Leadership U.S. Electoral Politics

Transcription

1 Leadership U.S. Electral Plitics Evan Schnidman y, and Nrman Sch eld z Nvember 19, 2011 Abstract Frmal wrk n the electral mdel ften suggests that parties r candidates shuld lcate themselves at the electral center (r rigin). Recent research has fund n evidence f such cnvergence. In rder t eplain nn-cnvergence, the stchastic electral mdel is etended by including varius kind f valences, sme f which are based n individual and electral perceptins. These electral perceptins can be in uenced by the cntributins made by activist grups t the varius parties. We present this frmal activist mdel and then discuss the US presidential electin in 2008 and the midterm US electin in 2010 in rder t evaluate the in uence f activist grups, such as the s-called "Tea Party." As an applicatin f the mdel, we discuss the cntinuing realignment f US plitics that has that has ccurred since the end f the nineteenth century as a result f cn ict between activist calitins. JEL classi catin: H10 Key wrds: stchastic electral mdel, valence, electral perceptins, lcal Nash equilibrium. 1 Activist Plitics This paper attempts t mdel electins by incrprating vter judgments abut candidate and leader cmpetence. In a sense the prpsed mdel can be linked t Madisn s understanding f the nature f the chice f Chief Magistrate (Madisn, 1999 [1787]). and Cndrcet s wrk n the s-called Jury Therem (Cndrcet, 1994 [1785]). This aspect f Cndrcet s wrk has recently received renewed attentin (McLennan, 1998) and can be seen as a cntributin t the develpment f a Madisnian cnceptin f electins in representative demcracies as methds f aggregatin f bth preferences and judgments. The literature n electral cmpetitin has fcused n preferences rather than judgments. Mdels f tw-party cmpetitin have typically been based n This is a revised and etended versin f Sch eld and Schnidman (2011). y PhD Student, Department f Gvernment, Harvard University, Cambridge Massachusetts (evan.schnidman@gmail.cm) z Taussig prfessr f Plitical Ecnmy, Center in Plitical Ecnmy, Washingtn University, Saint Luis, Missuri (sch eld.nrman@gmail.cm).tel

2 the assumptin that parties r candidates adpt psitins in rder t win, and has inferred that parties will cnverge t the electral median, under deterministic vting in ne dimensin (Dwns, 1957) r t the electral mean in stchastic mdels. 1 This median based mdel has been applied recently by Acemglu and Rbinsn (2000, 2006a) in a wide ranging accunt f plitical ecnmy, including the transfrmatin f the British plity t a mre demcratic mdel in the nineteenth century. In this paper we develp a thery f plitical chice in which the plitical space is f higher dimensin. This space is ne which is derived essentially frm the underlying factr structure f the plitical ecnmy. That is t say, the aes are based n the preferences f thse wh cntrl the factrs f land, capital and labr. Fr eample, Figure 1 presents an estimate f the distributin f preferences (r preferred psitins) in the U.S. presidential electin f The rst-left right dimensin represents preferences (r attitudes) twards gvernment ependiture and taes and can be interpreted as a capital ais. The secnd nrth-suth r scial dimensin re ects attitudes n scial plicy, particularly civil rights, and can be interpreted as a labr ais. [Insert Figure 1 here] Because the plitical space is tw-dimensinal, parties in the United States must be calitins f ppsed interests. Figure 1 als shws a partisan cleavage line btained frm a simple lgit mdel f the electin. This cleavage line jins the preferred pints f vters wh, accrding t the stchastic vte mdel, wuld chse the candidates with equal prbability f ne half. In gure 2 we present the results f a factr analysis f the 2004 ANES, shwing estimated mean partisan and activist psitins fr Demcrat and Republican vters in 2004 (errr bars are larger fr the mean activist psitins. This Figure, tgether with Figure 1 suggests that candidate psitins are very much e ected by activists wh are estimated t be lcated at mre etreme psitins in the plicy space. This inference is cmpatible with the mdel presented here. [Insert Figure 2 here] The Figure suggests that thugh the Republican party cntains bth scially cnservative and scially liberal grups, they bth tend t be pr-capital. Similarly the Demcrat party tends t be pr-labr. The increasing dminance f Tea Party scial cnservatives in the Republic Party, and indeed the fact that the Republican psitin in the recent mid term electin f 2010 appeared t be fairly radical in the lwer right quadrant f the plitical space, caused sme prminent Republicans t cnsider a change f party allegiance t the Demcrats. Shifts in the activist calitins fr the tw parties thus cause a transfrmatin f the partisan cleavage line. Miller and Sch eld (2003,2008) argue that this is a fundamental aspect f 1 See the earlier wrk by Enelw and Hinich (1989), Eriksn and Rmer (1990) and mre recent wrk by Duggan (2006), and Patty, Snyder and Ting (2009). 2 This gure is based n factr analysis f the American Natinal Electin Study (ANES) fr 2004 by Sch eld et al. (2011a). 2

