and policy outcomes capture rather different aspects of representation, but each can be useful in gauging the nature and quality of democratic responsiveness. This article reviews the literature on representative behaviors and outputs. The mechanics by which preferences are converted to policy are considered, along with extensions of the basic model—where the magnitude of opinion representation varies systematically across issues and political institutions. For most of the article, public opinion is an independent variable—an important driver of public policy change. In a concluding section, however, we reconsider opinion as a dependent variable; specifically, we discuss its responsiveness to policy change. The ongoing existence of both policy representation and public responsiveness to policy, we argue, is critical to the functioning of representative democracy.
A fundamental expectation of democratic governance is that policy will be a function of opinion (see, e.g., Dahl, 1971; Pitkin, 1967). 1 We can express this expectation formally as follows:
where P designates policy and O opinion, say, the public’s average preference for redistribution. This is not meant to be a complete model of policy, of course, as we know that many other things also matter (Jacobs & Shapiro, 2000). The equation is simply intended to characterize the general relationship between opinion and policy. To be absolutely clear, we expect a positive relationship—when the public wants a lot of redistribution, for example, they should get a lot of it. Whether and the extent to which this is true is a critical indicator of effective representation. Such “responsiveness” of policy to opinion does not mean that the public is getting what it wants, however. For this, we would need to assess the relationship between policy and the public’s preferred level of policy, P * , the latter of which can be difficult to capture. 2 Not surprisingly, few studies consider whether the public actually gets what it wants, though some research examines the “congruence” between preference majorities and policy adoption, as we will see.
A substantial body of research examines the relationship between the positions held by the public and elected officials. Much of this scholarship compares results of public opinion surveys with results of surveys of politicians. Other research compares public opinion surveys with the platforms or “manifestos” of political parties. Yet other studies consider actual roll call votes, particularly in systems with single member districts, where legislators represent geographic areas.
Early empirical work on the opinion–policy link was sparked in large part by Miller and Stokes’s (1963) “Constituency Influence in Congress.” These authors brought together data on public preferences by constituency, and both surveys and roll call voting behavior of U.S. members of Congress (MCs) on social welfare, foreign affairs, and civil rights. Correlations between constituency preferences and members of Congress’s behavior suggested that the latter was guided in part by constituency opinion. The finding was striking at the time, empirically demonstrating a mode of representation quite different from the party-centered work that had preceded it.
This seminal study spawned a vast literature seeking to establish links between the voting behavior of representatives and some combination of constituency opinion, constituency aggregate demographics, and representatives’ own demographic traits and party affiliations. The research is concentrated in the United States but has been extended to other countries with legislative districts (e.g., Converse & Pierce, 1986).
Referred to as studies of dyadic representation (Weissberg, 1978), the literature on roll call voting asserts that representation is to be found in the relationship between individual constituencies and individual representatives. A good amount of work bears out significant connections in the U.S. House of Representatives and Senate (e.g., Wright & Berkman, 1986). Recognizing important limitations in previous research relying on correlations, Bafumi and Herron (2010) attempt to directly match the ideal points of constituencies and representatives in an analysis of congressional roll calls. This research relies on broad ideological positions that encompass multiple issues (and other things), which has its own limitations, as Broockman (2016) illustrates in his fascinating analysis.
Of course, a positive relationship between opinion in districts and the roll call votes of their representatives does not mean that public opinion actually causes policy, as it could be that policymakers are reacting to something else, including the same things that determine public opinion. Although covariates may help identify the direction of causality in observational studies, there is no escaping the omitted variable(s) problem. Since is difficult to demonstrate causality in such studies, it is important that experimental work finds evidence of causal effects of opinion on representatives’ roll call voting behavior (Butler & Nickerson, 2011).
In countries without geographic constituencies, as in many of those that allocate legislative seats using proportional representation, scholars have examined “collective representation”—the match between public opinion and the positions of institutions, such as legislatures or executives. The first work of this type actually focused on the United States, where Weissberg (1978) proposed collective congressional representation as an alternative to dyadic representation. The relevance of collective representation even in a system that facilitates constituency-level representation is important: as Weissberg noted, concordance between individual legislators’ actions and constituency preferences is a helpful but not sufficient condition for policy representation; indeed, most individual representatives could vote against the majority opinion in their district and, so long as the various district preferences were reflected in the votes of other districts’ representatives, policy outcomes could still be representative of the (national) majority preference.
