A Theory of Divided Government The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters Citation Alesina, Alberto, and Howard Rosenthal. 1996. “A Theory of Divided Government.” Econometrica 64 (6) (November): 1311. doi:10.2307/2171833. Published Version doi:10.2307/2171833 Citable link http://nrs.harvard.edu/urn-3:HUL.InstRepos:34222831 Terms of Use This article was downloaded from Harvard University’s DASH repository, and is made available under the terms and conditions applicable to Other Posted Material, as set forth at http:// nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of- use#LAA Econometnica, Vol. 64, No. 6 (November, 1996), 1311-1341 A THEORY OF DIVIDED GOVERNMENT1 BY ALBERTOALESINA AND HOWARDROSENTHAL This paper extends the spatial theory of voting to an institutionalstructure in which policy choices depend upon not only the executive but also the compositionof the legislature.Voters have incentivesto be strategicsince policy reflects the outcome of a simultaneouselection of the legislature and the executive and since the legislature's impact on policy depends upon relative plurality.To analyze equilibriumin this game between voters, we apply "coalition proof' type refinements.The model has several testable implicationswhich are consistentwith voting behaviorin the United States. For instance,the model predicts:(a) split-ticketswhere some voters vote for one party for presidentand the other for congress;(b) for some parametervalues, a dividedgovernment with differentparties controlling the executiveand the majorityof the legislature;and (c) the midtermelectoral cycle with the partyholding the presidencyalways losing votes in midtermcongressional elections. KEYWORDS: Voting theory,checks and balances,conditional sincerity, midterm cycle, coalitionproof Nash equilibria. 1. INTRODUCTION NATIONALPOLICIES TYPICALLY REFLECT the composition of both the executive and the legislature. In the United States, for instance, the administration's influence on policy is affected by the compositionof congress. Traditionalspatial models of elections ignore this institutionalcomplexity and assume that the winner of a two-partyrace fully controlspolicy.2 These models focus on the strategicinteraction between the candidatesand treat each voter as a passive player,voting for the candidateoffering the more preferredpolicy. In our model, instead,the voters face two partieswith preferencesdefined on policy outcomes(Wittman (1977, 1990),Calvert (1985), Alesina (1988)).Without "checks"by the legislature,the executivesrepresenting the two parties pursue distinct and polarizedpolicies.3 However, policy outcomes in our model are a functionnot only of which partyholds the executivebut also of the composition of the legislature. The ideal policies of the two parties are identified by two points on an interval;voters to the left of the left party'spolicy alwaysvote for this party in both the legislative and presidentialelections; similarly,voters on the right of 1 This work was supportedby National Science FoundationGrant SES-8821441.Alesina also gratefullyacknowledges support from the Sloan Foundation.An earlierversion of this paper was circulatedwith the title "ModeratingElections." We thank Geir Asheim, Michael Whinston,a co-editor, and the referees for extremelyhelpful comments.The manuscriptwas revised while Rosenthalwas a fellow of the Centerfor AdvancedStudy in the BehavioralSciences. 2 For example,see Kramer(1973). For surveysof this literature,see Enelow and Hinich (1984) and Ordeshook(1986). 3The existenceof "polarization"in Americanpolitics has been widelydocumented by Poole and Rosenthal(1984a,b; 1991). This content downloaded from 128.103.149.52 on Sat, 7 Jun 2014 15:22:45 PM All use subject to JSTOR Terms and Conditions 1312 A. ALESINA AND H. ROSENTHAL the right partyhave a dominantstrategy to vote for the right party.In contrast, voters with ideal policies intermediatebetween the positionsof the parties take advantageof the institutionalstructure of "checks and balances"to moderate the executive and achieve a policy outcome closer to their ideal. This realistic feature of the model implies strategic voting. Unlike the traditional spatial model, the decisionof each "middle-of-the-road"voter dependsupon his beliefs about the behavior of other voters. Thus we examine strategic voting in the context of the "new institutionalism's"concern about the executive-legislative interaction(Shepsle (1986), Hammondand Miller (1988),Fiorina (1990, 1991)).4 Our model of "moderatingbehavior" has implicationsconsistent with three observationsconcerning American politics: split-ticket voting, the midtermcycle, and dividedgovernment. The model predictssplit-ticket voting when the execu- tive and the