Issue Ownership and Representation in the United States: A theory of legislative response to constituency opinion. Abstract

Transcription

1 Issue Ownership and epresentation in the United States: A theory of legislative response to constituency opinion Patrick J. Egan pategan@princeton.edu Visiting Scholar (006-07) Center for the Study of emocratic Politics Woodrow Wilson School of Public and International Affairs Princeton University Ph.. Candidate Charles and Louise Travers epartment of Political Science University of California, Berkeley this version: October 006 Abstract Issue ownership (Budge and Farlie 1983, Petrocik 1996) the extent to which the public trusts one party to better handle a particular issue than the opposing party is a persistent, important phenomenon in American politics. But formal models have failed to incorporate this notion in their analyses of the platforms candidates take in two-party electoral competition. By adding issue ownership to standard models of position-taking by policy-oriented candidates, I generate the prediction that candidates will take divergent, asymmetric positions in the policy space. Specifically, the candidate from the issue-owning party will leverage her advantage to take a position close to her own ideal point, while the other candidate s position will be highly responsive to voters preferences. These predictions are confirmed by examining the relationship between constituency opinion from U.S. Congressional districts (generated from the 000 National Annenberg Election Survey) and key roll call votes cast by Congress from 1997 through 00. The positions taken by emocrats on emocratic-owned issues such as education and the environment are much less sensitive to district opinion than those taken by epublicans. Similarly, epublican positions on issues like defense and crime are less responsive to public opinion than those taken by emocrats. Thanks to James Adams, Larry Bartels, Henry Brady, Bruce Cain, osalee Clawson, ui de Figueiredo, Patricia Hurley, Megan Mullin, Abe Newman, Ben Page, obert Powell, Laura Stoker, ob Van Houweling and participants in the American Politics Colloquium and the Positive Political Theory Seminar at UC Berkeley for comments and suggestions regarding this paper.

2 I. Introduction The notion that elected leaders represent citizens interests is at the core of our understanding of the democratic process. If, with ahl (1971), we define democracy as popular control of government by political equals, we naturally seek to examine the nature and extent of that control. o the decisions of elected officials correspond to the public s preferences? o elected officials change their actions when public preferences change? This paper concludes that members of the United States Congress generally reflect their constituents' preferences in their roll-call votes, but members of Congress can and do deviate from the public's preferences when the public believes that the member's political party "owns" an issue. In the United States, epublicans are generally thought to own national security issues such as terrorism and emocrats are thought to own issues such as the environment and social security. Issue ownership makes it possible to explain observed variations in the responsiveness of politicians across issue domains. It also helps us reach a more nuanced understanding of the distinction drawn by Stokes (1963) between position and valence issues. In addition, it requires that we rethink two broad streams of literature: that on formal models of candidates in electoral competition and that on the relationship between citizen opinion and government action. Formal models have been employed to generate predictions about the relationship between policy platforms adopted by candidates in democratic electoral competition and voters preferences. In general, this work has focused on extensions and refinements of the simple median voter model variously attributed to Hotelling (199), Black (1948) and owns (1957). A second stream of research consists primarily of empirical studies that seek to measure the relationship between citizen opinion and government action. This literature on responsiveness and representation has its origins in the pioneering work of scholars such as Miller and Stokes (1963), Achen (1977, 1978), and Page and Shapiro (1983). 1

3 This paper contributes to both streams of literature, and it addresses some of their shortcomings. On the one hand, critics of the formal models of party competition complain that the formal theories are rarely subjected to empirical tests and that the central prediction of the classic ownsian spatial model that candidates platforms should converge to that of the median voter in two-party electoral competition is seen infrequently in politics (see e.g., Osborne 1995, oemer 001, Grofman 004, Adams, Merrill and Grofman 005). Furthermore, more recent models that predict separation based on the possibility of third-candidate entry, uncertainty about the location of voters, policy-motivated politicians, and other mechanisms do not provide very clear predictions beyond the mere possibility of separation of candidate platforms. On the other hand, recent reviews of the empirical responsiveness literature have found it lacking in theoretical grounding. In particular, scholars have failed to generate strong theories that explain the observed variation in responsiveness across issue domains (Manza and Cook 00, Burstein 003) I address these critiques by incorporating the notion of issue ownership in a model of two-party electoral competition. Issue ownership is a phenomenon seen across representative democracies in which the electorate tends to associate political parties with long-term strengths and weaknesses on various issues. A political party owns an issue to the extent that the public assesses it can better handle the issue than its opponents. Budge and Farlie (1983) identified that the public s assessments of these strengths and weaknesses were important determinants of election outcomes. Petrocik (1996) extended this finding to the American context. He discovered that the emocratic and epublican parties make labored efforts to emphasize their owned issues in presidential campaigns, and that they perform better to the extent that they succeed in doing so. Issue ownership can be measured empirically by a commonly asked question on public opinion surveys: Which party do you trust to a better job handling [issue x]? In the United States, emocrats are generally trusted more than epublicans on issues such as the environment and education; epublicans are trusted more on foreign policy and taxation.