A good amount of work since has focused on other countries. Building on a decade’s work by a large number of scholars, Miller et al. (1999) examine various mass–elite linkages, including the representation of public opinion. A great deal of related research compares public opinion and the positions of parties, which Dalton et al. (2011) review and summarize. Powell (2000) assesses the match between the public’s ideological dispositions and those of governments, focusing especially on the effects of electoral systems. Budge et al. (2012) and Dalton et al. (2011) measure the positions of the public, political parties, and governments and assess their relationships, and find significant connections. While scholars have typically examined collective representation across space, a number of recent articles study dynamics and focus explicitly on the positions of the executive (e.g., Hakhverdian, 2010; Hobolt & Klemmensen, 2005; Jennings & John, 2009).
Research on representation in positions tells us something about the representation of opinion but tells us less about representation in actual policy. While positions and policies are related, after all, they are not the same things, and there is a growing body of work focused directly on policy. Scholarship in this area has tended to take one of five approaches: correspondence, consistency, covariation, dynamic representation, and congruence. These approaches are differentiated to a large extent by data availability, though, as we shall see, each has its advantages.
The policy correspondence approach to the study of representation is similar to most of the research on policy positions. Scholars in this tradition are interested in the correlation between expressed public preferences and policy across geographic areas, such as states or countries. The research asks: to what extent do levels of policy vary across areas alongside public preferences for policy? Erikson et al.’s (1993) Statehouse Democracy stands out as the classic example of this kind of research. These authors examine the relationship between estimated state ideology scores and a measure of state policy liberalism; results show quite a strong relationship between the two, and the work spawned a number of studies at the state level in the United States (see, e.g., Camobreco & Barnello, 2008; Carsey & Harden, 2010).
There is similar research on very local levels of government in the United States, including cities (Tausanovitch & Warshaw, 2014) and school districts (Berkman & Plutzer, 2005). Use of the approach outside the United States has been limited, though some scholars have assessed the relationship between opinion and policy across countries (e.g., Brooks & Manza, 2007).
We draw the policy “consistency” designation from Monroe (1998), whose work on the United States provides an archetypal example of this line of analysis. This research asks: to what extent is policy change consistent with a prior public preference for policy change? The approach involves identifying a single survey question asking about policy change and examining the relationship between the proportion of respondents favoring that change and the existence of proximate changes in policy. “Consistency,” then, refers to the match between public preferences for change and actual policy change, what Weissberg (1976) referred to as “majoritarian congruence.” Across more than 500 cases from 1981 to 1993, for instance, Monroe finds a consistency score of 55%.
Consistency scores can be estimated for separate policy domains or for different time periods. Indeed, this is where consistency scores are most interesting—they can indicate those domains (and times) in which opinion representation is particularly good (or bad). What consistency scores cannot do is establish a clear causal connection between public opinion and policy change. As Monroe (1998, p. 12) himself notes, the best this kind of analysis can do is to establish the coincidence of a public preference for change and actual policy change. A demonstration that preferences lead policy requires an analysis of data over time—data that can show, at least, that the public preference for change precedes the policy change.
The principal advantages of the consistency approach relate to the fact that it requires little data—indeed, each case requires just one survey result, and the capacity to assess whether there was a proximate change in policy in that domain. As a consequence, the approach can easily include a wide range of policy issues. Where overall policy responsiveness is concerned, the inclusion of as many policy domains as possible is critical. Polling questions deal with issues of some level of public salience, so estimated overall responsiveness will be based on a rather restricted set of policy domains. Moreover, because policy responsiveness is likely greatest for salient issues—as we shall discuss later—an estimate of overall policy responsiveness will almost necessarily be biased upward, relying as it does on only those salient issues about which pollsters ask questions (see Burstein, 2003). The consistency approach, by requiring just a single question, can encompass a broader spectrum of policy issues than the more data-intensive covariation or dynamic representation approaches described later.