3 U.S. plitics: as activists n the trailing edge 3 f the cleavage line change party allegiance, then the psitins f the tw parties shift. This can be interpreted as a clckwise rtatin in the plitical space. They suggest that in the 150 years since the Civil War, the partisan cleavage line has rtated nearly 180 degrees, with the Republicans nw ccupying the psitin nce ccupied by the Demcrats in the late nineteenth century. Miller and Sch eld cnjecture that in time, the Republican Party will adpt plicies that are analgus t thse prpsed by William Jennings Bryan in 1896: ppulist and anti-business. In parallel, the Demcratic Party will increasingly appeal t pr-business, scial liberal csmplitans. We argue that the fundamental changes in vter chice result nt nly frm changes in the distributin f electral preferences, but frm the shifts in electral perceptins abut the cmpetence f the plitical candidates. 4 These perceptins are in uenced by the resurces that the candidates cmmand. In turn, these changes in perceptins are the cnsequence f the shifting pattern f activist supprt fr the candidates. 5 The essence f the mdel presented here is that it attempts t endgenize the resurces available t candidates by mdeling the cntracts they can make with their supprting activists. The activists must slve their wn ptimizatin prblem by estimating the bene t they receive frm their cntributins and deciding what resurces t make available t their chsen candidate. In recent years, the imprtance f activist cntributins has increased, and this has enhanced the in uence f activist grups. 6 The empirical and frmal mdels that we discuss here prvide a reasn why electral plitics has becme s plarized in the United States. 7 Mrever, this plarizatin appears t have bene ted the wealthy in sciety and may well accunt fr the increase in inequality in incme and wealth distributin that has ccurred ver the last decade. (Hacker and Piersn 2006, 2010, Piersn and Skcpl, 2007). Essentially there is an arms race between candidates ver these resurces due t a feedback mechanism between plitics and ecnmics. As the utcme f the electin becmes mre imprtant, activists becme increasingly aware that the resurces they prvide have becme crucial t electin victries, and they becme mre demanding f their chsen candidates. Because f the er f resurces, candidates are frced t mve t mre radical psitins, and plarizatin in candidate psitins increases, even thugh there may be little change 3 These wuld n the ne hand be csmplitan, scially liberal but ecnmically cnservative Republicans r n the ther hand, ppulist, scially cnservative but ecnmically liberal Demcrats. 4 In the empirical mdels that we have develped, perceptins are linked t candidate character traits such as mral, caring, knwledgable, hnest, mral, strng, ptimistic, intelligent. 5 In recent electins, candidate resurces are epended thrugh the media. Even a hundred years ag, presidential candidates had t epend resurces in campaigning thrughut the cuntry. 6 Indeed, Herrera, Levine and Martinelli (2008) bserve that spending by parties in federal campaigns went frm 58 millin dllars in 1976 t ver 1 billin in 2004 in nminal terms. 7 See the wrks by Firina et al. (2005, 2009) and McCarty, Ple, and Rsenthal (2006) n plarizatin in the electrate and Layman et al. (2010) n plarizatin amng activists. Sch eld et al. (2011a,b) gives similar results fr the 2000, 2004 and 2008 electins. 3

4 in the degree f plarizatin f the electrate. Over the lng run we see tw frces at wrk: the cntinuing circumferential realignment and a radial plarizatin that ccurs at times f plitical quandaries, caused by ecnmic dwnturn r shcks t the glbal plitical ecnmy. In the net sectin we present an utline f the mdel that we use. In Sectin 3 we present the frmal details f the mdel, and then in Sectins 4 and 5 we apply it t the cnsideratin f the 2008 and 2010 electins in the United States. Sectin 6 applies the mdel t episdes in United States histry, cmmenting n the balance between land, labr and capital. Sectin 7 cncludes. 2 An Outline f the Mdel. In the standard spatial mdel, nly candidate psitins matter t vters. Hwever, as Stkes (1963, 1992) has emphasized, the nn-plicy evaluatins, r valences, f candidates by the electrate are equally imprtant. In empirical mdels, a party s valence is usually assumed t be independent f the party s psitin, and adds t the statistical signi cance f the mdel. In general, valence re ects the verall degree t which the party is perceived t have shwn itself able t gvern e ectively in the past, r is likely t be able t gvern well in the future (Penn, 2009). Over the last decade a new literature has develped that cnsiders deterministic r prbabilistic vting mdels including valence r bias twards ne r ther f the candidates. 8 Recent wrk 9 has develped an empirical and frmal stchastic electral mdel based n multinmial cnditinal lgit methds (MNL). In this mdel, each agent,j; was characterized by an intrinsic r egenus valence, j. This mdel can be cnsidered t be Dwnsian, since it was based n a pure spatial mdel, where the estimates f valence were btained frm the intercepts f the mdel. It was pssible t btain the cnditins fr eistence f a lcal Nash equilibrium (LNE) under vte maimizatin fr a parallel frmal mdel using the same stchastic assumptins as the MNL empirical mdel. A LNE is simply a vectr f agent psitins with the prperty that n agent may make a small unilateral mve and yet increase utility (r vte share). The mean vter therem asserts that all candidates shuld cnverge t the electral rigin. 10 Empirical analyses f the 2000, 2004 and 2008 U.S. presidential electins (Sch eld et al., 2011a,b) has crrbrated the earlier wrk by Enelw and Hinich (1989) and shwn, by simulatin n the basis f the MNL mdels, that presidential candidates shuld cnverge t the electral rigin. 11 Hwever, the empirical wrk als suggests that presidential candidates d nt 8 Adams and Merrill (2005), Anslabehere and Snyder (2000), Aragnes and Palfrey (2002), Ashwrth and Buen de Mesquita (2009), Banks and Duggan (2005), Grseclse (2001) and McKelvey and Patty (2006). 9 Sch eld and Sened (2006), Sch eld (2007). 10 The electral rigin is the mean f the distributin f vter preferred pints. 11 Sch eld et al., 2011c btains a similar result fr the electins in Britain in 2005 and

5 in fact adpt psitins clse t the electral center. Figure 1, mentined abve, shws the estimated psitins f the presidential candidates in the 2004 electin in the U.S. This gure is cmpatible with previus wrk empirical wrk by Ple and Rsenthal (1984) wh als nted that there was n evidence f candidate cnvergence in U.S. presidential electins. This paper ers a mre general mdel f electins that, we suggest, accunts fr the di erence between the estimates f equilibrium psitins and actual candidate psitins. The mdel is based n the assumptin that there is a secnd kind f valence is knwn as activist valence. When party, r candidate j adpts a plicy psitin z j, in the plicy space,x; then the activist valence f the party is dented j (z j ): Implicitly we adpt a mdel riginally due t Aldrich (1983). In this mdel, activists prvide crucial resurces f time and mney t their chsen party, and these resurces are dependent n the party psitin. 12 The party then uses these resurces t enhance its image befre the electrate, thus a ecting its verall valence. Althugh activist valence is a ected by party psitin, it des nt perate in the usual way by in uencing vter chice thrugh the distance between a vter s preferred plicy psitin, say i ; and the party psitin. In this rst mdel, as party j s activist supprt, j (z j ); increases due t increased cntributins t the party in cntrast t the supprt k (z k ) received by party k; then (in the mdel) all vters becme mre likely t supprt party j ver party k: The prblem fr each party is that activists are likely t be mre etreme than the typical vter. By chsing a plicy psitin t maimize activist supprt, the party will lse centrist vters. The party must therefre determine the "ptimal marginal cnditin" t maimize vte share. The Therem, presented in Sectin 3, gives this as a ( rst rder) balance cnditin. Mrever, because activist supprt is denminated in terms f time and mney, it is reasnable t suppse that the activist functin will ehibit decreasing returns. The Therem pints ut that when these activist functins are su ciently cncave, then the vte maimizing mdel will ehibit a Nash equilibrium. It is intrinsic t the mdel that vters evaluate candidates nt nly in terms f the vters preferences ver intended plicies, but als in terms f electral judgements abut the quality f the candidates. These judgements are in turn in uenced by the resurces that the candidates can raise frm their activist supprters. Grssman and Helpman (1996), in their game theretic mdel f activists, cnsider tw distinct mtives fr interest grups: Cntributrs with an electral mtive intend t prmte the electral prspects f preferred candidates, [while] thse with an in uence mtive aim t in uence the pliticians plicy prnuncements. In ur rst activist mdel the term j (z j ) in uences every vter and thus 12 Fr cnvenience, it is assumed that j (z j ) is nly dependent n z j, and nt n z k ; k 6= j; but this is nt a cucial assumptin. 5