4 The concept of issue ownership makes possible a more nuanced understanding of the commonly-cited distinction made by onald Stokes between position issues and valence issues in his 1963 critique of spatial models. Position issues, wrote Stokes, are those that involve advocacy of government actions from a set of alternatives over which a distribution of voter preferences is defined (373). By contrast, valence issues merely involve the linking of parties with some condition that is positively or negatively valued by the electorate (373). Stokes argued that spatial models of his day, which excluded valence issues, failed to capture important aspects of party competition. ecent formal models have addressed this shortcoming. Generally, these models build in the assumption that some parties or candidates get a valence benefit because they are viewed as more competent, charismatic or have some other advantage (e.g., Ansolabehere and Snyder 000, Groseclose 001, Schofield 003). One problem with many of these models is that simply posit that one candidate has a valence advantage over another without providing a particular rationale for that advantage and without providing any sense of the possible limitations of that advantage. 1 These models also neglect a final observation by Stokes regarding position and valence issues: that their distinction is not always clear. It is of course true that position-issues lurk behind many valence-issues. (373). Many valence debates (such that over President Bush s handling of the war in Iraq) are affected by discussions of ordered sets of alternatives (such as whether to withdraw forces from Iraq or commit more troops). Issue ownership provides an intuitive explanation for why a candidate or party would have a valence advantage by linking this advantage with candidates ability to handle specific policies. In this paper, I conceptualize issue ownership as follows: Candidates policy platforms are usually imperfect predictors of the outcomes they will actually deliver to voters should they be elected to office. Unforeseen events often arise that require officeholders to calibrate the policies upon which they campaigned. The policies enacted by elected officials interact with these events to produce the outcomes 1 An exception to this is one version of Schofield s model, which assumes that candidates accrue a valence advantage to the extent that they take positions favored by party activists who in return provide the candidates with money and support. 3

5 ultimately experienced by the public. Issue ownership, therefore, is the extent to which a party s elected leaders are better able than the opposing party to respond to events in an issue domain and deliver the outcome promised during the campaign. I incorporate these ideas in a model in which candidates from two parties one of which is considered by voters to be the issue-owning party compete for a seat in a national legislature. In the model, the candidate from the issue-owning party has the competency to observe events perfectly in the issue domain. Thus if elected, she can respond with policy choices that deliver precise outcomes. The opposing candidate only knows the probability distribution of these events, and thus if elected must respond to such events imprecisely. The model generates the prediction that the competency advantage associated with issue ownership leads some voters to support the issue-owning candidate, even if they prefer the outcomes promised by the opposing candidate. This in turn allows the issue-owning candidate to take a position that is unresponsive to the preferences of the typical voter. By contrast, the opposing candidate compensates for her lack of competency on the issue by making promises that closely track the preferences of the median voter. To the extent that an issue-ownership advantage exists, the parties positions diverge. In this paper, I briefly review the formal literature regarding models of two-party electoral competition and the empirical literature on representation of constituency opinion across policy domains. I further develop the notion of issue ownership and provide data indicating how issue ownership works in American politics. I then present a model that helps us understand how issue ownership affects representation of constituency opinion by incumbent legislators. Finally, I present analyses of Congressional roll call votes and public opinion data from the 000 National Annenberg Election Survey that comport with the empirical implications of the theory. This paper leaves aside the question of the origin of issue ownership. It focuses instead on the nature of issue ownership and its impact on representation. 4

6 II. Formal models of electoral competition and representation Formal models of electoral competition are a broad family of analyses that make predictions about the policy platforms adopted by office-seeking candidates in representative democracies. The point of departure for this work is generally the model proposed by owns (1957), who found that simple assumptions about two-party electoral competition yielded the powerful prediction that candidate platforms should converge to the preferences of the voter located at the median of the distribution of all voters preferences. The assumptions required to obtain this result are over the policy space (that it is one-dimensional), the electorate (that each citizen has single-peaked preferences in the policy space and that all citizens vote for the candidate located closer to them in the space), and the candidates (that they are motivated solely by winning office). As is well known, owns work was inspired by the models of firm location developed by Hotelling (199) and of voting in committees by Black (1948). The simplicity of owns model and the intuitiveness of his assumptions have led to the enduring appeal of the Median Voter Theorem (MVT). It is a critical component of many contemporary models of politics, including legislative activity (e.g. Krehbiel 1998), economic policy (Meltzer and ichard 1981, Persson and Tabellini 000) and political transitions (Acemoglu and obinson 005). But its central prediction does not square with the fact that the platforms of political parties differ sharply from one another in most representative democracies. Much of the formal work on electoral competition following owns has therefore focused on determining the changes to the MVT necessary to achieve the divergence of candidate platforms in equilibrium. Two important additions to the MVT that candidates care about policymaking (Wittman 1973), and that candidates do not know the preferences of the median voter perfectly (Coughlin 199) together generate the prediction that candidate platforms will diverge in ways similar to what is observed in actual campaigns (oemer 001). Further innovations for example, the possibility of entry by a third party (Palfrey 1984), allowing for abstention by disaffected voters (Adams and Merrill 003), and assuming 5