Relatively light data requirements also mean that the consistency approach has been quite easily exported outside the United States to countries where comparatively fewer opinion data are available (e.g., Brooks, 1987; Petry, 1999). These studies compare preferences for policy change at a single point in time with actual policy change in a subsequent period—usually the next 12 months—and in so doing add much to our understanding of the opinion–policy link across countries. The ongoing interaction between opinion and policy is, however, more appropriately captured by other approaches, as we shall see.
Policy covariation studies involve a slightly more data-intensive approach to the link between opinion and policy. While consistency studies measure preference for policy change at a single point in time, covariation studies rely on cases in which the same policy question was asked at two different points in time. Changes in the distribution of responses over that period are compared with proximate policy change. Measures of policy also tend to be more comprehensive in this approach. Policy is typically examined both before and after the period of opinion change, so it is clearer when opinion precedes policy, or vice versa. The central question, then, is: to what extent do changes in policy follow related changes in public preferences for policy?
Studies of policy covariation go further than consistency studies in examining both opinion and policy over time and are thus better equipped to examine the causal order of opinion and policy change. The best-known and most comprehensive study of policy covariation is Page and Shapiro’s (1983) study of more than 300 federal U.S. policy issues from the mid-1930s to the late 1970s. These authors compare measures of covariation across domains and institutions, similar to Monroe, but with the additional advantage of being able to ascertain whether policy change followed or preceded opinion change. Indeed, a critical insight offered by this approach is that policy change often precedes measured opinion change.
The covariation approach has been used outside the United States as well (e.g., Bélanger & Pétry, 2005). The approach has much to recommend it: it is not so data-intensive as to be difficult to apply outside the United States, but at the same time it gathers enough information to get a general sense for the direction of causality between opinion and policy. Still, as with consistency, the limited period over which preferences and policies are measured makes it difficult to ascertain which came first. And preferences can change in part because of previous policy changes (see Wlezien, 1995). Such interaction over time is missed by the covariation approach but can be captured in the dynamic representation approach.
Where regular, frequent readings of opinion and policy are available, it is possible to analyze the time-serial relationship(s) between them—that is, with a sufficient number of cases, we can statistically assess the effect of opinion on policy over time. The main research question is: does public opinion reliably influence policy change? Given the need for reasonably long time series, there is a surprisingly long history of research in the area, particularly in the United States.
Early studies of dynamic representation preceded the development of the time-series econometrics that has come to characterize the field. In The Attentive Public , Devine’s (1970) analysis includes plots of (survey-based) mean policy support measures for different publics, alongside appropriations in those domains; meanwhile, Weissberg (1978) plots opinion measures alongside spending measures for 11 different U.S. policy domains. In each case, over-time analysis consists mainly of visual interpretations of graphs. Nevertheless, these authors’ broader longitudinal outlook makes their work the clear precursor to more recent research on dynamic representation.
The term “dynamic representation” is drawn from Stimson et al.’s (1995) article of the same title, a critical and representative example of what these analyses have come to look like. The article posits a model in which policy is a function of public preferences, either directly through politicians’ reactions to shifts in opinion, or indirectly through elections that result in shifts in the partisan composition of the legislature. The authors then examine relationships between a survey-based measure of “opinion liberalism” and policy voting measures for the president, House, Senate, and Supreme Court. There is strong evidence that policymakers respond to changes in public opinion.
Wlezien (1995) developed a “thermostatic” model of the (dynamic) reciprocal links between preferences and government spending—that is, a model that examined both opinion representation over time and public responsiveness to policy change. Dynamic models such as these seemingly are best equipped for investigating the causal relationships between opinion and policy. Not surprisingly, there is a growing body of such work, including analyses of defense spending by Eichenberg and Stoll (2003); analysis of immigration policy by Jennings (2009); work looking across a range of policy domains by Erikson et al. (2002) and Soroka and Wlezien (2010); and research across various countries by Hobolt and Klemmensen (2008), Wlezien and Soroka (2012), and Soroka and Wlezien (2015).