legislatureare elected simultaneously.That is, some voters support differentparties in the two elections to, say, counterbalancea presidenton the right with a legislature on the left. The model also captures the persistent phenomenonof the "midtermvoting cycle."In midtermcongressional elections, the president's party loses plurality relative to the preceding congressional elections held simultaneouslywith the presidential election. Our explanation relies upon the fact that in presidential election years the electorate faces uncertaintyabout who will win the presidency.Moderate voters want neither a left-dominatedcongress for fear that the presidentmay turn out to be left, nor a right-dominatedcongress lest the presidentbe right.Their uncertaintyabout the outcome of the presidentialelection leads them to prefer a "hedged"congress. At midterm,once the identity of the presidentis known,voters move congress further in the direction opposite to that of the president in order to achieve better moderation.Empirical studies by Alesina and Rosenthal (1989), Erikson (1990), and Alesina, Londregan,and Rosenthal (1993) have shown that this model of the midterm effect performs at least as well and often better than traditionalempirical voting models that emphasizeincumbency advantage and retrospectivevoting based on the state of the economy.5A sufficientlystrong midtermcycle or substantialsplit-ticket voting will producedivided government where the party holding the presidency does not have a majority in the legislature. We present both a complete information and an incomplete information model. In the former,but not the latter, the distributionof voter ideal points is common knowledge.Split-ticket voting, moderatingbehavior, and dividedgov- 4Cox (1987)analyzes strategic voting when voters decide more than one contest in the contextof the double-memberdistricts of Victorian England. He finds evidence of substantialsplit-ticket voting until cabinet governmentis instituted.With cabinet government,policy depends only on a parliamentarymajority and not on, as in the Americansystem, an executive-legislativeinteraction that is affectedby the compositionof the legislature.The virtualdisappearance of split-ticketvoting undercabinet government is expectedfrom the perspectiveof our model. 5Erikson (1988)tests coattailsand other alternativeexplanations of the midtermcycle and finds that they are not supportedby the data. This content downloaded from 128.103.149.52 on Sat, 7 Jun 2014 15:22:45 PM All use subject to JSTOR Terms and Conditions THEORY OF DIVIDED GOVERNMENT 1313 emnment,present in both cases, can be more easily illustratedwith the complete informationmodel.6 A midtermcycle emerges only in the incompleteinforma- tion model. The Alesina and Rosenthal(1995, pp. 43-120) volume,written after the initial draft of this paper, motivates these models using the concept of conditionalsincerity briefly discussedin Section 3.2 of this paper but does not analyze equilibriumusing the game-theoreticconcepts of abstract stable sets and coalition proof Nash equilibrium. The paper is organizedas follows. In Section 2 we describe the one-period model and the game-theoreticsolution concepts which we employ.We analyze equilibriumin this basic model in Section 3. This section uses rather general assumptionsabout functional forms. In Section 4 we present the two-period complete informationcase with more restrictiveassumptions. In Section 5 we discuss the incompleteinformation version of the model for specific functional forms.Section 6 concludes. 2. THE ONE-PERIOD MODEL 2.1. Preferencesand Institutional Setup We consider a two-partysystem in which the parties (identified with the candidates)have policy preferences.In particular,the two parties, labelled D and R, and the voters have unidimensional,continuous, and strictly concave utility functions over a single policy issue. Utility functions are all identical except for the bliss point. The utility of a policy x is given as (1) u1 = U(X, 0i) where, for the two parties, we denote the bliss points as OD and OR with 0 < OD < OR < 1 and, for the generic voter i, we denote the bliss point as Oi. There is a continuumof voters. The bliss point cumulativedistribution H(O) is continuousand strictlyincreasing on the interval I = [0,1]. The median type is denoted ,u= H-1 (0.5). Any non-null subset, S, of I is termed a coalition. The preferences of both parties and voters are common knowledge.We choose to workwith a continuumof voters ratherthan with a finite numberfor reasonsof technicalconvenience clarified in footnote 13, followingProposition 2 below. The president is elected by majorityvote. If the executive had complete control of policy, the partywinning the presidencywould