7 that a party s platform is the result of bargaining among factions that make up its coalition (oemer 001) have also generated predictions of platform divergence. An additional line of research explains platform divergence with so-called nonpolicy factors. Scholars working in this vein begin with the notion that policy platforms are only one of many criteria voters use to evaluate candidates. A candidate may be viewed as more experienced, charismatic, or credible than her opponents and may thus enjoy an electoral advantage. Two recent papers have incorporated this valence advantage in models of electoral competition, and generate the counterintuitive result that the valence-advantaged candidate takes a relatively moderate position in equilibrium. A model developed by Ansolabehere and Snyder (000) features candidates who are purely office-seeking and have perfect knowledge of voters preferences. It yields the prediction that the valence-advantaged candidate always wins the election and generally takes a moderate position in the policy space. By contrast, Groseclose (001) proposes a model in which candidates are motivated by both policy and officeholding and have imperfect knowledge of voters policy preferences. He finds that, under certain assumptions about the shape of voters utility functions, small increases in valence advantage lead the advantaged candidate to take more moderate positions. The intuition behind these unexpected results is that the valence-advantaged candidate seeks to highlight the differences between herself and her disadvantaged opponent. To the extent that she takes policy positions that are similar to her opponent, she leads voters to choose primarily on the valence criterion. In both models, valence enters additively into the voters utility functions: voters realize a sort of bump if the valence-advantaged candidate wins. By incorporating the notion of issue ownership, the model outlined here takes a different approach to motivating valence advantage: it ties such an advantage directly to candidates ability to deliver policy outcomes on a particular issue. In doing so, the model links valence and position together. In this conceptualization, any valence advantage is issue-specific: a candidate who is considered better able to handle environmental issues, for example, is not necessarily similarly advantaged on (say) 6

8 foreign policy issues. As we will see, the issue-specificity of this advantage combined with the assumption that candidates care about policymaking as well as holding office yields the prediction that the position of the issue-owning candidate is less responsive to public opinion than that of the opposing candidate. This is a result that contrasts sharply with the predictions of the two papers discussed above. III. Empirical analyses of responsiveness and representation Empirical scholars have found varying degrees of responsiveness to public opinion across time, jurisdictions, and policies. In their pioneering work examining the relationship between the policy attitudes of survey respondents and their congressional representatives, Miller and Stokes (1963) found that a significant relationship existed in the area of civil rights, but not so in social welfare or foreign policy. Subsequent work pointing out methodological problems in this research has revised these findings, and demonstrated less variation in responsiveness across issue domains (Achen 1978, Erikson 1978). Page and Shapiro (1983) find that responsiveness is higher on domestic issues than foreign policy issues, and argue that this is because domestic issues tend to be more salient with the mass public. In a recent valuable contribution, Wlezien (004) finds strong relationships between public preferences and budget appropriations on defense, welfare, and health programs and significant, but weaker, relationships on education and environmental protection programs. Additional research on this topic has largely been conducted on an issue-to-issue basis, making comparisons between issues very difficult. Perhaps surprisingly given Page and Shapiro s result, defense issues have been found exhibit high degrees of responsiveness. Bartels (1991) uses National Election Studies data and roll-call votes cast on the defense budget by House members to estimate that constituency demand for increased defense spending accounted for about 10 percent of the total 198 fiscal year appropriation for the Pentagon. Additional studies have found relationships of varying degrees on environmental policy, taxation and health (for a review, see Burstein 003). 7

9 Another line of research in this field Lowi (1963), Wilson (1974), and Arnold (1990) concludes that the structure of policies and the organized interests supporting various policy outcomes should affect the representation of public opinion by elected legislatures. For example, Arnold argues that diffuse, unorganized interests can be victorious over concentrated, special interests when costs or benefits can be easily recognized by the public. This hypothesis corresponds somewhat with Carmines and Stimson s (1980) notion of easy versus hard issues easy issues being those that are more symbolic than technical, deal with policy ends rather than means, and have long been on the political agenda. We might expect a stronger relationship between opinion and policymaking on easy issues. Perhaps the biggest gap in the research is that we still do not have empirically tested hypotheses that identify the policy domains on which we should expect more or less responsiveness by elected officials to public opinion. Should we expect politicians to care more about public opinion regarding tax policy or the environment? efense spending or trade policy? Education or foreign policy? Political scientists have, at best, educated guesses as to the answers to these questions and in fact, much of the most highly regarded research in the field has yielded conflicting findings. In sum, the research offers support for the idea that government activity bears some relationship with public opinion regarding policy issues. But empirical work that offers us the opportunity to compare the strength of this relationship across issues is scarce. If government responds to public opinion to varying degrees on different policies, this phenomenon has yet to be decisively demonstrated by political scientists (Manza and Cook 00 and Burstein 003). IV. Conceptualizing issue ownership The idea of issue ownership provides a promising place from which to better explore variation in the character of representation across policy domains. In the framework described by Petrocik (1996), parties each own a set of issues on which the public views the party as substantially better able to handle than the other party. In 8