6 cntributes t the electral mtive fr candidate j: In additin, the candidate must chse a psitin t balance the electral and activist supprt, and thus change the psitin adpted. This change prvides the lgic f activist in uence. We argue that the in uence f activists n the tw candidates can be characterized in terms f activist gradients. Because each candidate is supprted by multiple activists, we etend the activist mdel by cnsidering a family f ptential activists, fa j g fr each candidate, j, where each k 2 A j is endwed with a utility functin, U k, which depends n candidate j s psitin z j ; and the preferred psitin f the activist. The resurces allcated t j by k are dented R jk (U k (z j )). Let jk (R jk (U k (z j ))) dente the e ect that activist k has n vters utility. Nte that the activist valence functin fr j is the same fr all vters. With multiple activists, the ttal activist valence functin fr agent j is the linear cmbinatin j (z j ) = P k2a j jk (R jk (U k (z j ))). We als btained infrmatin frm the American Natinal Electin Surveys n activists- thse wh cntributed resurces t ne r ther f the tw parties. Figure 1, abve, shwed the estimated psitins f activists fr the tw parties. The gure des suggest that activists in uence the candidate psitins. The balance cnditin suggests that the aggregrate activist gradients fr each f the tw candidates pint int ppsite quadrants f the plicy space. Bargains between the activists supprting candidate j then gives a cntract set f activist supprt fr candidate j; and this cntract set can be used frmally t determine the balance lcus, r set f ptimal psitins fr each candidate. This balance lcus can then be used t analyze the pre-electin cntracts between each candidate and the family f activist supprt grups. Cnsider nw the situatin where these cntracts have been agreed, and each candidate is cmmitted t a set f feasible cntracts as utlined in Grssman and Helpman (1994, 1996, 2001). Suppse further that the activists have prvided their resurces. Then at the time f the electin the e ect f this supprt is incrprated int the empirical estimates f the varius egenus, scidemgraphic and trait valences. Cnsequently, when we estimate these valences we als estimate the aggregate activist in uence. The estimated psitins f the candidates can then be regarded as incrprating plicy preferences f the activists. Electral mdels where candidates have plicy psitins 13 implicitly assume that candidates wuld be willing t accept defeat because f an adherence t particular plicy psitins. We argue that it is mre plausible that the estimated psitins f the candidates are the result f maimizing candidate utility functins that balance the electral cnsequences f psitin-taking with the necessity f btaining activist resurces t cntest the electin. This calculatin requires an estimate f the degree t which these resurces will in uence the perceptins that the electrate has f the varius valences assciated with the mdel. 13 As prpsed by Wittman (1977), Calvert (1985), Duggan and Fey (2005), Duggan (2006) and Peress (2010). 6

7 In the nal versin f the mdel we allw the activist valence functin t be individual speci c: The ttal resurces available t candidate j are nw dented j (z j ); and these may be allcated t individuals, with resurce m ij targeted n vter, r vter class, i by candidate j: Since m ij will depend n z j ; we write this allcatin as m ij (z j ); s the the budget cnstraint is R j (z j ) = X k2a j R jk (U k (z j )) = X i2n m ij (z j ): The ptimizatin prblem is nw a mre cmple ne, subject t this cnstraint. In actual fact, candidates will generally nt allcate resurces t individuals per se, but t vter classes via media utlets in di erent regins, r "zip cdes." The general balance cnditin speci es hw these resurces shuld be allcated thrughut the plity. A recent literature n electins has fcussed n the e ects f campaign ependiture n US electin results (Cate, 2004). 14 Herrera et al. (2008) suggest that electral vlatility frces candidates t spend mre, while Ashwrth and Buen de Mesquita (2007) suppse that candidates buy valence s as t increase their electin chances. Meirwitz (2008) ntes that candidates and parties spending this mney thught that it wuld in uence the electin utcme. Dwnsian mdels f cmpetitin cannt eplain hw candidates chse spending campaign levels r what factrs in uence these decisin. Meirwitz pries the chice f ependiture in terms f candidate chice f e rt, but his mdel des nt eplicitly deal with an endgenus budget cnstraint. 3 The Frmal Stchastic Mdel 3.1 The First Activist Mdel We develp an electral mdel that is an etensin f the multiparty stchastic mdel f McKelvey and Patty (2006), mdi ed by inducing asymmetries in terms f valence. The justi catin fr develping the mdel in this way is the empirical evidence that valence is a natural way t mdel the judgements made by vters f party leaders and candidates. There are a number f pssible chices fr the apprpriate mdel fr multiparty cmpetitin. The simplest ne, which is used here, is that the utility functin fr the candidate f party j is prprtinal t the vte share, V j, f the party in the electin. 15 With 14 An earlier paper by Grseclse and Snyder (1996) lked at vte buying, but in the legislature. 15 Fr re ning the mdel, and fr empirical analysis, it wuld be mre apprpriate t use the share f the electral cllege vtes, r a cmbinatin f this and the party vte shares in the electins t Cngress. We adpt this simplifying assumptin in rder t present the essential structure f the frmal mdel. 7