The drawback to dynamic models is they require a good deal of data, and to date this is available across many policy domains in a very limited number of countries: the United States, Canada, and the United Kingdom. Work on dynamic representation has thus been concentrated in polling-rich Anglo-American democracies, and mainly in salient policy domains. What the approach lacks in generalizability, however, it makes up for in the detail with which it can analyze opinion–policy relationships. This, we hope, will become clear in the sections that follow.
The research on policy discussed earlier is informative about responsiveness to public opinion but not about the actual match between public preferences and policy. Some recent work attempts to match the policy preferences of the public and policy decisions, and Lax and Phillips’s (2012) groundbreaking work stands out. 3 These scholars directly compare public support for specific policies in U.S. states and the corresponding state decisions on 39 specific policies in eight issue areas. Results reveal substantial responsiveness, but they also suggest that congruence between policy adoption and majoritarian public support is evident in only about half the cases. There is variation in congruence, and this reflects differences in issues and institutional context and is associated with issue salience, legislative professionalization, and term limits.
What Lax and Phillips have done is to be commended not only for what it shows but for the example it sets for scholars of policy representation; their approach has been adopted by others, including in other countries (Rasmussen et al., 2019). Even so, as Wlezien (2017) has shown, this work still cannot reveal whether the public is getting the policies it actually prefers. This requires more direct measures of the public’s preferred policies (e.g., the amount of defense spending it wants), which are difficult to come by in many areas. 4
Some scholars consider indirect ways to assess congruence, relying on relative preferences for policy —a preference for more or less policy. If people say that government spending is “about right,” they seemingly are happy with the status quo. By contrast, if they say the government is spending “too little” or “too much,” they appear to be indicating that they prefer policy change. 5 This is the assumption underlying work on majoritarian “consistency” mentioned earlier and provides the basis for Bartels’s (2015) assessment of the social welfare “deficit” and Ellis and Stimson’s (2012) characterization of “conflicted conservatives”—those who think of themselves as conservatives but support liberal spending policies. While an intriguing possibility, there are a number of problems with the assumption that the survey responses reflect congruence, the most basic of which may be question wording, which can matter a lot (see Wlezien, 2017). 6 Even assuming we have the right question wording, to assess congruence we need to match up preferences with spending, which is also difficult to get straight, as it depends on what the public has in mind. 7 Perhaps most importantly, using relative preferences to indicate congruence presumes that the public responds thermostatically to policy, taking into account both the level of policy it wants ( P * ) and the amount it gets ( P ) (Soroka & Wlezien, 2010; Wlezien, 1995). Otherwise, measured relative preferences really do not tell us anything about the public’s satisfaction with the policy status quo. Thus, while it may be tempting to interpret responses to relative preference items as telling whether the public is satisfied with current policy, doing so may conceal more than it reveals. 8 Responses nevertheless do provide useful information about the variation in public demand for spending over time, as much of the research on dynamic representation discussed earlier has made clear.
The congruence between preferences and policy clearly is an important scholarly concern, and there is a growing body of research on the subject. As we have seen, that research provides limited insight, and there is still work to be done on conceptualization and measurement. Thankfully, there has also been progress, particularly to improve the measurement of preferences (Barnes et al., forthcoming; Bonica, 2015; D’Attoma et al., 2018; Kolln & Wlezien, 2016; see also Buchanan et al., forthcoming).
Representation can occur in two familiar ways. The first way is indirect, through elections, where the public selects like-minded politicians who then deliver what it wants in policy. This is the traditional pathway to representation and is deeply rooted in the literature on responsible parties (Adams, 2001). In effect, the public chooses among alternative policy visions and then the winning parties put their programs into place after the election. The second way to representation is direct, where sitting politicians literally respond to what the public wants. This pathway reflects an active political class, one that endeavors to stay attuned to the ebb and flow of public opinion and adjust policy accordingly. The two ways to representation actually are related. That is, the first way implies the second, at least assuming incumbent politicians are interested in remaining in office or else are motivated to represent the public’s preferences for other reasons. This is how we think of representative democracy, how we think it should work—that is, we expect responsiveness. Responsiveness is dynamic—responsive politicians follow preferences as they change. Policy change is the result.