10 Petrocik s conceptualization, handling is the ability to resolve a problem of concern to voters. It is a reputation for policy and program interests, produced by a history of attention, initiative and innovation toward these problems, which leads voters to believe that one of the parties (and its candidates) is more sincere and committed to doing something about them. (86) Petrocik notes that on a wide range of issues, survey respondents said they trusted either the epublicans or the emocrats to handle those issues more than the other party, and sometimes by substantial margins. Figure 1 provides an example of recent data on issue ownership. It is a bar chart indicating results from an ABC/ Washington Post poll conducted in 00. Because partisan respondents are highly likely to identify their own party as the trusted party on any given issue, Figure 1 includes only self-identified independents to better illustrate variation across issues. Figure 1. Trust in Parties' Ability to Handle Policy Issues (among self-identified Independents only) source: ABC News/Washington Post Poll, September 00 war on terrorism Iraq and Saddam Hussein taxes federal budget economy and jobs financial fraud by corporations social security prescription drug benefits for elderly health care education environment % trusting emocrats % trusting epublicans 9

11 The data displayed in Figure 1 are generally in line with the expectations of observers of American politics: voters appear to trust epublicans more on foreign policy issues and taxes, and emocrats on domestic issues. On newly-emerging issues that don t fit pre-existing frameworks (such as the corporate financing scandals that had just begun to break in 00) there is little differentiation among the parties. An important note is that the relative ranking of issues owned by each party has remained steady over time that is, the emocrats and epublicans strongest issues in the 1980s and 1990s are their strongest issues today. The data in Figure 1 are from 00, when the political parties were held in approximately equal esteem by the American public. When one party falls into general disfavor (as is currently the case with the epublican Party), it can find itself at a disadvantage on most issues. But the relative ranking of issues over time has been stable since issue-ownership questions were first included on public opinion surveys in the 1980s. For example, an April 006 poll by the Winston Group found that while emocrats were more trusted than epublicans on most issues, the epublicans two best issues remained border security and the war in Iraq. 3 Given the strong, consistent nature of issue ownership and its evident importance to election campaigns and outcomes, a natural hypothesis emerges about the influence of issue ownership on representation: because voters trust parties more on issues they own, incumbents from the owning party should have more latitude in deciding how to vote on roll-calls on the issue that come before them in Congress. After a brief discussion of the nature of issue ownership and issue trust, I proceed to formalize and test this hypothesis. 3 The origins and dynamics of parties issue ownership advantages are beyond the scope of this paper. But we might intuitively locate the basis for issue ownership in the coalitions that contribute to parties history of attention, initiative and innovation toward a particular issue domain. These coalitions are undoubtedly sources of policy expertise for issue-owning parties. They provide new ideas, research, and people who can staff the parties key policymaking positions on the issue domain. These resources allow issue-owning parties to better respond to events in order to achieve the outcomes they want. 10

12 V. The nature of issue ownership and issue trust As described above, issue ownership is defined as the extent to which the public trusts a party to handle a particular issue. (To distinguish this issue-based notion of trust from the generalized idea of trust in government, I will often refer to the trust accompanying issue ownership as issue trust. ) Issue ownership has several components. It is a long-term phenomenon that is established over decades of position taking and accomplishments by political parties. Changes in the degree of issue trust assigned to one party or another are rare. Perhaps the most striking example of such a change is the issue evolution that occurred on race in the 1960s as documented by Carmines and Stimson (1990). Their findings can be reconceptualized in the issueownership framework as a shift in which party owned race issues among whites and blacks. The shift was precipitated by the dramatic change in the positions taken by the two parties on race. Until just prior to the Civil ights Act of 1964, emocrats had the trust of whites on the issue of race and epublicans had the trust of blacks. The reverse was true by the close of the 1960s. But such dramatic changes in the ownership of issues are unusual. emocrats have arguably owned Social Security since the program s creation by the oosevelt administration, and have owned health care since the inception of Medicare and Medicaid during the Johnson administration. epublicans ownership of the issue of taxation stretches back at least to the tax revolts of the 1970s, and the G.O.P. has owned the crime issue from the law-and-order campaigns of ichard Nixon onward. Issue trust has remained steady despite actions taken on issues by the owning parties that were clearly not desired by a majority of the public, including the disastrous Clinton healthcare reform plan of and the Iran-Contra scandal that clouded the epublicans foreign policy reputation in the late 1980s. Another characteristic of issue trust is that it is often correlated with the extent to which voters agree with the party s position on an issue. In practice, this means that a majority of the public tends to agree with the positions taken by a party on the issues it 11