8 this assumptin, we can eamine the cnditins n the parameters f the stchastic mdel which are necessary fr the eistence f a pure strategy Nash equilibrium (PNE). Because the vte share functins are di erentiable, we use calculus techniques t btain cnditins fr psitins t be lcally ptimal. Thus we eamine what we call lcal pure strategy Nash equilibria (LNE). Frm the de nitins f these equilibria it fllws that a PNE must be a LNE, but nt cnversely. The key idea underlying the frmal mdel is that party leaders attempt t estimate the electral e ects f plicy chices, and chse their wn psitins as best respnses t ther party declaratins, in rder t maimize their wn vte share. The stchastic mdel essentially assumes that candidates cannt predict vte respnse precisely, but that they can estimate the e ect f plicy prpsals n the epected vte share. In the mdel with valence, the stchastic element is assciated with the weight given by each vter, i, t the average perceived quality r valence f the candidate. De nitin 1.The Stchastic Vte Mdel M(; ; ;; ) with Activist Valence: The data f the spatial mdel is a distributin, f i 2 Xg i2n, f vter ideal pints fr the members f the electrate, N, f size n. We assume that X is a cmpact cnve subset f Euclidean space, R w, with w nite. Withut lss f generality, we adpt crdinate aes s that 1 n i = 0: By assumptin 0 2 X; and this pint is termed the electral mean, r alternatively, the electral rigin. Each f the parties in the set P = f1; : : : ; j; : : : ; pg chses a plicy, z j 2 X, t declare prir t the speci c electin t be mdeled. Let z = (z 1 ; : : : ; z p ) 2 X p be a typical vectr f candidate plicy psitins. We de ne a stchastic electral mdel, which utilizes sci-demgraphic variables and vter perceptins f character traits. Fr this mdel we assume that vter i utility is given by the epressin u i ( i ; z) = (u i1 ( i ; z 1 ); : : : ; u ip ( i ; z p )) where u ij ( i ; z j ) = j + j (z j ) + ( j i ) + ( j i ) k i z j k 2 + j (1) = u ij( i ; z j ) + j : (2) Here u ij ( i; z j ) is the bservable cmpnent f utility. The cnstant term, j ; is the intrinsic r egenus valence f party j, The functin j (z j ) is the cmpnent f valence generated by activist cntributins t agent j: The term is a psitive cnstant, called the spatial parameter, giving the imprtance f plicy di erence de ned in terms f a metric induced frm the Euclidean nrm, jj jj; n X. The vectr = ( 1 ;..., j ; ::; p ) is the stchastic errr, whse mutivariate cumulative distributin is the Type 1 etreme value distributin, dented by : Scidemgraphic aspects f vting are mdeled by, a set f k -vectrs f j : j 2 P g representing the e ect f the k di erent scidemgraphic parameters (class, dmicile, educatin, incme, religius rientatin, etc.) n vting fr party j while i is a k-vectr denting the i th individual s relevant scidemgraphic characteristics. The cmpsitins f( j i )g are scalar prducts, 8

9 called the scidemgraphic valences fr j. The terms ( j i ) are scalars giving vter i 0 s perceptin f the traits f the leader (r candidate) f party j: The ce cients, j ;crrespnd t di erent candidates. We let = ( p ; :::: 1 ): 16 The trait scre can be btained by factr analysis frm a set f survey questins asking respndents abut the traits f the candidate. including mral, caring, knwledgable, strng, dishnest, intelligent, ut f tuch. Sch eld et al. (2011a,b) shw that the electral perceptins f candidate traits are statistically relevant fr mdeling US presidential electins. It is assumed that the intrinsic valence vectr = ( 1 ; 2 ; ::; p ) satis es p p : Vter behavir is mdeled by a prbability vectr. The prbability that a vter i chses party j at the vectr z is ij (z) = Pr[[u ij ( i ; z j ) > u il ( i ; z l )], fr all l 6= j]: (3) = Pr[ l j < u ij( i ; z j ) u il( i ; z l ), fr all l 6= j]: (4) Here Pr stands fr the prbability peratr generated by the distributin assumptin n. The epected vte share f agent j is V j (z) = 1 X n ij (z): (5) The di erentiable functin V : X p! R p is called the party pr le functin. The mst cmmn assumptin in empirical analyses is that is the Type I etreme value distributin (als called the Gumbel (maimum) distributin). The therem in this paper is based n this assumptin. This distributin assumptin is the basis fr much empirical wrk based n multinmial lgit estimatin. De nitin 2: The Type I Etreme Value Distributin, : (i) The cumulative distributin, ; has the clsed frm with prbability density functin i2n (h) = ep [ ep [ h]] ; (h) = ep[ h] ep [ ep [ h]] and variance : (ii) Fr each vter i; and party j; the prbability that a vter i chses party j at the vectr z is ij (z) = ep[u ij ( i; z j )] px : (6) ep u ik ( i; z k ) k=1 16 Fr US electins we talk f the traits f candidate j; rather than party leader j: 9

10 See Train (2003:79) In this stchastic electral mdel it is assumed that each party j chses z j t maimize V j, cnditinal n z j = (z 1 ; ::; z j 1 ; z j+1 ; ::; z p ). De nitin 3. Equilibrium Cncepts. (i) A vectr z =(z1; :::; zj 1 ; z j ; z j+1 ; :::; z p) is a lcal Nash equilibrium (LNE) i, fr each agent j; there eists a neighbrhd X j f zj in X such that V j (z 1; :::; z j 1; z j ; z j+1; :::; z p) V j (z 1; :::; z j ; :::; z p) fr all z j 2 X j : (ii) A vectr z =(z1; :::; zj 1 ; z j ; z j+1 ; :::; z p) is a pure strategy Nash equilibrium (PNE) i X j can be replaced by X in (i).. (iii) The strategy zj is termed a lcal strict best respnse, a lcal weak best respnse, r a glbal best respnse, respectively t z j =(z 1; :::; zj 1 ; z j+1 ; :::; z p) depending n which f the abve cnditins is satis ed. (iv) Strict lcal Nash equilibria (SLNE) and strict Nash equilibria (SPNE) are de ned analgusly by requiring strict inequalities in the de nitin. Frm the de nitins, it fllws that if z is a PNE it must be an LNE. Ntice that in this mdel, each agent is uncertain abut the precise electral utcme, because f the stchastic cmpnent f vter chice. Nne the less, we presume that each agent uses pinin pll data, etc. t estimate epected vte share, and then respnds t this infrmatin by searching fr a "lcal equilibrium" plicy psitin.in rder t gain as many vtes as pssible. It fllws frm (6) that fr vter i, with ideal pint, i ; the prbability, ij (z); that i picks j at z is given by ij (z) = [1 + k6=j [ep(f jk )]] 1 (7) where f jk = u ik( i ; z j ) u ij( i ; z j ): We use (9) t shw that the rst rder cnditin fr z t be a LNE is that it be a balance slutin. De nitin 4: The balance slutin fr the mdel M(; ; ;; ): Let [ ij (z)] =[ ij ] be the n by p matri f vter prbabilities at the vectr z, and let " # [$ ij ] = ij 2 ij n k=1 ( kj 2 kj ) be the n by p matri f weighting ce cients. The balance equatin fr zj is given by epressin zj = 1 d j + 2 dz j (8) nx $ ij i : (9) i=1 The vectr X i $ ij i is a cnve cmbinatin f the set f vter ideal pints. This vectr is called the weighted electral mean fr party j: De ne z el j = X i $ ij i : (10) 10