We can formally express these expectations by revising our equation 1 for policy ( P ) as follows:
where O still is opinion and I is introduced to represent partisan control of government. Here policy is conceived to be directly responsive to opinion and also indirectly responsive, through changes in partisan composition owing to elections. Of course, the indirect linkage presupposes a connection between public opinion and party control of government:
These models apply across both space and time. We can characterize the relationships between opinion and governments and policy across countries or, say, provinces or states within a country. There are relatively few studies across countries, as good comparative data are hard to come by, though scholarly explorations are underway. There is more research on the U.S. states, as we have seen, and Erikson et al.’s (1993) classic examination reveals both connections: general policy differences across states reflect the partisan composition of government and opinion, and the partisan composition reflects opinion.
We can also characterize relationships over time, as preferences change, following the study of dynamic representation. This sort of analysis allows us to more explicitly assess policy “responsiveness.” Erikson et al. (2002) do just this, focusing on the number of major pieces of legislation in the United States. They show that policy change nicely follows opinion over time independently of party control. Wlezien (2004) and Soroka and Wlezien (2010) show the same, focusing on different budgetary policy domains in the United States and also Canada and the United Kingdom. Pacheco (2013) and Caughey and Warshaw (2018) show much the same in the American states. These results do not mean that politicians actually respond to changing public preferences, as it may be that they and the public both respond to something else, for example, the perceived “need” for spending. All we can say for sure is that the research captures policy responsiveness in a statistical sense—whether and the extent to which public preferences directly influence policy change, other things being equal. There is also indirect representation owing to elections; in some studies, these effects actually exceed the more direct responsiveness (Bartle et al., 2019).
Of course, policy representation is an institutional outcome. In parliamentary systems, this is fairly straightforward—the government can change policy directly, assuming it does not face a realistic threat of a vote of (no) confidence. In presidential systems, agreement across institutions is usually required, as in the United States. Presidential responsiveness to public preferences is conceptually quite simple: the president represents a national constituency and is expected to follow national preferences. Congressional responsiveness is more complex, even putting aside bicameralism, as members of the legislature represent districts. Although preferences may differ across constituencies, there is reason to suppose that preferences in different constituencies move together over time, just as movement of opinion across states and various demographic subcategories of the American public (Page & Shapiro, 1992) is largely parallel. To the extent that they are responsive to public preferences, both the president and Congress should move in tandem, and predictable policy change is the logical consequence, possibly even in the presence of divided government. Here we have a good amount of evidence, as we have seen, though party polarization may weaken the tendency.
How exactly do politicians know what public preferences are? Elections likely provide some information, but direct representation between elections requires something further. Politicians may learn about preferences through interactions with constituents; they may just have a good intuition for public preferences (Fenno, 1978). Polls likely also play a critical role. We know that policymakers’ use and interpretation of polls can vary (e.g., Kingdon, 1995); at the same time, particularly given developments in polling technology, policymakers have relatively easy access to public opinion on policy matters and there is considerable evidence of the importance of polls, both public and private, in policymaking (e.g., Geer, 1996). This work is critical: it shows one means by which politicians learn about public preferences. Of course, politicians have other, more direct sources of information as well, and these may be of greater importance to them. 9
Representation does not occur in all policy domains in all countries. The characteristics of domains appear to matter, for instance. Representation is likely to reflect the political importance (or “salience”) of issues, if only due to the possible electoral consequences. Let us briefly trace the logic.