13 owns. Majorities of Americans not only trust epublicans over emocrats on taxation and defense; they also tend to share the epublicans preferences for lower tax burdens and a hawkish foreign policy. The reverse is true for the emocratic-owned issues of education and healthcare. Table 1 displays two additional examples: on the emocraticowned issue of the environment, a strong majority of Americans sides with the party on whether federal efforts should be expanded in this area. On the epublican-owned issue of crime, a substantial majority favors the death penalty for some crimes, a position taken much more frequently by epublican candidates than emocrats. Table 1. Americans Policy Preferences on Issues Owned by the Two Parties Amount of federal government effort to protect the environment eath penalty sentences for some crimes more 66.6 favor 74.7 same, less, or none 33.4 oppose 5.3 source: 000 National Annenberg Election Study. However, issue ownership is a notion that is broader than simply one of shared policy preferences between voters and parties. It also incorporates voters assessment of the competency of parties to solve particular policy problems. A voter may be a dove on defense issues and thus agree with emocrats on those policies, but he may ultimately believe that epublicans are better at defending the country from a terrorist attack. Another voter may prefer the free-market approach taken to environmental regulation by epublicans, but thinks the emocrats will actually do a better job on ensuring the cleanup of toxic waste sites. Survey data that allow us to assess whether issue ownership goes beyond shared policy preferences between voters and parties are scarce, because issue-ownership questions are rarely asked in concert with policy preference questions in the same public opinion survey. The American National Election Studies did so in 199, 1994, and 1998 on four different issue domains, and multivariate analysis of these survey data is 1

14 presented in Table. If issue-ownership is nothing more than policy congruence between voters and parties, then we would expect that it would be completely explained by this congruence and party identification. In Table, participants responses to issueownership questions were regressed on measures of their policy placements of the parties as well as their party identification. All variables were scaled -.5 to.5 (emocratic to epublican), with zero indicating a neutral position between the two parties. The intercept in each regression is thus the estimated issue-ownership advantage accorded to the epublican Party (or, if negative, the emocratic Party) that cannot be explained away by party identification or policy preferences. As shown in the table, the intercepts are of the expected sign and statistically significant. on each of the four issues analyzed: epublicans are advantaged on the issues of foreign affairs and crime, emocrats on healthcare and the environment. 4 On each issue, the advantage is of roughly the same absolute magnitude (.09 to.18): that is, the advantaged party enjoys a 9-to-18-point ownership advantage among Independents whose policy preferences are equally distant from those of both parties. 4 The dependent variable takes on three possible values (trust emocrats, trust both parties equally, trust epublicans) and could arguably be considered an ordinal, rather than interval level, measure. Analyses of the data using an ordered probit model produced substantively similar but less easily interpretable results than the OLS analyses presented here. 13

15 Table. eterminants of Issue Ownership on Four Issues, defense/ foreign affairs (199) crime (1994) health care (1994) environment (1998) party to which feels closer on issue (.07) (.0) (.05) (.07) s party identification (.03) (.03) (.03) (.04) epublican issue-ownership advantage (intercept) (.01) (.01) (.01) (.01) N 1,710 1,480 1, Adjusted -squared SEE Source for data: American National Election Studies. ependent variable: Party trusts to handle issue. Cell entries are OLS regression coefficients and (in parentheses) their standard errors. All variables scaled -.5 (in direction favoring emocrats) to.5 (in direction favoring epublicans) 5 The data presented here make it clear that issue ownership is an important, consistent factor in how Americans think about the two major political parties which goes beyond mere policy congruence between voters and parties. It is a concept that is distinct from shared policy preferences or party affiliation. In the next section, I explore how incorporating this notion in a model of electoral competition might influence how the parties respond to constituency opinion. V. A Note on epresentation A brief discussion is needed regarding how we measure representation. Following Achen (1978), I distinguish between two notions of representativeness. Centrism is the extent to which an elected representative locates close to the typical voter 5 Limited data availability prevented a perfect alignment of policy preference and issue ownership questions. In the defense/foreign affairs model, the NES issue-ownership question asked about foreign affairs; the policy question asked about defense spending. For the crime model, no spatial-location question was available. I used respondents stated positions on President Clinton s 1994 crime bill instead. 14

16 in her district. By contrast, responsiveness is the extent to which there is an association between representatives locations and district opinion across districts or over time. To help clarify these concepts, Figure depicts the stylized relationships between district opinion (on the x-axis) and the roll-call records established by legislators (the y-axis) in three hypothetical representative systems. Figure. Centrism and esponsiveness in Three Hypothetical epresentative Systems records established by legislators A B C ideal points of district median voters Line A in Figure depicts a system in which legislators are both perfectly centrist and perfectly responsive. Legislators do exactly as wished by their districts median voters, and thus line A lies directly on the graph s 45-degree line. Line B is a system in which legislators are perfectly responsive to district opinion, but not particularly centrist: each legislator takes a position that is further to the right than desired by her district. Finally, line C is a system that is not particularly centrist and is completely unresponsive. Very few legislators take positions near the ideal points of their district median voters, and there is no relationship between opinion and roll-call records across districts. Ideally, we would like to be able to assess a system s centrism and its responsiveness with empirical tests. But a limitation of the data has led scholars to focus on responsiveness: it is rare that opinion and policymaking are measured in the same units (particularly in the legislative context, where legislative records are derived from 15