11 The balance equatins fr j = 1; :::; p can then be written as z el j zj 1 d j + = 0: (11) 2 dz j The bracketed term n the left f this epressin is termed the marginal electral pull f party j and is a gradient vectr pinting frm zj twards the weighted electral mean, zj el ; f the party. This weighted electral mean is that pint where the electral pull is zer. Ntice that the each entry in the vectr z el = (z1 el ; z2 el ; :::zp el ) depends n all ther entries. The vectr d j dz j is called the marginal activist pull fr party j. If z satis es the system f balance equatins, fr all j; then call z* a balance slutin. Fr the fllwing discussin nte again that by suitable chice f crdinates, the equi-weighted electral mean 1 n i = 0, and is termed the electral rigin. The fllwing therem is prved in Sch eld (2006a). Activist Therem 1. Cnsider the electral mdel M(; ; ;; ) based n the Type I etreme value distributin,and including bth intrinsic and activist valences. (i) The rst rder cnditin fr z* t be an LNE is that it is a balance slutin. (ii) If all activist valence functins are highly cncave, in the sense f having negative eigenvalues f su ciently great magnitude, then a balance slutin will be a LNE. Ntice that if X is pen, then this rst rder cnditin at z* is necessary fr z* t be a PNE. 3.2 Etensin t the case with multiple activist grups (i) Fr each party leader, j, let fa j g be a family f ptential activists, where each k 2 A j is endwed with a utility functin, U k ; which is a functin f the psitin z j :The resurces allcated t j by k are dented R jk (U k (z j )). The ttal activist valence functin fr leader j is the linear cmbinatin j (z j ) = X jk (R jk (U k (z j ))): (12) k2a j where f jk g are functins f the cntributins fr jk (U k (z j )g; and each jk is a cncave functin f R jk. (ii) Assume the gradients f the valence functins fr j are given by d jk dz j = a dr jk k dz j = a ka du k k (13) dz j where the ce cients, f a k ; a k g > 0; and are di erentiable functins f z j: 11

12 (iii) Under these assumptins, the rst rder equatin d j dz j d j = X dz j k2a j = X k2a j (a = 0 becmes d dz j [ jk (R jk (U k (z j ))) (14) k a k) du k = 0: (15) dz j The Cntract Curve generated by the family fa j g is the lcus f pints satisfying the gradient equatin X du k b k = 0; where X b k = 1 and all a k > 0: (16) dz j k2a j k2a j Here we let b k = (a k a k ) and renrmalize. The Balance Lcus fr the leader j; de ned by the family, fa j g; is the slutin t the rst-rder gradient equatin 2 3 z el j zj X du k a k 5 = 0: (17) 2 dz j k2a j The simplest case, discussed in Miller and Sch eld (2003) is in tw dimensins, where each leader has tw activist grups. 17 In this case, the cntract curve fr each leader s supprters will, generically, be a ne-dimensinal arc. Miller and Sch eld als suppsed that the activist utility functins were ellipsidal, mirrring di ering saliences n the tw aes. In this case the cntract curves wuld be catenaries, and the balance lcus wuld be a ne dimensinal arc. The balance slutin fr each leader naturally depends n the psitins f ppsed leaders, and n the ce cients, as indicated abve, f the varius activists. The determinatin f the balance slutin can be btained by cmputing the vte share Hessian alng the balance lcus. Because the activist valence functins can be epected t be cncave in the activist resurces, the Hessian f the verall activist valence, j ; can be epected t have negative eigenvalues. Fr this reasn, the Activist Therem 1 gives a frmal reasn t epect eistence f a PNE. In Figure 3, the pint z1(z 2 ) satis es the balance equatin fr a Republican candidate. This pint lies n the balance lcus f the Republican party, and is als a functin f the Demcrat candidate lcatin, z 2. A similar balance lcus can be cnstructed fr the Demcrat candidate. Nte that Figure 1 is cmpatible with Figure 3. [Insert Figure 3 here] If we assciate the utilities fu k g with leaders f the activist grups fr the parties, then the cmbinatin X du k a k dz j k2a j 17 See als Sch eld and Cataife (2007) fr eample. 12

13 may be interpreted as the marginal utility f the candidate fr party j, induced by the activist supprt. Ntice that the mdel presented here is frmally identical t ne where the party leader has plicy preferences. This activist mdel can be given a game-theretic fundatin, as in Grssman and Helpman (2001), and can in principle be etended t the case where there are multiple activist grups which have the ptin f chsing frm amng a set f pssible party leaders, all with varying intrinsic valences and preferences (Galiani, Sch eld and Trrens, 2012). 3.3 Etensin f the Activist Mdel: Targeting Vters We nw reinterpret j (z j ) = X k2a j R jk (U k (z j )): (18) as the ttal resurces btained by agent j frm the varius activist grups. These resurces are denminated in terms f time (times skilled labr rate) r mney, s we can take the units as mnetry. These resurces are used t target the individual vters and the vter utility functin is nw u ij ( i ; z j ) = j + i (m ij ) + ( j i ) + ( j i ) k i z j k 2 + " j = u ij( i ; z j ) + " j : Here i (m ij ) is the valence e ect f the ependiture f resurces, (m ij ) n the targeting f vter i; by agent j: We assume that the greater the resurces m ij spent n persuading vter i; the greater the implicit valence assciated with candidate j; s d i (mij) > 0: We may als assume decreasing returns s that d 2 i (m ij) dm 2 j dm j < 0: Obviusly we can partitin the vters int di erent categries, in terms f their scidemgraphic valences. Nte that di erent agents may target the same vter r grup f vters. We assume that fr each j the budget cnstraint is satis ed: R j (z j ) = X X R jk (U k (z j )) = m ij (19) k2a j i2n We nw assume that j slves the ptimizatin prblem that we nw cnstruct. Since j (z j ) determines the budget cnstraint fr j; we can write m ij m ij (z j ), s i (m ij ) i (m ij (z j ) ij (z j ): : We shall als assume that the slutin t the ptimizatin prblem is smth, in the sense that ij () is a di erentiable functin f z j : 13