In its simplest sense, a salient issue is politically important to the public. People care about the issue and have meaningful opinions that structure party support and candidate evaluation. Candidates are likely to take positions on the issue and it is likely to form the subject of political debate. People are more likely to pay attention to politicians’ behavior on an important issue, as reflected in news media reporting or as communicated in other ways. Politicians, meanwhile, are likely to pay attention to public opinion on the issue—it is in their self-interest to do so. There are many different and clear expressions of this conception of importance. In issue domains that are not important, conversely, people are not likely to pay attention to politicians’ behavior, and politicians are by implication expected to pay less attention to public opinion in these areas. This reflects a now classic perspective (see, e.g., Geer, 1996; Moniz & Wlezien, 2020). 10
This not only implies variation in representation across issues, which Lax and Phillips (2012) found in their analysis of representation in the American states; it also implies variation in responsiveness within domains over time, as salience changes. When an issue is not very salient to the public, politicians are expected to be less responsive. As salience increases, however, the relationship should increase. That is, to the extent that salience varies over time, the relationship between opinion and policy itself may vary. Though the expectation is clear, there is little research on the subject. We simply do not know whether representation varies much over time, and we still do not know much about the variation in issue importance (see Wlezien, 2005; see also Moniz & Wlezien, 2020). 11
Public preferences in the different policy domains are not entirely unique—they tend to move together over time. This patterned movement in preferences is well documented in the United States (Erikson et al., 2002; Wlezien, 1995) but is also evident elsewhere, in the United Kingdom and Canada (Bartle et al., 2011; Soroka & Wlezien, 2010). The pattern has led some scholars to conclude that the public does not have preferences for policy in different areas but rather a single, very general preference for government activity (e.g., Stimson et al., 1995). From this perspective, measured preferences in various domains largely represent (multiple) indicators of a single, underlying preference for government action. When compared with the more traditional perspective, this characterization of public opinion implies a very different, global pattern of representation.
Some research shows that, although preferences in different areas do move together over time, the movement is not entirely common (Wlezien, 2004). Preferences in some domains share little in common with preferences in others; these preferences often move quite independently over time. In short, the work indicates that preferences are some combination of the global and the specific—moving together to some degree but exhibiting some independent variation as well. This research also shows that policymakers reflect the specific variation, at least in some policy domains (Wlezien, 2004; see also Druckman & Jacobs, 2006). Not surprisingly, these domains tend to be highly salient to voters, ones on which they pay attention to what policymakers do. In other less salient domains, policy only follows the general global signal. In yet other, very low salience domains, policy seemingly does not follow preferences at all.
Polities differ in many ways, and some of these differences should have significant implications for the nature and degree of representation. Of fundamental importance are media openness and political competition. Without some degree of media openness, people cannot easily receive information about what government actors do and thus cannot effectively hold politicians accountable for their actions. A small but growing literature considers the role of media in representation (e.g., Hiaeshutter-Rice et al., 2021; Neuner et al., 2019; Williams & Schoonvelde, 2018).
Even where we have essential levels of media and political competition, institutional differences may have important implications for policy representation. Here we have a growing body of empirical work, particularly on electoral systems.
Most of this research focuses on the differences between the majoritarian and proportional visions, using Powell’s (2000) language, and mostly on how these differences matter for policy representation. Lijphart (1984) provides the first direct statement on the matter. He distinguishes between “consensual” democracies—characterized by, most notably, proportional representation, multiparty systems, and coalition governments—and “majoritarian” systems—characterized by simple plurality election rules, a two-party system, and single-party government. Most importantly, Lijphart suggests that consensual democracies provide better descriptive representation and general policy congruence than do majoritarian systems.
Powell (2000) provides further empirical support, focusing specifically on the differences between majoritarian and proportional election rules and their implications for representation. Powell finds that proportional representation tends to produce greater congruence between the government and the public; specifically, the general ideological disposition of government and the ideological bent of the electorate tend to match up better in proportional systems. According to Powell, this reflects the greater, direct participation of constituencies the vision affords in governing coalitions (see also Miller et al., 1999).
Powell’s results pertain to elections and their immediate consequences. But what about in the periods between elections? Are coalition governments more responsive to ongoing changes in opinion? Although proportional systems may provide more indirect representation, it is not clear that they afford greater direct representation. There is reason to think that governments in majoritarian systems actually are more responsive to opinion change. First, presumably it is easier for a single party to respond to changes than a multiparty coalition, as coordination in the latter is more difficult and costly. Second, majoritarian governments may have more of an incentive to respond to opinion change. Since a shift in electoral sentiment has bigger consequences on election day in majoritarian systems, governments there are likely to pay especially close attention to the ebb and flow of opinion. Thus, it may be that the two systems both work to serve representation, but in different ways, where proportional systems provide better indirect representation via elections and majoritarian systems better direct representation in between elections. Basic empirical work (Soroka & Wlezien, 2015; Wlezien & Soroka, 2012) supports the expectation.