17 roll-call votes on many different topics). It is therefore impossible to assess where the 45-degree line lies in an actual representative system, and thus very difficult to measure the extent to which the system is centrist. By contrast, differing units do not prevent us from assessing the slope of the relationship between opinion and policymaking across districts: this can be done easily with measures of correlation or with linear regression. I will therefore focus on my formal model s implications for responsiveness in the issueownership context, and limit my empirical tests to assessing responsiveness. VI. The Model Two candidates, N (from Party N) and O (from Party O), compete to represent a legislative district by announcing binding promises to deliver outcomes x N and x O in a one-dimensional policy space 6 that is in the interval [1,7]. 7 The candidates have known ideal points x N * and x O * over outcomes in this space. The election for the legislative seat is determined by an (odd-numbered) constituency of voters, each of whom has an ideal point in the policy space. Voters and parties derive quadratic utility from any outcome, x, such that U i (x) = -(x i x) for any actor i. The profile of voter preferences therefore is single-peaked, and thus any outcome preferred by the voter whose preference is at the median of the distribution of voters preferences will be preferred by a majority of the voters. Call this median voter V and his ideal point x * V. I assume that x * V is located between the two candidates ideal points and (without loss of generality) that x * N < x * V < x * O. Policy uncertainty and issue ownership Unlike many models of electoral competition, I assume that programs and outcomes are distinct. Specifically, any outcome x from which voters and candidates realize utility is the result of an interaction between the program p enacted by the winning candidate and a state of affairs S, which is a random variable whose realized value is 6 I have extended this model to the case where candidates take positions in a two-issue policy space, and found similar results. These results will be included in a future version of this paper. 7 I employ this interval for easy comparison with the scale used by the American National Election Studies to assess the locations of voters and parties in an issue space. For another example of this practice, see Adams, Merrill, and Grofman (005). 16

18 unknown at the time of the campaign. The realization of this random variable is an event s which affects the outcome x by interacting with the program p so that x = ps. In effect, this is a production function for outcomes where nature supplies events s that are operated upon by programs p to produce outcomes x. The random variable S is assumed to have a mean of w and a variance of σ. The justification for such a model is that it realistically captures how policymakers respond to events. 8 Consider for example a candidate whose party campaigns by promising an outcome of reduced crime. If elected, she may face an event such as the unexpected arrival on the market of a new, highly addictive illegal drug that requires her party to adjust their program to achieve the level of crime reduction (the outcome) she promised in the campaign. Examples of where events affect outcomes exist across issue domains, including defense spending (where a policymaker might have to respond to unforeseen threats to national security), environmental protection (where an economic downturn might affect how strong environmental regulation should be), and education (where test scores may unexpectedly rise or fall). Some politicians are better than others at responding to events like these. More relevant to this argument, the survey data discussed earlier in this paper indicate that on any given issue, Americans believe the issue-owning party is better than the other party at this task. I therefore incorporate issue ownership into the model in the following way: I assume that legislators from the issue-owning party are better able to assess the state of affairs S and respond with a program p that delivers the outcome x promised in the campaign. Call Party O the issue-owning party. Should candidate O be elected to office, her party s competency on the issue allows her to observe s perfectly and thus respond 8 This notion of policymaking under uncertainty is thus similar to that proposed by Gilligan and Krehbiel (1987) in their model of legislative committee expertise, with a crucial difference: I model policy outcomes as the multiplicative product of events and policy, rather than assuming (as do Gilligan and Krehbiel) that outcomes are the additive sum of events and enacted policy. As is discussed below, a multiplicative model captures the notion of issue ownership with the greatest fidelity. 17

19 with the policy that produces her promised outcome of x O. Specifically, she will enact a policy p O * = x O / s, which will result in the outcome x = p O * s = x O. Thus the median voter s expected utility from O s promised outcome x O, is: EU x E x p s * * [ V( O)] = [ ( V O ) ] = E x x * [ ( V O) ] ( ) ( ) * * = E xv xv xo + x * * V o V O = x x + x x o (1) By contrast, candidate N is assumed to have no competency on the issue except for knowledge of the distribution of S. Thus if N is elected to office, the best she can accomplish is to enact a policy p * N that takes into account the expected value of S to maximize the utility of a voter whose ideal point is the same as her promised platform of * x N. This policy is found by first substituting p N into the utility function: EU p E x p S * * [ x ( )] [ ( ) ] i= xn N = N N = Ex + p S x p S * * [ N ( N ) N N ] = x ( p ) ( w + σ ) + x p w * * N N N N Now we find the value of p N * that maximizes this expression by writing the first * order condition for a maximum of EU [ ( p )]: V N EU * [ x ( )] i= x p N N * pn = + + = * ( pn )( w σ ) xnw 0 which implies that p * N = xn w w + σ. () I pause here to note the consequences of modeling the outcome, x, as an interaction between S and p. Consider an alternative model one typically employed in 18