14 Then just as abve, the rst rder cnditin gives a mre general balance cnditin as fllws: 0 = dv j(z) dz j S z j X = 1 n X i2n = 1 n X i2n i2n[ ij 2 ij] = X i2n d ij dz j [ ij 2 ij]f2( i z j ) + d ij dz j (z j )g: [ ij 2 ij]f i + 1 d ij (z j )g: 2 dz j Hence z j = r z j = X i nx i=1 h ii h[ ij 2 ij ] i + 1 d ij 2 dz j (z j ) P k2n [ kj 2 kj ] $ ij ( i + i ) where i = 1 d ij (z j ) 2 dz j [ ij 2 ij and $ ij = P ] k2n [ kj 2 kj ] This can be written z j zj el = n X i=1 $ ij i where z el j = nx $ ij i. i=1 When d ij dz j (z j ) = d j dz j (z j ) this reduces t the previus result (11). The di erence nw is that instead f there being a single centrifugal marginal 1 d activist pull j 2 dz j (z j ) there is an aggregrate activist pull nx $ ij i i=1 = 1 2 nx i=1 [ ij 2 ij ] P k2n [ kj 2 kj ] d ij dz j (z j ) determined by the budget cnstraint given in Eq (31). Ntice that the rst rder cnditin depends n the marginal terms, d ij dz j (z j ); assciated with plicy psitins, and these will depend n the marginal valence e ects d i (mij) dm j : Althugh these valence e ects can be assumed t ehibit decreasing returns, these will vary acrss di erent classes f vters. The plausibility f eistence f Nash equilibria turns n whether the induced secnd 14

15 rder terms d2 ij (z dzj 2 j ) have negative eigenvalues. The assumptin f negative eigenvalues wuld give a versin f the activist therem. Nte als that if ij is clse t 0 r 1, then $ ij will be clse t 0, s the ptimal calculatin will be cmple, thugh in principle slvable. It is plausible the candidate shuld epend resurces n pivtal vters fr whm ij is clse t 1/2. 18 T sketch an utline f a general mdel t endgenize activist supprt, we rst let : [X B n ] p! [0; 1] np specify the vter prbabilities in terms f candidate psitins in X p and the distributin, in B np ; f resurces fm ij g t all vters. 19 We then let V = V 1 :: V p : [X B n ] p! [0; 1] p be the party pr le functin, mapping party psitins and vter distributins t vte shares, as given by the abve mdels. Indeed, fr a mre general mdel we culd cnsider multiparty systems where agents frm beliefs abut calitins behavir, as suggested in Sch eld and Sened (2006). In this case the mapping wuld be V = V 1 :: V p : [X B n ] p! R p : We assume that each f the k activists ers a distributin f resurces t the p party leaders, which we take t be a vectr in B k : We seek is an equilibrium t a game frm which may be written U V: W =B k [X B n ] p! R k R p : This is an etremely cmple dynamical game, and we d nt attempt t eplre the full rami catins f this mdel here. 20 One way t deal with it is t cnsider a dynamical versin by cnsidering a preference eld fr each party, r activist. This will be a cne in the tangent space f the agent s strategy space which speci es thse changes in the agent s behavir which increase the agents utility. We dente the jint preference eld by H UV : W! TW 18 Stkes (2005) make a smwhat similar inference, discussing clientist mdels f plitics, where m ij is simply a mnetary bribe t i: Obviusly the marginal bene t t a pr vter is greater than t a wealthy vter, under the usual assumptin f decreasing marginal utility fr mney. 19 It is reasnable t assume that the resurce distributins lie in a cmpact ball, namely B np : 20 See Cram (2010) fr a dynamical versin f a similar mdel. Acemglu and Rbinsn 2008) als develp a mdel based n Markv Perfect Equilibrium where the elite, activists, have di erent preferences fr the public gd, in X and cntribute t the de fact pwer f the plitical leader. Hwever, they d nt assume cmpeting plitical leaders.the matching mdel prpsed by Jacksn and Watts (2010) embeds the Nash equilibrium within a calitin game, and wuld allw the principals t switch frm ne agent calitin t anther. 15

16 where TW stands fr the tangent bundle abve W: A result in Sch eld (2011) shws that if the tangent eld H UV satisies a half pen prperty then there will eist a critical Nash equilibrium satisfying the rst rder cnditin fr equilibrium. Earlier results f Sch eld (1978) and McKelvey (1979) had suggested chas culd be be generic in electral mdels. 21 The applicatin f this mdel (in sectin 6) t the histrical develpment f the U.S. plitical ecnmy suggests that the equilibria f the mdel are subject t bth circumferential and radial transfrmatins ver time. 4 Activist Supprt fr Parties in the United States T apply the abve mdel, suppse there are tw dimensins f plicy, ne ecnmic, and ne scial. These can be fund by factr analysis f survey data. As in Figure 2 indicates, we represent the cn icting interests r bargains between the tw activist grups f supprters fr the Republican Party, lcated at R and C, by a "cntract curve." This represents the set f plicies that these tw grups wuld prefer their candidate t adpt. It can be shwn that this cntract curve is a catenary whse curvature is determined by the eccentricity f the utility functins f the activist grups (Miller and Sch eld, 2003). We call this the Republican cntract curve. The Demcrat activist grups may be described by a similar cntract curve (This is the simplest case with just tw activist grups fr each candidate. As the previus sectin shws, this idea can be generalized t many activist grups.) The therem presented abve gives the rst rder cnditin fr the candidate psitins (zdem ; z rep) t be a Nash equilibrium in the vte share maimizing game. This cnditin is that the party psitins satisfy a balance equatin. This means that, fr each party, j = dem r rep; there is a weighted electral mean fr party j, given by the epressin z el j = X i $ ij i : (20) This is determined by the set f vter preferred pints { i g:ntice that the ce - cients f$ ij g fr candidate jwill depend n the psitin f the ther candidate,k: As presented in the frmal mdel, the balance equatin fr each j is given by: z el j 4.1 The 2008 electin zj dj dz j j z = 0: (21) The previus sectins have suggested that a candidate s valence at electin time is due t the ability f activist grups t raise resurces fr the candidate. At the same time, the candidate psitins are the result f a balancing act between 21 See als Riker (1980, 1982, 1986). 16