Just as electoral systems may matter, so too may government institutions. In particular, research suggests that the horizontal division of powers may structure the relationships between opinion and policy over time. 12 The concentration of powers in parliamentary systems—as opposed to presidential systems—affords voters more direct control over government on election day. This presumably aids indirect representation: to the extent election outcomes reflect public opinion, then policy representation will follow quite naturally, at least to the extent we have responsible parties.
The same seemingly is not true of direct representation, and there is reason to suppose that parliamentary governments are less reliable in their attendance to public opinion over time. Scholars have long noted the dominance of cabinets over parliaments (see, e.g., Cox, 1987). Scholars portray a world in which governments exercise substantial discretion, where the cabinet is the proposer, putting legislation to a legislature that ultimately has only a limited check on what the government does. Strom (2003) concludes that parliamentary government deals much better with selecting representatives—avoiding what political scientists refer to as “adverse selection”—than it does with ensuring that they actually do the right thing—avoiding what political scientists call “moral hazard.” Put simply, cabinet governments are difficult to control on a recurring basis.
This has fairly direct implications for government responsiveness. When there are differences between what the cabinet and parliament want, the latter cannot effectively impose its own contrary will. The process of amendment and veto is compromised, at least by comparison with presidential systems, especially “Madisonian” ones in which executive and legislative powers are balanced. In the latter the executive cannot effectively act without the legislature, at least with respect to statute. The legislature is the proposer—it puts statute to the executive—and while the executive can veto legislation, the legislature can typically override the veto. Most changes in policy require agreement between the executive and legislature, or else a supermajority in the latter. This can help reduce disjunctures between public opinion and policy change.
Although the separation of powers makes presidential systems more deliberate in their actions, therefore, it may also make them more reliably responsive to public opinion over time. 13 We still expect representation in parliamentary systems, of course—after all, governments in these systems are more easily held accountable for their actions, as responsibility is far clearer, particularly in a majoritarian context. In between elections, however, there is little to make parliamentary cabinets accountable except for the prospect of a future electoral competition. Though important, the incentive is imperfect, and comparative research bears out these expectations (Wlezien & Soroka, 2012). Recent work by Rasmussen et al. (2019) considers the competing influences of government and electoral institutions.
We make regular reference to “public opinion” and “public preferences.” But what exactly is the public? Is it the collection of all of us, with each person’s preferences given equal weight? Or is it a more narrowly drawn public, including some people’s preferences but not others? Who gets what they want in policy?
In one conception, the public consists of all citizens, or at least all adults. Citizens are all, more or less, equally entitled to vote, and each person has but one vote. Perhaps then we should all have equal weight where policymaking is concerned. This is an ideal, the stuff of civics textbooks; in reality, however, there is good reason to think that preferences are not equal and that some people’s preferences are more important than others. In particular, we might expect politicians to pay special attention to the preferences of active voters. These are the people who matter on election day, after all—the ones who put (and keep) politicians in office.
The representation of voters rather than citizens would not matter much if voters were a random sample of citizens. But we know that there are differences between the voting and nonvoting publics: voters tend to be better educated, have better jobs, and have higher incomes. Not surprisingly, voters tend to be more conservative than their nonvoting counterparts, at least on some economic issues. If politicians are more attentive to this group, and follow the median voter, then policy will be more conservative than the median citizen would like. This is of obvious importance. We still know relatively little empirically, however, though scholarly interest is on the rise: Griffin and Newman (2005) reveal that U.S. politicians pay more attention to the opinions of voters than to those of nonvoters, and recent work by Dassonneville et al. (2021) suggests similar findings across OECD countries.