20 models of policymaking under uncertainty in which x were simply the sum of S and p. The expected utility derived by a voter with ideal point x N from any program p would be: U x E x x x ( ) [ ( ) ] i= x = N N = E x S+ p [( N ( )) ] = E x x S+ p + S + ps+ p [( N N( ) )] = x + x p+ x w w σ pw p N N N The p that maximizes this voter s utility (denoted p ** ) is found as follows: du dp ** N = x p w=0 ** p = xn w (3) The difference between the optimal values for p announced by N in the interactive case (shown in equation ) and the additive case (equation 3) are substantial. When x is modeled as the multiplicative product of p and S, the disadvantaged candidate must take into account the mean value and the variability of the state of the world in developing an appropriate platform. When x is simply (S + p), candidate N must only take into account the mean value of S. To understand why the multiplicative model makes more sense, consider the difference between the optimal policy approaches to crime prevention and preventing acts of terrorism. Both may produce the same amount of suffering on average, but the amount of suffering attributable to terrorism is much more variable. The additive model of policy uncertainty suggests that optimal preparation policies in these two domains should be similar; the interactive model (more realistically) suggests that optimal policies should differ due to the difference in the variability of the damage they can cause. eturning to the model, we can now calculate the median voter s expected utility from N s platform by substituting the expression for p * N in equation () into V s utility function: 19

21 EU x E x p S * * [ V( N)] = [ ( V N ) ] w E x x S w + σ * = V N * w * w = E ( xv ) + xn S xv xn S w σ + w + σ * xn w * w = ( xv ) E[ S ] + x [ ] V xn E S ( w + σ ) w + σ * xn w * w = ( xv ) ( w + σ ) + x V xn w ( w + σ ) w + σ * * w = ( xv ) ( xn xv xn). w + σ (4) Candidate preferences The candidates value outcomes as well as holding office. Following Groseclose (001), I model the extent to which candidates value office holding over outcomes with the parameter λ [0,1]. Where λ equals one, candidates value winning office only, and where λ equals zero they value outcomes only. O s expected utility function is therefore: EU [ ( x, x )] O O N * * w * w = λf (1 λ) F( xo xo) + (1 F) ( xo ) + xn x O xn w + σ w + σ (5) where F is an indicator variable that takes on the value one if O wins the election and zero if N wins. Similarly, the expected utility function for N is: EU [ ( x, x )] N O N * * w * w = λ(1 F) (1 λ) F( xn xo) + (1 F) ( xn ) + xn x N xn w + σ w + σ (6) 0

22 Play of the game 1. N and O simultaneously announce their promised outcomes x N and x O. (As is the standard case in models of electoral competition, these promises are binding and the winning candidate must implement the outcome upon which she campaigns.). An election is then held in which the median voter V elects one of the two candidates to office. V makes his choice by comparing the utility he expects to derive from the outcomes promised by N and O and electing the candidate yielding the higher expected utility. (I assume that if both candidates offer positions yielding equal utility, V breaks the tie by choosing O.) 3. Nature then draws s. If O was elected, she observes s perfectly. If N was elected, she learns nothing about s. 4. The winning candidate i enacts the program p i * that is most likely to result in the outcome x i she promised during the campaign. 5. Candidates and voters then realize utility from the resulting outcome x = p i * s. The game then ends. The solution concept employed is pure-strategy Nash equilibrium. esults This game results in the candidates locating apart, with O proposing an outcome x O that is less responsive to the district s preferences than is N s proposed outcome. Specifically, in equilibrium N proposes the outcome x N equal to x V *, while O proposes the outcome in the interval [x * V,, x * O ] that makes V just indifferent between the two candidates. This assures O s election. This result holds regardless of the value of λ (the degree to which candidates value office holding over policy outcomes). Hence, the issueowning party can afford to choose an outcome that is between the median voter and its ideal point. Thus the model predicts that candidates from the issue-owning party are able to be less responsive to the median voter than those from the other party. 1

23 As an illustration, I examine the case where candidates are completely outcome motivated and thus λ equals zero. We begin by considering the median voter s expected utility from x N and x O, as shown in equations (1) and (). Note that where x N = x O, the voter obtains utility from O s position that is greater by a factor of w /(w +σ ) than his utility derived from N s position. Thus O can always win the election by matching N s platform. But because O cares about policy, she can do better than this. Consider O s best reply to any proposed outcome x N played by N. This is found first by calculating the value of x O at which V is indifferent between the two candidates: EU [ ( x )] = EU [ ( x )] v O v N * * * w ( xv xo) = ( xv ) ( xn xv xn) w + σ * * * w xo = xv ± ( xv ) + ( xn xv xn) w + σ Because I have assumed that O s ideal point, x O *, is to the right of x V *, O s optimal platform as a function of x N is ( ) * * * w xo = xv + xv + ( xn xv xn). w + σ Since N is from the left party, she wishes for x to be as small as possible. Because x O O * reaches a minimum when x N = x V, a move by N to either the right or left of x * V allows O to take a more rightward position and still win the election. Substituting x * V for x N in the expression for x O and rearranging yields x ( x = x ) = x + * * O N V V * xv σ. The profile ( w + σ ) * * * xv σ xv,min xv +, x * O ( w + σ ) is a thus a Nash equilibrium. It can be shown that the same equilibrium holds for all values of λ. O s issueownership advantage allows it to always win the election even when N s platform is equivalent to the ideal point of the median voter. O exploits this advantage by