17 chsing an electrally ptimal psitin and being able t persuade activist grups t prvide these resurces.we brie y prvide sme infrmatin abut this balancing act: Figure 4 shws the estimated psitins f Republican and Demcrat Presidential primary candidate psitins prir t the 2008 electin. The gure clearly suggests that Obama adpted a fairly etreme plicy psitin, very liberal n bth ecnmic and scial aes. Figures 5 and 6 shw the relatinship between electral ppularity f the candidates and their campaign ependitures as f January [Insert Figures 4,5 and 6 here] Figure 5 suggests that Obama and Hilary Clintn were bth very successful in raising campaign resurces, and that these were highly crrelated with the electral supprt. Other candidates fell far behind and drpped ut f the race. Figure 6 suggests that McCain was als etremely ppular, even thugh his campaign, in January 2008, had nt been very successful in raising cntributins. This inference is cmpatible with McCain s estimated fairly mderate psitin in Figure 4. Obviusly, the relatinship between campaign resurces and ppular vte in primaries and in the general electin is etremely cmple. Further research will attempt t utilize the mdel presented here t clarify this relatinship. Obama s victry n Nvember 4, 2008 suggests that it was the result f an verall shift in the relative valences f the Demcrat and Republican candidates frm the electin f Indeed, Sch eld et al (2011b) analyse a spatial mdel f the 2008 electin and btain a gure f 0.84 fr the estimate f Obama s valence advantage ver McCain. In fact there were di erential shifts in di erent regins f the cuntry. In a regin f the cuntry frm West Virginia thrugh Tennessee, Arkansas and Oklahma, there was an increase f 20% in the Republican vte ver the share fr 2004, suggesting a reginal change f abut 0.6 in McCain s valence advantage. Obama s victry in 2008 suggests that plicy utcmes during his administratin ught t lie in the upper left hand quadrant f the plicy space. Figure 7 prvides an estimate (taken frm Sch eld et al. 2011a) f the lcatin f McCain and Obama at the Nvember 2008 electin. The Figure als shws the lcatin f Demcrat and Republican activists. Again, there is sme evidence that etreme activists in uence the plicy chices f the candidates. [Insert Figure 7 here] 5 Pst 2008 Electin The precise plicy utcme frm Obama s administratin have thus far depended largely n the degree t which Republicans in the Senate have blcked Demcratic plicies thrugh the use f the libuster. Early in his administratin sme f Obama s plicy initiatives successfully passed thrugh Cngress but nly after navigating Republican ppsitin in the Senate. Fr eample, n January 15, 2009, the Senate vted 52 against and 42 in supprt f Obama s ecnmic recvery prgram. On February 6, 2009 an agreement was reached in 17

18 the Senate t reduce the size f the stimulus bill t $780 billin, in return fr the supprt f three Republican senatrs. On February 9 the Senate did indeed vte by the required majrity f 61 t halt discussin f the stimulus bill, thus blcking a libuster. A cmprmise bill f $787 billin, including sme ta cuts, was agreed upn by bth the Huse and Senate within a few days; the bill passed the Huse with 245 Demcrats vting in favr and.183 Republicans vting against while the Senate passed it with just 60 vtes. The bill was immediately signed by President Obama. As Obama cmmented afterwards: Nw I have t say that given that [the Republicans] were running the shw fr a pretty lng time prir t me getting there, and that their thery was tested pretty thrughly and its landed us in the situatin where we ve gt ver a trillin dllars wrth f debt and the biggest ecnmic crisis since the Great Depressin, I think I have a better argument in terms f ecnmic thinking. On February 26, 2009 Obama prpsed a 10 year budget that revised the pririties f the past, with an estimated budget de cit fr 2009 at $1.75 trillin (ver 12% f GDP). It included prmises t address glbal warming and t reverse the trend f grwing inequality. The $3.6 trillin Federal budget prpsal passed the Huse n April 2, 2009 by 233 t 196, with even "blue dg" cnservative Demcrats supprting it, but, again, n Republicans. Obama s scial plicies even received a mdicum f success; n January 22, 2009 a bill against pay discriminatin passed the Senate 61 t 36. The Huse als gave nal apprval n February 4, by a vte f 290 t 135, t a bill etending health insurance t millins f lw-incme children. Frty Republicans vted fr the bill, and 2 Demcrats vted against it. When the bill was signed by President Obama, it was seen as the rst f many steps t guarantee health cverage fr all Americans but it was nt clear that the battle ver brader healthcare legislatin wuld take mst f Obama gained anther imprtant victry when the Senate cn rmed Snia Stmayr as Supreme Curt Justice n August 6, 2009, by a vte f 68 t 31. She is the rst Hispanic and the third wman t serve n the Curt. Similarly, Obama nminated anther wman, Elena Kagan, t the high curt and she was cn rmed almst eactly ne year after Stmayr n August 7, 2010 by a vte f 63 t 37. Thugh adding tw left-leaning female justices t the curt has increased the number f wmen n the Supreme Curt t an all time high f 3, it has nt fundamentally changed the idelgical make-up f the current curt which still regularly splits 5 t 4 in favr f mre right-leaning rulings. Events in 2009 and 2010 are cnsistent with the mdel presented in Sch eld and Miller (2007). Obama is attempting t attract and retain pr-business scial liberals with his respnse t the ecnmic crisis while his massive budget prpsal addresses the ecnmic dwn-turn but has angered mst Republicans. It is pssible that the Republican Party will gain vtes frm the blue-cllar vters wh are su ering the mst frm the ecnmic cllapse. Hwever, if 18