Much research focuses on differences relating to income. Bartels’s (2008) and Gilens’s (2012) early research on the subject suggests that U.S. politicians are most attuned to the opinions of high-income citizens, and some of the research that followed is confirming (e.g., Gilens & Page, 2014; Peters & Ensink, 2015). The results of other research are more mixed (Elkjær & Iversen, 2020; Rigby & Wright, 2011, 2013; Wlezien & Soroka, 2011). Analysis here is complicated by the striking similarity in preferences across income groups (Soroka & Wlezien, 2008, 2010), which has led some scholars to consider who wins when preference majorities differ (Bashir, 2015; Branham et al., 2017; Enns, 2015). Interestingly, that research shows little difference in the success rates of different groups, though the rich appear to do better than we would expect were politicians relying on the preferences of the median voter. For more on this, see Enns and Wlezien’s early (2011) volume on the subject and Elkjær and Klitgaard’s (forthcoming) more recent review and assessment of this literature.
Of course, there may be related sociodemographic manifestations, across race for example, that scholars are only beginning to examine. Political equality also may have explicitly partisan expressions. It may be, for instance, that politicians are more responsive to in-partisans, as Hill and Hurley (2003) and Soroka and Wlezien (2008) have suggested. Of course, effective responsiveness to public preferences depends on the accuracy of politicians’ perceptions of public attitudes, and it may be that misperceptions help produce inequality. For example, Sevenans et al. (forthcoming) argue that politicians may respond to a perception of public preferences that is biased toward educated and politically interested citizens. This is part of a growing literature on the ways in which biased perceptions of preferences may produce inequalities in representation, or mute representation of the average citizen (see especially Broockman & Skovron, 2018). Of course, it may simply be that politicians tend to see things much as the more highly educated and higher income citizens do, partly because they share those characteristics.
This and the other work on inequality in representation is important, but it only scratches the surface. We need to know more about the breadth and depth of the inequality, both at particular points in time and over time, so much work clearly remains to be done.
We have thus far concentrated on policy representation—the effect of public opinion on public policy. But policy representation ultimately requires that the public notices and responds to what policymakers do. Without such responsiveness, policymakers would have little incentive to represent what the public wants in policy—there would be no real benefit for doing so, and there would be no real cost for not doing so. Moreover, expressed preferences would be of little use even to those politicians motivated to represent the public for other reasons.
Despite ongoing concerns about the ignorance and irrationality of the average citizen, a growing body of recent work shows that the average citizen may be more informed than initially assumed. This is not to say that the average citizen knows very much about politics, but there is accumulating evidence that individuals may be capable of basic, rational political judgments. Moreover, even in the face of individual ignorance, aggregate preferences often react sensibly to real-world trends (Page & Shapiro, 1992). The public reacts to both real-world affairs and policy itself, much like a thermostat (Wlezien, 1995); that is, the public adjusts its preferences for “more” or “less” policy in response to policy change, favoring less (more) policy in the wake of policy increases (decreases), ceteris paribus. This conceptualization fits nicely with classic functionalist models of the policy process, where policy outputs feed back on public inputs into the policymaking process.
Empirical analysis shows that public responsiveness, like policy representation, varies across policy domains and political institutions (Soroka & Wlezien, 2010). That representation is likely to be greater in salient domains is largely the product of representatives reacting in domains in which publics themselves are monitoring and reacting to policy change, for instance. Salient domains are characterized by a higher degree of both representation and responsiveness; more precisely, public responsiveness and policy representation covary. This is not equally true across contexts, however. Fundamental to public responsiveness is the acquisition of accurate information about what policymakers are doing, and so responsiveness will be lower when the acquisition of information is more difficult. For instance, federalism, by increasing the number of different governments making policy and thus making less clear what each level of government is doing, may decrease responsiveness and representation. 14 The horizontal division of powers may also be important, though here our expectations are less clear. Regardless, where information is easier to acquire, public responsiveness—and by implication policy representation—should be greater.
Ultimately, we expect variation across domains and institutions in both policy representation and public responsiveness. Yet the existence of each connection between opinion and policy—indeed, the existence of both connections—is critical to the functioning of representative democracy. Insofar as research seeks to understand what public preferences are and how they are formed, then, it can be viewed as an examination of the potential for, or success of, representative democratic institutions. The literature makes a contribution to our understanding of one of the most significant and enduring questions in the study of politics: does representative democracy work? It does not work perfectly, to be sure, but in some cases it appears to work better than many of us might expect.
We thank the editor, Russell Dalton, for helpful comments, and also Robert Shapiro for his input in the past.