24 announcing a platform that is to the right of x V. Not surprisingly, O s advantage rises with σ that is, as the state of the world becomes less predictable because as σ increases, the precision with which N can deliver the outcome she promises decreases. Finally, note that as the variability of the state of the world declines and σ tends toward zero the platforms of both parties become responsive to the preferences of V. Therefore, on issues for which an announced program will surely yield the desired outcome, we should expect both parties to announce programs close to that of the median voter. Probabilistic voting The solution presented above is notable because it predicts that candidates outcomes will not converge upon the ideal point of the median voter. This result holds for purely outcome-motivated candidates, purely office-motivated candidates, and all those in between. Incorporating issue ownership into the model of electoral competition thus generates results different than those found by owns (1957). It is also notable for the fact that it reaches this result in a deterministic voting framework, where candidates have perfect knowledge of voters ideal points and how voter preferences translate into electoral outcomes. Typically, deterministic models lead to convergence upon the median voter s position as in the classic ownsian model (oemer 001). But the result is somewhat unsatisfying in that the model predicts that the candidate from party O will always win the election. It is likely that there are other, unobserved factors at work that may determine election results, and from time to time cause the candidate from party N to win despite her disadvantage. In addition, the candidates locations are implausibly unaffected by the extent to which they value outcomes versus office. We can incorporate these notions by moving to the probabilistic voting framework (Coughlin (199)), in which candidates are unsure of how the median voter will act and specifically, how the quantity EU [ ( x )] EU [ ( x )] will translate into an V O V N 3

25 election outcome. I now assume that the candidates assess that the probability O wins the election is equal to the probability that EU [ ( x )] EU [ ( x )] + ε > 0, where ε is a V O V N random variable distributed according to the logistic distribution function (denoted as Λ). Thus the probability O wins the election is Λ evaluated at EU [ ( x )] EU [ ( x )], or: V O V N exp( EU [ V( xo)] EU [ V( xn)]]) Pr( O wins) =Λ( E[ UV( xo)] E[ UV( xn)]) =. 1 + exp( EU [ ( x )] EU [ ( x )]) V O V N To complete the setup, I substitute this function for the indicator variable F in the expected utility functions for N and O (equations 5 and 6). Solving the issue-ownership game with probabilistic voting The solution concept of pure-strategy Nash equilibrium requires that we look for values of x N and x O that are best replies to one another that is, pairs of platforms in which neither candidate wishes to change her platform given the other candidate s * platform. I simplify the calculations by assuming that E[S] = w = x V that is, events are equally likely to require correction in the liberal direction from the median voter s perspective as they are to require correction in the conservative direction. We first calculate the platforms at which the candidates expected utility is maximized by finding EU [ i( xo, xn)] = 0 for i { N, O} : x i * ( xv ) ( V ) EU [ N( xo, xn)] = λλ (1 λ) ( x * ) (1 ) ( * ) ( * N xo Λ+ Λ xn + xn xnx N ) x * N x N x + σ and * ( xv ) ( V ) EU [ O( xo, xn)] = λλ (1 λ) ( x * ) (1 ) ( * ) ( * O xo Λ+ Λ xo + xn xnx O ), x * O x O x + σ 4

26 exp( EU [ v( xo)] EU [ v( xn)]) where Λ=. 1 + exp( EU [ ( x )] EU [ ( x )]) v O v N Performing these calculations, setting these expressions equal to zero, simplifying and collecting terms yields the following first order conditions: ( x * ) ( x x x * ) * ( xv ) + σ λ( x x ) λ x x ( x ) x x ( x x )( e * * * * V N N O * d O V = (1 ) ( O V ) O ( O O) + O O 1+ and ( x * ) ( x x x * ) * ( xv ) + σ λ( x x ) λ ( x x ) ( x ) ( x x ) ( x x )( e * * * * V N N N * d N V = (1 ) N V N N O + + N N 1+ ) ) where ( ) * ( xv ) + σ x x x x d = E[ U ( x )] E[ U ( x )] = x + x x * * V N N V * v O v N O O V. The complex nature of these first order conditions makes the model impossible to solve analytically. This is not unusual in models of electoral competition with policymotivated candidates. oemer (001) notes that there are no general proofs of the existence of these Wittman-style equilibria and that it is rare that these equilibria can be found analytically. I therefore proceed with numerical solutions to this game for a range of values of σ and λ. For illustrative purposes, I set x N = and x O = 6. Only one purestrategy Nash equilibrium exists for each {λ, σ} pair. Table 3 displays the candidates equilibrium platforms that are associated with a range of values for λ and σ when V s ideal point, x * V, is located directly in the middle of the policy space at 4. The table shows how O is able to exploit her issue-ownership advantage to announce a platform that becomes closer to her ideal point as the variability of the state of the world increases. As was the case in the deterministic model, this holds true in the scenario where candidates care only about policy (where λ equals zero), the 5