19 there is any ecnmic recvery by the 2012 electin, it is pssible that many f the pr-business grups in the cuntry will respnd t Obama s attempt t get the ecnmy mving by supprting him. That culd leave the Republican Party with nthing but the ld-style ppulism f William Jennings Bryan: anti- Wall Street, anti-banking, anti-detrit, anti-immigratin, and pr-evangelical religin. This will result in a party realignment t a situatin where the scially liberal and ecnmically cnservative csmplitan Demcrats are ppsed t ppulist Republicans. 22 In Octber, 2009, ne grup identifying as ppulist Republicans, the Tea Party" activists ppsed Obama s plicies n health care s much that they began lining up against the centrist Gvernr Charlie Crist in the GOP Senate primary. Ultimately, Crist was frced t becme an Independent and a Tea Party darling, Marc Rubi, was nminated as the GOP candidate fr the Flrida Senate seat (and ultimately wn the seat, beating Crist handily). Similarly, n Nvember 1, 2009 the centrist Republican candidate, Dede Sczzafava, decided t drp ut f the special electin in New Yrk s 23rd cngressinal district and endrse the Demcrat candidate, Bill Owens. Owens wn the electin in a district that had been Republican since As the Healthcare debate heated up ver summer and fall f 2009 it became clear that Republicans were intending t cntinue utilize their blcking calitin as lng as pssible t stimy Obama and the Demcrats. Interestingly, sme Demcrats cntributed t this ppsitin as well; in the health bill vte in the Huse in early Nvember 2009, 219 Demcrats with 1 Republican vted fr the bill, while 176 Republicans and 39 "Blue Dg" Demcrats vted against. 23 By December 19, Senatr Bernie Sanders f Vermnt, an independent wh caucuses with the Demcrats, as well as Demcrat Senatrs Ben Nelsn and Sherrd Brwn, had agreed t a cmprmise bill. This brught the size f the calitin t the critical size f 60 vtes, su cient t frce a decisin in the Senate. 24 Finally n Christmas Eve, 2009, the health bill passed in the Senate, again by 60 vtes with 39 Republicans ppsed. Hwever, the victry by Republican Sctt Brwn in the special Senate electin in Massachusetts n January 19 deprived the Demcrats f the 60 seat majrity required t push thrugh the legislatin. On February 25, 2010, an attempt t reach a bipartisan cmprmise failed, and there was talk f using a maneuvre knwn as recnciliatin t frce thugh a health bill using simple majrity rule. 25 Finally, n March 25, after strenuus e rts by President Obama and Huse speaker, Nancy Pelsi, the 22 That is, unlike the situatin in the previus gures, the Republican Party will mve t the lwer left quadrant f the plicy space, while business interests in the upper right quadrant will switch t the Demcrats. It is indicative f this trend that n April 28 Arlen Specter, the senatr frm Pennsylvania, shifted his allegiance frm the Republican Party t the Demcrats. 23 On Saturday, Nvember 21, the Senate vted 60 t 40, alng partisan lines, t mve t the nal discussin n the health care bill. 24 Clture is a mtin aimed at bringing debate t an end. It riginally required a tw-thirds majrity, but since 1975 has required a super-majrity f Recnciliatin is a measure whereby a bill can pass the Senate with a simple majrity; the legislatin must be shwn t be budget neutral ver a ten-year span in accrdance with the Byrd rule. 19

20 Huse vted t send a health care bill t the President. Republicans vted unanimusly against the legislatin, jined by 33 dissident Demcrats. The Senate passed the bill by simple majrity f 56 t 43, as required under recnciliatin and the President signed a draft f the bill, the "Patient Prtectin and A rdable Care Act," n March 23, 2010 and an updated versin f the bill n March 30, While it seemed that "gridlck" ensued ver the health care legislatin, several ther majr pieces f legislatin passed with far less ppsitin. On February 22, 2010 and again n March 17, 2010 the Senate vted and respectively t implement tw multi-billin "jbs creatin" prgrams. Even thugh the vte t end debate n the Financial Regulatin bill failed t btain the required supramajrity n May 19, 2010, it eventually passed the Senate. On July 15, 2010 the Senate vted fr the Ddd-Frank Wall Street Refrm and Cnsumer Prtectin Act, and this was signed int law by President Obama n July President Obama als signed int law a bill t restre unemplyment bene ts fr millins f Americans wh have been ut f wrk fr si mnths r mre. Further cmplicating issues f partisan discntent in Cngress has been the intrductin f ever increasing quantities f mney in the American plitical system. Fr eample, in 2009, health care, pharmaceutical and insurance lbbyists 27 spent apprimately $650 millin n lbbying itself, and abut $210 millin n media advertising, while the il and gas industry spent abut $560 millin. 28 It wuld seem inevitable that the imprtance f lbbying can nly increase in the future. 29 The Supreme Curt decisin, Citizens United v. Federal Electin Cmmissin, n January 21, 2010, remved limits n campaign cntributins and will further increase the imprtance f activist cntributins. An earlier Curt decisin, Federal Electin Cmmissin v. Wiscnsin Right t Life Inc. had allwed crpratins t buy advertisements supprting candidates as lng as they did nt appeal eplicitly fr the electin r defeat f a particular candidate. Citizens United remved this limitatin. In his State f the Unin address in late January, 2010 President Obama said the curt had reversed a century f law that I believe will pen the dgates fr special interests including freign crpratins t spend withut limit in ur electins. 30 Dwrkin (2010) later called the Supreme Curt decisin an 26 This cmple bill was 2300 pages lng. Russ Feingld, a Demcrat, vted against the bill, because it was nt strng enugh. Three mderate New England Republicans, Snwe and Cllins f Maine, and Sctt Brwn f Massachusetts, vted fr the bill. The death f Senatr Rbert Byrd f West Virginia made it mre di cult t summn the required 60 vtes fr clture. 27 The pharmaceutical industry was a strng supprter f refrm f health care, because f an agreement with Obama t prtect the industry s pr ts. 28 Tmasky (2010) gives a gure f $3.47 billin fr spending by lbbyists in the nn electin year f 2009, citing data frm the Center fr Respnsive Plitics. 29 Indeed, Herrera, Levine and Martinelli (2008) bserve that spending by parties in federal campaigns went frm 58 millin dllars in 1976 t ver 1 billin in 2004 in nminal terms. 30 Ntably, Gerge W. Bush appinted Supreme Curt Justice Samuel Alit brke frm traditinal judicial decrum at State f the Unn speeches t shake his head in disagreement with the President reprtedly muttering the wrds "that s nt true." 20